A tunnel orthographic image generation processing method, system and platform based on global constraint optimization and contour refinement distance setting, and a storage medium

Abstract

This invention discloses a method, system, platform, and storage medium for generating and processing tunnel orthophotos based on global constraint optimization and contour refinement distance determination. It generates and acquires first data corresponding to the tunnel; based on a longitudinal geometric reference with known physical spacing, it uses a global constraint optimization method to longitudinally calibrate the first data, generating corresponding second data; it then refines the contour distance determination for each section in the second data, generating corresponding third data. This method overcomes the limitations of local correction in dynamic filtering, achieving global compensation for longitudinal mileage errors and improving the geometric consistency of long-distance tunnel images. It accurately extracts the main lining contour in complex sections, effectively eliminating interference from auxiliary structures such as pipelines and cable troughs, ensuring that the distance division is based on the actual tunnel wall. Based on the distance division, it achieves adaptive grayscale equalization, compensating for intensity unevenness caused by changes in laser incident angle, point density, and curvature, thus improving image visual quality and detail recognition.

Description

Technical Field

[0001] This invention belongs to the field of surveying and tunnel engineering inspection technology, specifically relating to a method, system, platform and storage medium for generating and processing tunnel orthophotos based on global constraint optimization and contour refinement distance determination. Background Technology

[0002] With the continuous expansion of tunnel construction in my country, safety monitoring and maintenance during tunnel operation has become a key task in ensuring transportation safety. Generating grayscale images of tunnel interior walls using mobile laser scanning technology has become a mainstream method for tunnel defect detection and structural health monitoring due to its advantages such as high precision, high efficiency, and insensitivity to ambient light.

[0003] In the prior art, a method, system, platform, and storage medium for high-precision generation and processing of tunnel orthophotos based on adaptive fitting and fixed-distance equal division are disclosed (hereinafter referred to as "the prior application"). The technical solution includes using dynamic filtering (such as Kalman filtering) combined with cubic spline interpolation for longitudinal calibration, using the sleeper as the geometric reference to correct mileage errors; employing a recursive simplification method to adaptively fit the tunnel cross-section polygon, and using the polygon's centroid as the projection center, dividing the entire polygon's perimeter into fixed-distance equal divisions to establish a mapping relationship between pixels and circumferential arc lengths. This solution is the first to realize the generation of orthophotos of tunnels of arbitrary shapes, providing fundamental technical support for tunnel defect detection.

[0004] However, subsequent engineering practice revealed that the previously applied scheme still had certain technical bottlenecks, such as the global optimality problem of longitudinal calibration, the robustness problem of complex cross-sectional contour extraction, and the visual quality problem of grayscale assignment.

[0005] Regarding the global optimality problem of longitudinal calibration, the dynamic filtering method used in the previous application is essentially based on real-time tracking and correction of the process model. However, it is difficult to achieve global optimal compensation for complex nonlinear velocity fluctuations caused by track irregularities and carrier sway. Longitudinal mileage errors can still reach several centimeters under certain adverse conditions, affecting the geometric consistency of long-distance tunnel images.

[0006] Regarding the robustness of complex cross-sectional contour extraction, the recursive simplified fitting method in the previous application can generate polygonal contours, but when faced with complex cross-sections containing local protruding auxiliary structures such as pipelines, cable troughs, and pedestrian platforms, the fitted contour is easily affected by the auxiliary structures, resulting in inaccurate extraction of the main lining contour, which in turn affects the accuracy of subsequent equal division.

[0007] Regarding the visual quality issues of grayscale assignment, the previous application used direct assignment or simple averaging for grayscale mapping, without considering the impact of laser incident angle, local point density, and curvature variations on intensity information. The generated images exhibit uneven grayscale and blurred details in certain areas (such as the transition zone between straight walls and curved surfaces, and around auxiliary structures), affecting the accuracy of human visual interpretation and AI recognition.

[0008] Therefore, in view of the above-mentioned technical problems and defects, there is an urgent need to design and develop a method, system, platform and storage medium for generating and processing tunnel orthophotos based on global constraint optimization and contour refinement distance determination. Summary of the Invention

[0009] To overcome the shortcomings and difficulties of the existing technology, the present invention provides a method, system, platform and storage medium for generating and processing tunnel orthophotos based on global constraint optimization and contour refinement distance. Based on the prior application, it further improves the global optimality of longitudinal calibration, enhances the robustness of circumferential contour extraction and improves the visual quality of grayscale images.

[0010] The first objective of this invention is to provide a method for generating and processing tunnel orthophotos based on global constraint optimization and contour refinement distance determination; the second objective of this invention is to provide a system for generating and processing tunnel orthophotos based on global constraint optimization and contour refinement distance determination; the third objective of this invention is to provide a platform for generating and processing tunnel orthophotos based on global constraint optimization and contour refinement distance determination; and the fourth objective of this invention is to provide a computer-readable storage medium.

[0011] The first objective of this invention is achieved as follows: the method comprises the following steps:

[0012] Generate and acquire first data corresponding to the tunnel; wherein, the first data is the original point cloud data inside the tunnel, and the original point cloud data includes spatial coordinate information data and reflection intensity information data;

[0013] Based on a longitudinal geometric datum with known physical spacing, and combined with a global constraint optimization method, the first data is longitudinally calibrated and the second data corresponding to the first data is generated; wherein, the second data is a point cloud cross-section sequence with uniformly distributed scan lines and accurate mileage in the longitudinal direction of the tunnel.

[0014] The contour refinement and distance determination process is performed on each section in the second data to generate third data corresponding to the second data; wherein, the third data is tunnel orthophoto data without geometric distortion; the contour refinement and distance determination process includes extracting the main lining contour polygon of each section, and based on the main lining contour polygon, establishing a mapping relationship between pixels and circumferential real physical distance through a distance equal division method.

[0015] The second objective of this invention is achieved as follows: the system is used to implement the tunnel orthophoto generation and processing method based on global constraint optimization and contour refinement distance determination. The system includes: a data generation and acquisition unit, used to generate and acquire first data corresponding to the tunnel; wherein, the first data is the original point cloud data inside the tunnel, and the original point cloud data includes spatial coordinate information data and reflection intensity information data.

[0016] The first data processing and generation unit is used to longitudinally calibrate the first data based on a longitudinal geometric benchmark with known physical spacing and in combination with a global constraint optimization method, and generate second data corresponding to the first data; wherein, the second data is a point cloud cross-section sequence with uniformly distributed scan lines and accurate mileage in the longitudinal direction of the tunnel.

[0017] The second data processing and generation unit is used for contour refinement and distance determination processing of each section in the second data, and to generate third data corresponding to the second data; wherein, the third data is tunnel orthophoto data without geometric distortion; the contour refinement and distance determination processing includes extracting the main lining contour polygon of each section, and based on the main lining contour polygon, establishing a mapping relationship between pixels and circumferential real physical distance through a distance equal division method.

[0018] The third objective of this invention is achieved as follows: the platform includes a processor, a memory, and a control program for a tunnel orthophoto generation and processing platform based on global constraint optimization and contour refinement distance determination; wherein the control program for the tunnel orthophoto generation and processing platform based on global constraint optimization and contour refinement distance determination is executed on the processor, and the control program for the tunnel orthophoto generation and processing platform based on global constraint optimization and contour refinement distance determination is stored in the memory; the control program for the tunnel orthophoto generation and processing platform based on global constraint optimization and contour refinement distance determination implements the tunnel orthophoto generation and processing method based on global constraint optimization and contour refinement distance determination.

[0019] The fourth objective of this invention is achieved as follows: the computer-readable storage medium stores a control program for a tunnel orthophoto generation and processing platform based on global constraint optimization and contour refinement distance determination, and the control program for the tunnel orthophoto generation and processing platform based on global constraint optimization and contour refinement distance determination implements the tunnel orthophoto generation and processing method based on global constraint optimization and contour refinement distance determination.

[0020] This invention generates and acquires first data corresponding to a tunnel through a method; wherein the first data is raw point cloud data inside the tunnel, the raw point cloud data including spatial coordinate information data and reflection intensity information data; based on a longitudinal geometric reference with known physical spacing, and combined with a global constraint optimization method, the first data is longitudinally calibrated and processed to generate second data corresponding to the first data; wherein the second data is a sequence of point cloud cross-sections with uniformly distributed scan lines and accurate mileage along the longitudinal direction of the tunnel; each cross-section in the second data is refined and distance-fixed, and a third data corresponding to the second data is generated; wherein the third data is orthophoto data of the tunnel without geometric distortion; the contour refinement and distance-fixing processing includes... This method includes extracting the main lining contour polygon of each cross section, and establishing a mapping relationship between pixels and the actual physical distance in the circumferential direction based on the main lining contour polygon through a fixed-distance equal division method. It also includes the corresponding system, platform, and storage medium. This method can overcome the limitations of local correction in dynamic filtering, achieve global optimal compensation for longitudinal mileage errors, and improve the geometric consistency of long-distance tunnel images. Furthermore, it accurately extracts the main lining contour in complex cross sections, effectively eliminating interference from auxiliary structures such as pipelines and cable troughs, ensuring that the fixed-distance equal division is based on the actual tunnel wall. Based on the fixed-distance equal division, it achieves adaptive grayscale equalization, compensating for intensity unevenness caused by changes in laser incident angle, point density, and curvature, thereby improving the visual quality and detail recognition of the image.

[0021] In other words, based on the adaptive fitting and fixed-distance equal division framework, this invention replaces traditional dynamic filtering with a global constraint optimization method in the longitudinal direction. It uses known physical distances as hard constraints and combines them with a smoothing regularization term to achieve globally optimal compensation for nonlinear velocity fluctuations, improving the longitudinal geometric consistency of long-distance tunnel images. In the circumferential tunnel cross-section adaptive fitting method (TSAM), feature-priority retention dynamic threshold refinement and area-shape consistency vertex optimization, combined with automatic identification and removal technology for auxiliary structures, effectively solve the problem of interference from pipelines, cable troughs, etc., on the main lining contour under complex cross-sections, improving the robustness and accuracy of contour extraction. Simultaneously, it introduces a density-curvature adaptive grayscale weight mapping mechanism, dynamically adjusting weights according to local point density and curvature, effectively compensating for intensity unevenness caused by laser incident angle and local geometric changes, resulting in natural grayscale transitions and clear details in the image. Compared with existing technologies, the distortion-free orthophotos generated by this invention achieve comprehensive optimization in three dimensions: geometric accuracy, contour fidelity, and visual quality, providing a high-fidelity data foundation for intelligent tunnel defect identification, auxiliary facility statistics, and digital operation and maintenance. 2 Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a schematic diagram of the technical route of the tunnel orthophoto generation and processing method based on global constraint optimization and contour refinement distance determination according to the present invention.

[0024] Figure 2 This is a schematic diagram illustrating outlier removal in a tunnel orthophoto generation and processing method based on global constraint optimization and contour refinement distance determination according to the present invention.

[0025] Figure 3 This is a schematic diagram of point cloud sparsification in a tunnel orthophoto generation and processing method based on global constraint optimization and contour refinement distance according to the present invention.

[0026] Figure 4 This is a schematic diagram of the tunnel point cloud coordinate system in the tunnel orthophoto generation and processing method based on global constraint optimization and contour refinement distance according to the present invention.

[0027] Figure 5 This is a schematic diagram illustrating the annotation of trackless pillow scene reflection targets in a tunnel orthophoto generation and processing method based on global constraint optimization and contour refinement distance determination according to the present invention.

[0028] Figure 6 This is a schematic diagram of the Canny edge detection and Hough line detection results of a tunnel orthophoto generation and processing method based on global constraint optimization and contour refinement distance according to the present invention.

[0029] Figure 7 This is a schematic diagram of the longitudinal calibration of point cloud in a tunnel orthophoto generation and processing method based on global constraint optimization and contour refinement distance according to the present invention.

[0030] Figure 8 This is a schematic diagram of the tunnel cross-section adaptive method of the tunnel orthophoto generation and processing method based on global constraint optimization and contour refinement distance of the present invention.

[0031] Figure 9 This is a schematic diagram illustrating the removal of prominent contours in a tunnel orthophoto generation and processing method based on global constraint optimization and contour refinement distance determination according to the present invention.

[0032] Figure 10 This is a schematic diagram of the distance division in the tunnel orthophoto generation and processing method based on global constraint optimization and contour refinement distance determination according to the present invention.

[0033] Figure 11 This is a schematic diagram comparing images of tunnels with different cross-sectional shapes before and after correction, based on a tunnel orthophoto generation and processing method for global constraint optimization and contour refinement distance determination according to the present invention.

[0034] Figure 12 This is a detailed comparison diagram of images before and after correction in the tunnel orthophoto generation and processing method based on global constraint optimization and contour refinement distance according to the present invention.

[0035] Figure 13 This is a schematic diagram comparing the length measurement accuracy (20 samples) of a tunnel orthophoto generation and processing method based on global constraint optimization and contour refinement distance determination according to the present invention.

[0036] Figure 14 This is a schematic diagram comparing the area measurement accuracy of a circular fitting method based on RANSAC (20 samples) in a tunnel orthophoto generation and processing method based on global constraint optimization and contour refinement distance determination according to the present invention.

[0037] Figure 15 This is a schematic diagram comparing the area measurement values ​​before and after correction of a tunnel orthophoto generation and processing method based on global constraint optimization and contour refinement distance according to the present invention.

[0038] Figure 16 This is a schematic diagram of the image area measurement results (20 samples) of a tunnel orthophoto generation and processing method based on global constraint optimization and contour refinement distance according to the present invention.

[0039] Figure 17 This is a schematic diagram comparing tunnel images generated by different methods of the tunnel orthophoto generation and processing method based on global constraint optimization and contour refinement distance, and images generated by the method proposed in this paper.

[0040] Figure 18 This is a schematic diagram of the process steps of a tunnel orthophoto generation and processing method based on global constraint optimization and contour refinement distance determination according to the present invention.

[0041] Figure 19 This is a schematic diagram of the tunnel orthophoto generation and processing system architecture based on global constraint optimization and contour refinement distance determination according to the present invention.

[0042] Figure 20 This is a schematic diagram of the tunnel orthophoto generation and processing platform architecture based on global constraint optimization and contour refinement distance determination according to the present invention.

[0043] Figure 21 This is a schematic diagram of a computer-readable storage medium architecture in one embodiment of the present invention. Detailed Implementation

[0044] To facilitate a clearer understanding of the objectives, technical solutions, and advantages of this invention, the invention will be further described below in conjunction with the accompanying drawings and specific embodiments. Those skilled in the art can easily understand other advantages and effects of this invention from the content disclosed in this specification.

[0045] This invention can also be implemented or applied through other different specific examples, and various details in this specification can also be modified and changed based on different viewpoints and applications without departing from the spirit of this invention.

[0046] It should be noted that if the embodiments of the present invention involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of the components in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicators will also change accordingly.

[0047] Furthermore, if the embodiments of this invention involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Secondly, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0048] Preferably, the tunnel orthophoto generation and processing method based on global constraint optimization and contour refinement distance determination of the present invention is applied in one or more terminals or servers. The terminal is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0049] The terminal can be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal can interact with the customer via a keyboard, mouse, remote control, touchpad, or voice control device.

[0050] This invention provides a method, system, and platform for generating and processing tunnel orthophotos based on global constraint optimization and contour refinement.

[0051] like Figure 18 The diagram shown is a flowchart of a tunnel orthophoto generation and processing method based on global constraint optimization and contour refinement distance provided in an embodiment of the present invention.

[0052] In this embodiment, the tunnel orthophoto generation and processing method based on global constraint optimization and contour refinement distance can be applied to terminals or fixed terminals with display functions. The terminals are not limited to personal computers, smartphones, tablets, desktop computers or all-in-one computers with cameras, etc.

[0053] The tunnel orthophoto generation and processing method based on global constraint optimization and contour refinement distance determination can also be applied to a hardware environment consisting of a terminal and a server connected to the terminal via a network. The network includes, but is not limited to, a wide area network (WAN), a metropolitan area network (MAN), or a local area network (LAN). The tunnel orthophoto generation and processing method based on global constraint optimization and contour refinement distance determination in this embodiment of the invention can be executed by a server, by a terminal, or by both a server and a terminal.

[0054] For example, for a terminal that needs to perform tunnel orthophoto generation and processing based on global constraint optimization and contour refinement distance determination, the tunnel orthophoto generation and processing function based on global constraint optimization and contour refinement distance determination provided by the method of this invention can be directly integrated into the terminal, or a client for implementing the method of this invention can be installed. Alternatively, the method provided by this invention can also run on servers or other devices in the form of a Software Development Kit (SDK), providing an interface for the tunnel orthophoto generation and processing function based on global constraint optimization and contour refinement distance determination in the form of an SDK. Terminals or other devices can then implement the tunnel orthophoto generation and processing function based on global constraint optimization and contour refinement distance determination through the provided interface. The invention will be further described below with reference to the accompanying drawings.

[0055] like Figures 1-18As shown, this invention provides a method for generating and processing tunnel orthophotos based on global constraint optimization and contour refinement distance determination. The method includes the following steps: S001, generating and acquiring first data corresponding to the tunnel; wherein, the first data is the original point cloud data inside the tunnel, and the original point cloud data includes spatial coordinate information data and reflection intensity information data; S002, based on a longitudinal geometric reference with known physical spacing, and combined with a global constraint optimization method, longitudinally calibrating the first data, and generating second data corresponding to the first data; wherein, the second data is a point cloud cross-section sequence with uniformly distributed scan lines and accurate mileage in the longitudinal direction of the tunnel; S003, contour refinement distance determination processing for each cross-section in the second data, and generating third data corresponding to the second data; wherein, the third data is tunnel orthophoto data without geometric distortion; the contour refinement distance determination processing includes extracting the main lining contour polygon of each cross-section, and based on the main lining contour polygon, establishing a mapping relationship between pixels and circumferential real physical distances through a distance equal division method.

[0056] The longitudinal geometric reference includes the sleeper centerline automatically detected from the initial grayscale image generated from the original point cloud using image processing technology, or artificial markers with known arc length spacing laid out by external measurement means.

[0057] The process of longitudinally calibrating the first data based on a longitudinal geometric datum with known physical spacing, and combining a global constraint optimization method, to generate second data corresponding to the first data, further includes: S0021, constructing a continuous first function corresponding to the scan line number as the independent variable and the corrected mileage as the dependent variable; wherein, the first function is a mapping function; wherein, the first function is a natural cubic smooth spline function; S0022, using the known physical spacing of the longitudinal geometric datum as a hard constraint condition, and introducing a smoothing regularization term, constructing a corresponding second function; wherein, the second function is an objective function; the smoothing regularization term is an integral penalty term on the second derivative of the spline function; S0023, based on the second function, calculating and generating a corresponding first function, and mapping the pseudo mileage of each original scan line to the corrected real physical mileage.

[0058] The contour refinement and distance determination process for each cross section in the second data, and the generation of third data corresponding to the second data, further includes: S0031, sorting and processing the cross section point cloud, and constructing the corresponding initial polygonal contour; S0032, combining a feature-priority retention dynamic threshold refinement strategy, inserting new vertices layer by layer on the basis of the initial polygonal contour to approximate the true shape of the point cloud; S0033, optimizing the vertex position of the polygonal contour after the vertices are inserted by minimizing an energy function that includes an area fidelity term and a side length smoothing term, and generating the corresponding main lining contour polygon.

[0059] The dynamic threshold refinement strategy that prioritizes feature retention includes prioritizing points that are most far from the current edge, have a large curvature change, or have a significant azimuth change when determining whether to insert a new vertex.

[0060] The contour refinement and distance-fixed processing of each section in the second data, and the generation of third data corresponding to the second data, further includes: S0034, identifying and deleting contour segments corresponding to local protruding auxiliary structures on the main lining contour polygon, and creating a corresponding simplified main lining contour; S0035, sampling each edge of the simplified main lining contour at a fixed distance according to a preset target circumferential resolution, and generating a series of corresponding division points; S0036, assigning corresponding gray values ​​to the division points based on the correspondence between the division points and the original point cloud points, combined with adaptive weights; wherein, the adaptive weights are dynamically adjusted based on the local point density and / or curvature of the original point cloud points.

[0061] The step of identifying and deleting the contour segments corresponding to the local protruding auxiliary structures on the main lining contour polygon and creating the corresponding simplified main lining contour further includes: S00341, identifying and cutting off the continuous vertex sequence corresponding to the auxiliary structures based on the radial distance change of the polygon vertex relative to the centroid of the cross section and / or the curvature change at the vertex.

[0062] Specifically, in this embodiment of the invention, a high-precision orthophoto grayscale image reconstruction and calibration framework applicable to tunnels of arbitrary shapes is proposed. The core of this framework lies in the organic combination of longitudinal and circumferential calibration: longitudinal calibration uses sleeper spacing or manually laid high-precision reference points as reliable geometric benchmarks, integrating automated edge detection, global smoothing spacing constraint optimization, and cubic spline interpolation to achieve high-precision correction of uniform scan line distribution and mileage accumulation error, supporting unified processing of railway tunnels with sleepers and subway tunnels without sleepers; circumferential calibration first develops a tunnel cross-section adaptive fitting method (TSAM), which combines a feature-priority retention dynamic threshold layer-by-layer refinement mechanism with vertex smoothing optimization driven by area and shape consistency to generate polygonal contours with high robustness to various complex cross-sections such as circles, horseshoes, irregular shapes, and squares. Subsequently, an innovative distance correction algorithm based on contour optimization is constructed, which performs strict distance equal-division sampling on each side of the simplified main lining contour, and introduces a grayscale adaptive weight mapping mechanism based on local point density and curvature, ultimately achieving accurate correspondence between grayscale pixels and the actual circumferential physical distance. The orthophotos generated by this method exhibit minimal distortion, strong geometric consistency, uniform brightness, and rich detail. They not only support high-precision measurement of the length and area of ​​defects down to the centimeter level, but can also be widely applied in fields such as intelligent identification of tunnel defects, statistics of ancillary facilities, structural health monitoring, and service performance evaluation. For example... Figure 1 The diagram shown is the technical roadmap for this article.

[0063] In this solution, regarding laser point cloud data acquisition and preprocessing, the acquisition of point cloud data is based on the fundamental work of laser tunnel deformation and defect measurement. This solution utilizes a self-developed TLSD track inspection vehicle, with Faro, Leica, and Z+F lidars as the main measuring equipment. A schematic diagram of the equipment is shown below. Figure 2 As shown, during the mobile measurement process, as the inspection vehicle moves forward, the lidar scans along a two-dimensional spiral path to capture point cloud information of the tunnel cross-section. The data acquired by the equipment is all relative measurement data in the scanner coordinate system, requiring no absolute coordinate transformation. Since the point cloud of the tunnel inner wall acquired by the TLSD system inherently contains intensity information, the tunnel cross-section can be presented in image form based on this intensity information.

[0064] The algorithm implementation in this figure uses data collected by the Z+F PROFILER9012 lidar. This lidar model exhibits excellent performance and is highly representative. The parameters of the lidar mounted on the mobile detection equipment are shown in Table 1.

[0065] Table 1 LiDAR Parameters

[0066] LiDAR Name Distance measurement accuracy Scan speed Angle accuracy Vertical field of view Z+F PROFILER9012 ±0.1mm 1,016,000 points / second 0.002° / 0.002° 360°

[0067] For data preprocessing, after acquiring the raw point cloud, it is necessary to preprocess it to ensure data accuracy and usability. First, point cloud denoising is performed to remove outliers and anomalies; second, point cloud downsampling is performed to reduce redundant data and improve computational efficiency; finally, intensity enhancement is carried out by piecewise linear stretching

[45] to optimize grayscale contrast and highlight the surface texture and disease features of the tunnel. This sequence conforms to the standard pipeline of point cloud processing: first, clean up the data quality, then control the scale, and finally improve the appearance recognizability, providing a high-quality foundation for subsequent longitudinal and circumferential calibration.

[0068] Point cloud denoising aims to remove outliers and erroneous points, thereby significantly improving the accuracy and recognizability of the subsequently generated orthophoto grayscale image. This scheme employs a two-step combined strategy: first, based on the distance threshold method, the distance from each point on each cross-section to the origin is calculated, and obvious outliers exceeding a preset threshold are removed; then, the K-means clustering algorithm is applied to further identify and remove abnormal points (such as noise clusters) that are too close to the tunnel surface. Figure 2 As shown in the diagram, the process is illustrated, and the results demonstrate that the method effectively eliminates obvious outliers and improves the overall quality of the point cloud.

[0069] Excessive density in the original point cloud significantly increases computational burden, necessitating thinning. This scheme combines the local density of the point cloud with the target grayscale image resolution requirements, employing a thinning algorithm based on uniformly distributed random numbers. Data simplification is achieved by randomly selecting a subset. A schematic diagram of the algorithm is shown below. Figure 3 As shown in the experiment, a high thinning ratio (40%) significantly improves computation speed but leads to decreased accuracy, loss of reconstruction details, and increased errors. A medium thinning ratio (20%) achieves the best balance between computational efficiency and accuracy, with the smallest reconstruction error. A low thinning ratio (10%) produces the closest reconstruction result to the original data, but offers limited computational speedup. If the thinning ratio is too high (>40%), insufficient point cloud data may amplify distance and area calculation errors. Therefore, the thinning ratio should be determined based on a comprehensive consideration of specific accuracy requirements and computational resources.

[0070] After denoising and thinning, the point cloud intensity information is converted into grayscale values ​​and enhanced using a piecewise linear stretching algorithm. This algorithm constructs a piecewise linear transformation function to achieve targeted adjustments: enhancing the dynamic range of low-contrast regions while compressing excessive stretching in high-contrast regions, thereby optimizing overall contrast and improving detail recognition. This method is particularly suitable for tunnel scenes with non-uniform laser intensity signal distribution, effectively highlighting surface texture and defect features. Subsequently, the grayscale image matrix is ​​filled according to the original point cloud storage order before correction to generate an initial grayscale image of the tunnel inner wall. This initial image retains the original intensity information of the point cloud but has not yet eliminated the effects of non-uniform scan line distribution and projection distortion, providing a reliable input basis for subsequent longitudinal and circumferential calibration modules.

[0071] For the point cloud calibration algorithm, this scheme first needs to clarify the coordinate system definition of the tunnel point cloud to facilitate calculation. For example... Figure 4 As shown, the point cloud collected by the lidar in one rotation is considered a complete cross-section, where the forward direction is defined as the Y-axis, and the X and Z axes together form the cross-sectional plane. All points within the same cross-section have the same Y-value, which represents the mileage information. The Y-value of the original point cloud is assigned based on the speed of the inspection vehicle, but due to factors such as track irregularities, uneven track surfaces, and vehicle swaying, the actual speed often deviates from the preset uniform speed, resulting in cumulative errors in the mileage value. This error not only distorts the true distance mapping between cross-sections in the generated grayscale image but also significantly reduces the accuracy of subsequent measurements of the length and area of ​​defects. Therefore, the longitudinal mileage information of the point cloud must be calibrated, i.e., longitudinal calibration. Simultaneously, due to the large number of points and irregular shapes in each cross-section of the original tunnel point cloud, accurately establishing the correspondence between grayscale pixels and real spatial geometric distances while calibrating the image is another core challenge, requiring circumferential calibration.

[0072] The technical approach of this scheme is therefore divided into two main modules: longitudinal point cloud calibration and circumferential point cloud calibration. Its core objective is to establish a quantitative mapping relationship between the pixels of the grayscale image generated by laser point cloud and the actual geometric distance in three-dimensional space. The innovation of this scheme lies in proposing a complete high-precision orthophoto grayscale image reconstruction and calibration framework applicable to tunnels of arbitrary shapes. For longitudinal calibration, this scheme proposes a global smoothing spacing constraint optimization method. This method, based on natural cubic smoothing splines combined with hard constraints and second-order smoothing regularization terms, uses the sleeper spacing or a known arc length reference point as the geometric benchmark. It achieves effective correction of uniform scan line distribution and cumulative mileage error, and supports unified high-precision processing for both sleeper-equipped and sleeperless tunnels, significantly improving the robustness and engineering applicability of longitudinal geometric mapping. For circumferential calibration, this scheme proposes an adaptive fitting method for tunnel cross-sections. By combining feature-priority retention with dynamic threshold layer-by-layer refinement, a tunnel scene priority insertion strategy, and vertex smoothing optimization driven by area and shape consistency, it achieves highly robust polygon contour extraction for various complex cross-sections such as circles, horseshoes, irregular shapes, and squares. Subsequently, this scheme innovatively designs a fixed-distance correction algorithm based on contour optimization. It performs strict fixed-distance equal-division sampling on each side of the simplified main lining contour and introduces an adaptive grayscale weight mapping mechanism for local point density and curvature. Finally, it constructs a distortion-free orthophoto grayscale image that accurately corresponds to the pixel and the real circumferential physical distance, supporting centimeter-level defect measurement and area statistics.

[0073] In longitudinal point cloud calibration of tunnels, this paper proposes an automated calibration method based on global smoothing spacing constraint optimization to address the limitations of traditional longitudinal calibration methods, such as low efficiency due to manual annotation, neglect of nonlinear errors by linear interpolation, and uneven distribution of scan lines. This method utilizes fully automated sleeper edge detection and recognition, using sleeper spacing or known arc length reference points as reliable geometric benchmarks. It integrates natural cubic smoothing splines with hard constraints and second-order smoothing regularization terms to achieve high-precision correction of uniform scan line distribution and cumulative mileage errors. This method is suitable for complex track environments in both sleeper-equipped railway tunnels and sleeperless subway tunnels.

[0074] For reference feature extraction, to overcome the subjective errors and low efficiency of manual annotation, this module uses image processing technology to automatically extract the sleeper edges as geometric references. The standard sleeper spacing is 0.6m, with an allowable deviation of ±20mm. The processing flow begins by applying Gaussian filtering to the uncorrected grayscale image (generated from point cloud intensity) to smooth noise and preserve sleeper edge features. The filter kernel size is 3×3, the standard deviation σ=1.5, and the filtering formula is:

[0075] (1)

[0076] In the formula, Represents the original pixel grayscale value. These are the filtered pixel values. The standard deviation is a Gaussian distribution. This Gaussian filter effectively smooths noise caused by uneven illumination or intensity fluctuations, while maintaining the continuity of the edges of sleepers or reflectors, providing a high-quality image foundation for subsequent accurate extraction.

[0077] For scenarios involving railway sleepers, the sleepers appear as high-grayscale horizontal lines in grayscale images within railway tunnels. Therefore, the Canny edge detection algorithm combined with Hough transform is used for fully automatic extraction. The main reason for choosing railway sleepers as the longitudinal correction benchmark is that, for example, some subway systems have a standardized sleeper spacing of 0.6m (allowing ±20mm deviation). This fixed physical spacing is highly consistent in railway design and construction, serving as a highly reliable real mileage benchmark and effectively correcting for non-uniform distribution of scan lines and cumulative mileage errors caused by vehicle vibration, wheel slippage, or track irregularities. Compared to circumferential seams or manual calibration points, sleepers are densely distributed, have sharp grayscale contrast, and are numerous, facilitating automatic image detection without requiring additional installation.

[0078] The Canny algorithm first uses the Sobel operator to calculate the image gradient. The gradient magnitude is used for subsequent non-maximum suppression, and the gradient direction is used to determine the edge normal.

[0079] (2)

[0080] In the formula, and The gradients in the horizontal and vertical directions are respectively used. The Canny algorithm generates continuous sleeper edge lines through thresholding and edge connection. Then, the Hough transform is used to identify straight line features in the edges, as shown in the formula:

[0081] (3)

[0082] In the formula, This represents the distance from the straight line to the origin of the image. This represents the angle between the line and the horizontal axis. The Hough transform modulates edge points from image space. Convert to parameter space The system searches for voting peaks in the parameter space, identifies those that conform to the geometric characteristics of the sleepers, and records the center positions of the sleepers, thereby generating a sequence of track mileage points. each This indicates the longitudinal position of a sleeper. This automated process significantly improves efficiency, generates accurate sleeper mileage sequences, and provides a reliable benchmark for subsequent calibration.

[0083] In sleeper-less scenarios, due to the lack of standardized periodic geometric features (such as the fixed spacing of railway sleepers), longitudinal mileage cannot be reliably constrained by natural scales automatically extracted from images. Therefore, it is necessary to introduce high-precision manual reference points to ensure correction accuracy. When setting up the total station, Leica black and white checkerboard square reflectors with dimensions of 40mm×40mm or 60mm×60mm are used. The black and white checkerboard pattern forms a distinct high-value area on the LiDAR intensity map, facilitating manual center identification. The total station is set up at fixed intervals, either freely or by backsight of known control points, starting from the tunnel entrance or a known station number, along the tunnel's designed centerline.

[0084] like Figure 5 As shown, during the deployment process, the total station, referencing tunnel design drawings or BIM models, calculates the cumulative arc length in real time based on centerline parameters to ensure that the spacing w between adjacent reflectors is a known fixed value (typically 30m for straight sections, and appropriately densified to 10m for curved sections). On straight sections, the total station collects the three-dimensional coordinates of each reflector and directly adds the straight-line distance as the spacing w. On curved sections, the total station calculates the arc length; the arc length of a circular curve is equal to the radius R multiplied by the deflection angle Δθ, and for transition curves, an integral formula is used to record the actual arc length spacing w. This fixed-spacing deployment method ensures that each reflector position corresponds to a known physical distance in advance, requiring only manual labeling of the center row number in subsequent inspection images. This allows for a direct correspondence with the actual arc length mileage, avoiding the complexity of subsequent individual measurements or registration. The placement should prioritize stable areas on the sidewalls, at a height of 1.5–2.0m above the base plate, avoiding areas with water accumulation or disturbance on the inside of bends. Use specialized adhesives or magnetic supports for fixation, ensuring the surface is vertical and unobstructed. After placement, a reference point sequence is created, including number, three-dimensional coordinates, cumulative arc length mileage along the centerline, and archived photographs.

[0085] After the inspection, manually label the center image row number of each reflector in the uncorrected grayscale image. Based on the known arc length interval w preset by the total station, an observation sequence is established:

[0086] (4)

[0087] Correspondence with actual relative arc length mileage:

[0088] (5)

[0089] In the formula, This is a pseudo-relative mileage based on the scan sequence number (starting point set to 0). This represents the position of the j-th reference point (center of the sleeper or center of the reflector) in the pseudo-mileage coordinate system. w represents the known physical spacing (for scenarios with sleepers, w is the sleeper spacing; for scenarios without sleepers, w is the preset arc length spacing of the total station). Although the image projection spacing of the turning segment is locally compressed or stretched due to the curve geometry, w is the true centerline arc length. This hard constraint directly corrects the nonlinear drift caused by the curve, including errors caused by vehicle lateral offset, attitude tilt, and angular velocity changes. When the reflectors are sparse, the turning error mainly comes from the vehicle trajectory deviating from the centerline, LiDAR projection distortion, and the pseudo-mileage linear assumption; these errors are effectively suppressed by the second-order smoothing term of the subsequent global smoothing spline optimization. The hard constraint ensures that the mileage at each reflector strictly matches the preset true arc length.

[0090] In the scheme of this invention, for global smooth spacing constraint optimization and uniform resampling, the reference point observation sequence has been extracted above. ,in This is a pseudo-relative mileage based on scan sequence number (starting point set to 0). To achieve unified high-precision longitudinal correction for both sleeper-equipped and sleeper-free scenarios, this scheme proposes a global smoothing spacing constraint optimization method. This method learns a continuous mapping function. Minimize the objective function It is the true relative mileage value of the i-th scan line after longitudinal correction.

[0091] (6)

[0092] In the formula, the first term is the data fitting term (hard constraint), the squared error, and the forced function. The value at each reference point j strictly matches the actual mileage. This achieves scale correction and spacing constraints; the second term is a smoothing regularization term (penalty term), which, by integrating the square of the second derivative, penalizes the drastic changes in the curvature of the function, making... The overall design is smooth and physically sound, avoiding overfitting noise or local abrupt changes. The smoothing parameter controls the trade-off between fitting accuracy and smoothness: The smaller the value, the closer the function is to the reference point; The larger the value, the more linear or low-order smooth the function tends to be.

[0093] function Represented using naturally cubic smooth splines, its specific form is as follows:

[0094] (7)

[0095] In the formula, The coefficients of the linear terms describe the global trend; represents the spline coefficients at the j-th reference point, and the plus sign indicates the truncation of the power function; To truncate the power basis functions, only Effective immediately. This form guarantees... It is continuous, first-order and second-order differentiable, smoothly connected between reference points, and satisfies natural boundary conditions at the boundaries.

[0096] The optimization solution process is as follows: Substituting the spline form Transformed into coefficients The quadratic programming problem; constructing a design matrix based on all reference points. With smoothing penalty matrix Then, solve the system of linear equations.

[0097] (8)

[0098] Solve to obtain the spline coefficients c, and determine the continuous mapping function. .in, Corresponding to actual relative mileage. After optimization, Arbitrary pseudo-mileage Mapping to calibration mileage This is used for subsequent uniform resampling.

[0099] The standard scan line count num is the theoretical number of scan lines that each reference spacing w should contain, which can be roughly calculated as follows:

[0100] (9)

[0101] In the formula, w is the reference spacing and f is the scanning frequency. This is the nominal speed. This value is for reference only; due to speed fluctuations, the actual number of scan lines may deviate from num. After correction, num is used as a reference for the target resolution Δy. During resampling... Uniform sampling of target mileage points in the domain (L is the relative total length of the tunnel).

[0102] If the original scan line count is less than the target count, use cubic spline or linear interpolation to pad the lines, and bilinear interpolation to adjust the grayscale. If the original scan line count is greater than the target count, downsample and retain the closest value. Remove redundant rows. Output a relatively uniform mileage sequence and a grayscale image with longitudinal scaling correction. Through this global optimization framework, longitudinal calibration achieves unified support for both sleeper-equipped and sleeperless tunnels, relying only on known spacing constraints and smoothing regularization, with the starting point set to relative 0, significantly improving the algorithm's versatility, robustness, and engineering applicability.

[0103] For the calibration of tunnel point cloud circumferential point cloud, the present invention approximates the true contour of the cross section by fitting polygons to obtain true distance information, and then obtains the mapping relationship between pixels and true distance by projection center transformation and fixed distance equal division method to generate an orthophoto grayscale image with distance information without distortion.

[0104] To address the challenge of generating orthorectified grayscale images of tunnels of arbitrary shapes from moving laser scanning point clouds, this paper proposes an innovative polygon fitting method called "Tunnel Section Adaptive Method (TSAM)." This method simplifies point cloud data through dynamic thresholding and vertex smoothing adjustments, significantly improving fitting accuracy and robustness. Compared to traditional methods' sensitivity to noise and complex cross-sections (such as horseshoe-shaped or irregular tunnels), the TSAM method can adapt to various tunnel cross-section shapes, including circular, horseshoe-shaped, and irregular shapes. This ensures that the generated orthorectified grayscale image more closely approximates the actual tunnel cross-section in terms of geometry and area calculation, meeting the high requirements for tunnel defect detection and area measurement.

[0105] The TSAM method takes a single tunnel cross-section point cloud dataset as input. , where each point It includes 3D coordinates, which are projected onto the cross-sectional plane (XZ plane) during processing. The goal is to generate a closed vertex sequence with a number of vertices much smaller than N. ,satisfy This is done to preserve the main geometric features of the tunnel cross-section as much as possible while significantly reducing computational complexity.

[0106] Phase 1: Dynamic threshold refinement with feature-priority preservation

[0107] To construct the initial vertex sequence, first calculate the geometric centroid of the point cloud.

[0108] (10)

[0109] With the center of the form Using the origin as the coordinate point, assign all points in the point cloud to the azimuth angle. Sort to obtain an ordered sequence of closed loop points. Initial vertex sequence It consists of the largest outward convex points of the point cloud along the main directions (0°, 90°, 180°, 270°), and typically contains 4 vertices. This initial selection strategy fully considers the common geometric symmetry and main structural directions of tunnel cross-sections, and has strong engineering relevance.

[0110] The dynamic threshold refinement process sets an initial distance threshold. Where R is the diagonal length of the point cloud bounding box, and k ranges from 15 to 25 (this scheme defaults to k=20). For the current vertex sequence... Each edge in Calculate all point cloud points not included in the current sequence (from...) The perpendicular distance d from this edge:

[0111] (11)

[0112] If there exists a vertex on some edge that satisfies Then, new vertices are inserted according to the priority of tunnel scene features (priority order: ① the one with the largest deviation distance; ② points in areas with large curvature changes; ③ points with significant azimuth changes), and the original edge is split into two sub-edges. After each vertex insertion, the threshold is updated globally.

[0113] (12)

[0114] It is a dynamic threshold decay factor (or reduction factor) used to gradually decrease the distance threshold ε in each iteration, thereby achieving a layer-by-layer refinement from a "coarse outline" to a "fine outline". Simultaneously, it calculates the area of ​​the polygon enclosed by the current vertex sequence in real time. Reference area of ​​point cloud The relative error. Point cloud reference area polygon area. Calculated using the cross product formula of vertex coordinates.

[0115] (13)

[0116] In the formula, Here are the coordinates of the polygon's vertices. The refinement process terminates when any of the following conditions are met:

[0117] (14)

[0118] Or it may reach the maximum number of iterations.

[0119] Phase Two: Area-Shape Consistency-Driven Vertex Smoothing Optimization

[0120] Obtaining the initial vertex sequence Then, the vertex positions are further fine-tuned by minimizing energy so that the final contour simultaneously meets the requirements of overall geometric scale fidelity and uniform side length distribution.

[0121] The bi-objective energy function is defined as follows:

[0122] (15)

[0123] The first term is the relative area error square term (dimensionless), which is used to constrain the consistency between the overall scale of the polygon and the actual cross-section; the second term is the relative side length fluctuation square term (dimensionless), which is used to promote the uniformity of vertex distribution and reduce local geometric distortion during subsequent interval sampling. This represents the current average side length;

[0124] Weighting coefficient and Calibration was performed through comparative experiments on typical tunnel cross-sections (circular, horseshoe-shaped, and square) of Chongqing Metro Line 16. Under the premise of fixing other parameters, different ratios (1:1 to 10:1) were tested, and the relative area error, relative standard deviation of side length, and comprehensive score (area fidelity weight 0.6, uniformity weight 0.4) were used for evaluation. The results showed that when… The area error is too large. Uneven distribution of side lengths can easily lead to distortion in subsequent sampling. The 3:1 to 6:1 range shows the best overall performance, with 4:1 achieving the highest overall score while maintaining a relative area error of less than 0.5% and a relative standard deviation of side length of less than 10%, demonstrating the strongest adaptability to various cross-sections. Therefore, this scheme recommends a weight ratio range of [range missing]. A 4:1 ratio was consistently used in all experiments to effectively balance the requirements of geometric scale fidelity and vertex distribution uniformity. The vertex positions were iteratively updated using gradient descent.

[0125] (16)

[0126] In the formula, For learning rate, For energy function pairs The gradient is calculated. Iteration continues until the energy E converges. The final output is a sequence of closed vertices. While maintaining a geometric scale consistent with the actual tunnel cross-section, it possesses smooth and natural vertex distribution characteristics, laying a solid foundation for subsequent circumferential interval division and high-precision orthophoto generation. A schematic diagram of the TSAM method flow is shown below. Figure 8 As shown. Through the above two stages, TSAM achieves adaptive contour extraction from coarse to fine, taking into account both scale fidelity and shape uniformity. The entire process does not rely on any specific geometric prior assumptions and exhibits excellent robustness and engineering applicability for various tunnel cross-sections such as circles, horseshoes, irregular shapes, and squares.

[0127] In contour-optimized distance-corrected cross-sectional point clouds, this paper proposes an image generation method based on edge-by-edge strict distance division and adaptive gray-level weight mapping. This addresses the problems of uneven vertex distribution, unnatural gray-level transitions, local stretching artifacts, and detail loss that are easily caused by traditional equidistant sampling of the overall perimeter, particularly in areas with uneven vertex distribution, straight-to-curved transition zones, and local protruding auxiliary structures. First, this method identifies and removes local protruding auxiliary structures through joint detection of radial height and curvature abrupt changes, obtaining a simplified main lining contour. Then, it performs strict distance division sampling on each edge of the simplified contour to ensure high consistency of pixel physical resolution within each edge and globally. Finally, it introduces an adaptive gray-level weight mapping mechanism based on local point density and curvature to dynamically compensate for the gray-level assignment of sampling points. This significantly improves the visual continuity, detail balance, and disease identification friendliness of the image while maintaining centimeter-level geometric measurement accuracy.

[0128] The polygon vertex sequence obtained by TSAM fitting is ,in , The original total number of vertices. Represents the planar coordinates (XZ plane) of the i-th vertex.

[0129] First, calculate the Euclidean distance (actual side length) between adjacent vertices:

[0130] (17)

[0131] Based on this, the initial cumulative chord length parameter is obtained (used for initialization and as a reference for the original non-uniform distribution):

[0132] (18)

[0133] In the formula, S is the total perimeter of the original polygon.

[0134] The discrete curvature of each vertex is then calculated for adaptive adjustment of subsequent convex contour removal and smoothing intensity:

[0135] (19)

[0136] The process of identifying and deleting locally protruding attachment structures is as follows: Traverse the vertex sequence, taking the centroid c as the origin, for each vertex... Calculate its radial distance Identify segments whose continuous radial distance is significantly greater than the local mean or median. ,in and (This refers to the neighborhood statistical mean and standard deviation); it also assists in examining areas of drastic curvature changes (continuous) ) or local height abrupt change (radial difference between adjacent vertices) If a candidate region simultaneously satisfies both radial protrusion and abrupt curvature / height changes, it is identified as a locally protruding ancillary structure such as a pedestrian platform, cable, or pipeline. The connecting edges of this candidate region are removed from the main polygon, thus deleting it from the main contour; the remaining vertices are reconnected and closed to form a simplified main lining contour. The total number of its vertices is Circumference is A diagram illustrating the removal of prominent outlines is shown below. Figure 9 As shown.

[0137] Simplified main outline Each edge i (endpoint) ,length ), calculate the number of sampling points required for this edge based on the target circumferential resolution d:

[0138] (20)

[0139] Sampling is performed on the i-th edge with strict fixed intervals: starting from the starting point Initially, linear interpolation is performed along the edge direction at intervals to generate... sampling points :

[0140] (twenty one)

[0141] The sampling points of all edges are concatenated sequentially to form a global equidistant point sequence. (v=0,1,…,H-1), where the total number of points is... The physical spacing between pixels inside each edge is strictly 1. The global average resolution accurately matches the target d.

[0142] For each sampling point Use kd-tree to find distance The k most recent original point cloud points (k=8~15) carry the original intensity values ​​of the lidar. (Reflection intensity, preprocessed by piecewise linear stretching to convert to grayscale range [0, 255]). An adaptive grayscale weighting mapping mechanism based on local point density and curvature is introduced for grayscale calculation:

[0143] (twenty two)

[0144] Among them, weight Employing a density-curvature joint adaptive form:

[0145] (twenty three)

[0146] In the formula, Let m be the local point density of the m-th point in the point cloud. This is an estimate of the curvature near that point. For the m-th point cloud point The Euclidean distance is used. This weight appropriately increases the contribution of a single point in low-density or high-curvature regions, effectively compensating for the gray-scale sampling imbalance caused by the laser scanning incident angle effect and curvature changes, thereby alleviating the problems of blur stretching in straight-wall regions and detail compression in curved regions. Periodic boundary processing is used for the first and last pixels to ensure seamless closure. A schematic diagram of the adaptive gray-scale weight mapping mechanism of local point density and curvature is shown below. Figure 10 As shown.

[0147] The proposed method effectively eliminates the interference of protruding parts on the perimeter of the main contour by identifying and deleting locally protruding attachment structures. It more realistically reflects the length of the outer surface of the tunnel lining; strict equal-distance division on each side ensures that the physical interval of pixels is consistent within each side, the average resolution of the whole image accurately matches the target d, the circumferential geometric mapping relationship is precisely controllable, and the length and area measurement of surface defects of the main tunnel lining such as water seepage, spalling, and cracks are directly calculated by multiplying the number of pixels by the circumferential pixel resolution d, and then by the vertical pixel resolution. The measurement accuracy is completely equivalent to and more reliable than the traditional equal-distance method on each side; at the same time, the adaptive gray-scale weight mapping mechanism significantly improves the visual continuity and detail recognition of the image.

[0148] In the accuracy verification of this invention, to verify the accuracy and feasibility of the method, this scheme conducted on-site scanning of point cloud data from a section of Chongqing Metro Line 16 and Shenzhen Metro Line 17 Phase I. Chongqing Metro Line 16 employs a construction method combining mining and shield tunneling. The mining method sections have large cross-sectional areas and diverse shapes (including horseshoe-shaped, irregular, and near-square / rectangular tunnels), and the cross-sectional form changes within the section, effectively simulating common geometric variations in actual engineering projects. For the shield tunneling sections, curved segments were selected to compare the accuracy of this scheme's algorithm with traditional methods and to verify the image generation quality of the curved segments. Shenzhen Metro Line 17 Phase I is a typical shield tunneling line, using a circular cross-section throughout, with complex geological conditions including heterogeneous soft and hard strata and underground structures.

[0149] The proposed algorithm comprises three core modules: data preprocessing, longitudinal calibration, and circumferential calibration. Its computational complexity is primarily influenced by the point cloud size, the number of cross-sections, the number of iterations, and the sampling resolution. Taking a 25m tunnel segment as an example, on a test platform with an Intel Core i9-13900K CPU (24 cores), 64GB RAM, and an NVIDIA RTX4090 GPU, the total processing time (in seconds / 25m segment) for each module in single-threaded mode is as follows: preprocessing 18–32s; longitudinal calibration 4–9s; circumferential calibration approximately 23–44s. The total processing time is 45–85s / 25m. The current offline implementation cannot strictly meet the stringent real-time requirements of "data acquisition and generation simultaneously."

[0150] During data acquisition, the carrier's operating speed was 1.8 km / h (0.5 m / s), the lidar rotation speed was 100 rps, and the actual distance between adjacent longitudinal sections was 5 mm. In the actual measurements, the perimeter of the horseshoe-shaped section was 29.8 m before and 31.4 m after the shape change, the perimeter of the irregular-shaped section was 31.5 m, and the perimeter of the square section was 31.5 m. The experiment set the image height of the horseshoe-shaped and irregular-shaped tunnels to 5000 pixels, the circular shield tunnel to 4000 pixels, and the square tunnel to 4500 pixels. The calculated pixel resolutions are as follows: horseshoe-shaped cross-section (before change): 5.0 mm / px (vertical) × 5.96 mm / px (horizontal); after change: 5.0 mm / px × 6.28 mm / px; irregular cross-section: 5.0 mm / px × 6.30 mm / px; circular shield cross-section: 5.0 mm / px × 4.20 mm / px; square cross-section: 5.0 mm / px × 7.00 mm / px. Except for the circular cross-section, which has a slightly smaller horizontal resolution than the vertical resolution, the horizontal resolution of other cross-sections is between 5.96 and 7.00 mm / px, which is close to the vertical resolution of 5 mm / px. The maximum aspect ratio is 1:1.40, the overall geometric distortion is small, the image settings are reasonable, and it can well support centimeter-level defect identification and measurement.

[0151] Before experimental verification, the Canny edge detection algorithm was first used to extract the sleeper edge features from the original grayscale image, and then combined with Hough transform to identify straight line parameters, achieving fully automated sleeper labeling. This method significantly improves labeling efficiency and accuracy, avoids the subjective bias of manual labeling, and provides a reliable geometric benchmark for longitudinal calibration. To fully verify the effectiveness and robustness of the algorithm, this scheme selected four typical tunnel cross-sections for field measurements: a horseshoe-shaped test section, an irregular-shaped test section, a square tunnel test section, and a shield tunnel curve test section.

[0152] This method first performs longitudinal calibration: Through fully automated sleeper edge detection and recognition, the sleeper center position is extracted as the geometric reference, and global smoothing spacing constraint optimization is used to achieve high-precision correction of uniform scan line distribution and mileage accumulation error. Then, circumferential calibration is performed: First, a robust polygonal profile is generated based on the Tunnel Cross-Section Adaptive Fitting Method (TSAM). This method uses a dynamic threshold layer-by-layer refinement with feature priority retention combined with vertex smoothing optimization driven by area-shape consistency, adapting to various cross-sections such as circular, horseshoe, irregular, and square shapes. Then, a fixed-distance correction algorithm based on profile optimization is used to perform strict fixed-distance equal-division sampling on each side of the simplified main lining profile, and an adaptive grayscale weight mapping mechanism of local point density and curvature is introduced to achieve accurate correspondence between pixels and the actual circumferential physical distance. Finally, a distortion-free, high-resolution orthophoto grayscale image of the tunnel inner wall is generated, supporting centimeter-level defect measurement and area statistics.

[0153] Traditional method comparison experiment: To ensure the reliability of accuracy verification, the true ground values ​​for all lengths and areas were obtained using a high-precision Leica TS16 total station (range accuracy 1mm + 1.5ppm, angle accuracy 0.5″). The measurements were conducted using a combination of free station setup and backsight control points. Each target was measured at least three times independently (a total of 12 angle and distance readings). The final coordinates were calculated as a weighted average, and the standard deviation of the three-dimensional coordinates of a single point was controlled within 1.2mm.

[0154] True length value: For the length measurement of tunnel ancillary facilities, a multi-point tracking measurement mode along an approximate path on the lining surface is adopted. The length of each segment is obtained by summing the chord lengths between two adjacent points. The expanded uncertainty of a single segment length measurement is ±3mm.

[0155] True area value: For typical defective areas or ancillary facilities (such as electrical box panels, water leakage patches), the area is obtained by collecting coordinates from multiple points on site and calculating the area using the polygon area calculation formula. The expanded uncertainty of the area measurement is ±0.012m², mainly affected by boundary point positioning errors and local micro-curvature of the lining.

[0156] To clearly demonstrate the innovation and advantages of our proposed method in the field of tunnel orthophoto generation, Table 2 provides a systematic comparison with existing representative methods. This comparison covers core technologies, applicable cross-sectional shapes, qualitative advantages and limitations, quantitative accuracy indicators, and their relative merits and demerits compared to our proposed method.

[0157] Table 2 Comparison of Existing Algorithms

[0158] method Core technologies Applicable cross-sectional shapes Does it support arbitrary shapes? Does it support centimeter-level area measurement? The length RMSE (cm) in this dataset. The area MAE (m²) on the dataset of this scheme. Brief review of main advantages and limitations Sun et al. (2020) RANSAC circle fitting + arc length projection Only circular no Only roughly supports circular tunnels (not centimeter-level high precision). 6.8 0.020 (Valid only for circular segments; invalid for non-circular segments) Applicable to standard shield-tunnel circular tunnels; deformation accuracy is in the millimeter range, but area measurement is limited by projection distortion. Chen (2023) Least squares elliptic fitting + cylindrical projection Circular / Near-elliptical no Only a rough estimate is possible for circular tunnels (area accuracy is not reported in the literature). 5.4 0.018 (Valid only for circular segments; invalid for non-circular segments) Shield tunnel crack length extraction is relatively good; however, the non-circular cross-section of the mining method fails, resulting in limited area accuracy. Liu et al. (2020) Harmonic mapping parameterization Theoretically arbitrary yes No (No physical distance mapping, unreliable measurement is not possible) Reliable measurement is not supported. Area measurement is not supported. High geometric fidelity and low distortion parameterization; however, it lacks precise distance mapping and cannot quantify area / length. Ruan et al. (2021) Circumferential seam marking + orthographic projection Only circular no Partial support (circular tunnel) 4.9 0.0142 Longitudinal correction is relatively effective; limited to circular shield tunnels, with moderate area accuracy. This solution method Longitudinal + circumferential correction Any (circle / horseshoe / square / irregular) yes Yes (full-shape, centimeter-level high precision) 2.5 0.0096 Full-section shape adaptation; longitudinal and circumferential dual correction; centimeter-level length and area measurement.

[0159] By comparing typical algorithms such as circle fitting, ellipse fitting, harmonic mapping, and arc length expansion, it can be seen that existing methods are mostly limited to circular or near-circular cross sections, have insufficient longitudinal correction, and lack pixel-level physical distance mapping. In contrast, the longitudinal global smooth spacing constraint optimization, circumferential TSAM+ distance correction, and adaptive gray-scale weight mapping of this scheme achieve universal, high-precision geometric correction and reliable area measurement for any cross section, providing significant advantages for tunnel defect assessment.

[0160] like Figure 17 As shown, in the ablation experiment, to accurately evaluate the independent contribution of each core module proposed in this scheme to the geometric accuracy, measurement accuracy, and visual quality of orthophotos, this section conducted module ablation experiments on a unified dataset (the mixed section of the mining method section and the shield tunneling curve section of Chongqing Metro Line 16, n=20). All experiments were based on the complete algorithm framework of this scheme, with only specific modules removed or disabled. The results are shown in Table 3.

[0161] Table 3 Ablation Experiment Data Results

[0162] experimental group Length RMSE (cm) Area (MAE) (m²) A. No correction 11.4 0.1246 B. Vertical calibration only 4.7 0.092 C. Loop calibration only 5.8 0.035 D. Remove global smoothing spacing constraints 3.9 0.0185 E. Remove contour optimization 3.2 0.0132 F. Complete Algorithm of this Scheme 2.62 0.0096

[0163] The longitudinal calibration module is the core of longitudinal geometric stabilization. Enabling this module alone (Group B) can reduce the length RMSE from 11.4 cm to 4.7 cm (a reduction of approximately 59%), significantly eliminating uneven distribution of scan lines and mileage accumulation errors, and significantly improving subjective image continuity.

[0164] The circumferential calibration module makes a significant contribution to area accuracy and adaptability to arbitrary shapes. Enabling only the circumferential calibration (Group C) reduces the area MAE to 0.034 m² and the length RMSE to 5.6 cm, demonstrating that the module can effectively compensate for projection distortion of non-circular cross-sections and improve grayscale naturalness. However, the mileage error remains significant when longitudinal calibration is lacking.

[0165] Global smoothing spacing constraint optimization is crucial for precise and uniform control of scan line spacing. After removal (Group D vs. Group F), the length RMSE increased from 2.62 cm to 3.9 cm (an increase of approximately 49%), the area MAE slightly increased to 0.0182 m², and slight spacing fluctuation artifacts appeared in the longitudinal direction of the image, indicating that this constraint is a key guarantee for high longitudinal uniformity.

[0166] Contour optimization significantly improved contour accuracy and the stability of subsequent interval sampling. After removal (Group E vs. Group F), the area MAE increased from 0.0099 m² to 0.0131 m² (an increase of approximately 32%), and slight jaggedness and local geometric distortion appeared at the boundaries of complex cross-sections, resulting in a decrease in subjective quality.

[0167] The ablation results fully demonstrate that the longitudinal calibration, circumferential calibration, global smoothing spacing constraint, and contour optimization modules of this scheme work together in a progressive manner to achieve distortion-free high-precision reconstruction of orthophotos of tunnels of arbitrary shapes and centimeter-level defect measurement capabilities.

[0168] For visual comparison of images before and after calibration, such as Figure 11 As shown, the effectiveness of the calibration algorithm was verified through visual comparison, demonstrating the differences between the original and calibrated images of four cross-sections: the left sub-image shows significant distortion in the velocity-mileage direction and cross-sectional direction of the original image, resulting in disproportionate length ratios, blurred object edges, and poor identification of tunnel wall ancillary facilities, failing to meet the needs of defect identification or area measurement; the right sub-image presents the calibrated image, with uniform scan line distribution, sharp contours, and significantly enhanced identification, clearly reflecting the geometric details and surface features of the tunnel wall. The test sections of this scheme are mainly straight sections or gentle curve sections with small curvature, so the impact of curve offset on the experimental results is limited. However, subway tunnels generally have sections with small turning radii and large curvature. In such scenarios, relying solely on the relative measurement of laser point clouds can easily introduce attitude errors. To overcome this limitation, the TLSD motion detection device developed in this invention has reserved an IMU interface and installation space, supporting multimodal data fusion.

[0169] like Figure 12 As shown, this image compares the local details of the tunnel inner wall before and after calibration for four different cross-sectional shapes, highlighting the differences between the measured length and area values ​​and the actual values. Before calibration, the measured results showed significant deviations and blurred details due to uneven longitudinal mileage and circumferential projection distortion, making it impossible to accurately identify ancillary facilities or damaged areas. After calibration, the geometric consistency of the image was significantly improved, and the measured results closely matched the actual values, validating the algorithm's effectiveness.

[0170] For the measurement accuracy verification, the circumferential and longitudinal correction methods proposed in this scheme were analyzed and verified using measurement results from 20 sets of actual length data for tunnel ancillary facilities, including horseshoe-shaped, irregular-shaped, circular, and square shield tunnels. The results are as follows: Figure 13 As shown, key accuracy indicators for length data were calculated, including mean absolute error, root mean square error, and standard deviation of absolute error. Specific calculation results show that the mean absolute error of this method is 2.15 cm, the root mean square error is 2.5 cm, and the standard deviation of absolute error is 1.43 cm. These indicators demonstrate that the overall deviation of this method is controlled within the centimeter level, exhibiting low sensitivity to length variations and minimal dispersion in the measurement results, thus ensuring high stability and accuracy in the measurement of the length of auxiliary facilities.

[0171] Under the condition of a circular shield tunnel, the circumferential and longitudinal correction methods proposed in this scheme were compared and analyzed with the traditional RANSAC algorithm circle fitting method based on the measurement results of 20 sets of actual area data of auxiliary facilities in the tunnel. The results are as follows: Figure 14 As shown. These auxiliary facilities include typical components such as pipelines, electrical boxes, and signs. By calculating key accuracy indicators for all positive area data, including mean absolute error, root mean square error, and standard deviation of absolute error, the performance of the two methods can be objectively evaluated. Specifically, the calculation results show that the mean absolute error of the proposed method is 0.0083 m², the root mean square error is 0.010 m², and the standard deviation of absolute error is 0.0056 m²; the corresponding values ​​for the traditional random sampling consensus algorithm circle fitting method are a mean absolute error of 0.0185 m², a root mean square error of 0.0228 m², and a standard deviation of absolute error of 0.0133 m². Based on these indicators, the proposed method significantly outperforms the traditional method in all three dimensions: mean absolute error, root mean square error, and standard deviation of absolute error.

[0172] Meanwhile, to further verify the accuracy, this scheme was evaluated based on two different sets of tunnel ancillary facility area measurement data. The first set of data included 20 samples with known areas, covering horseshoe-shaped cross-section tunnels, irregular cross-section tunnels, shield-shaped cross-section tunnels, and square cross-section tunnels. The results before and after correction were compared with the true values ​​to quantitatively evaluate the calibration effect of the algorithm. Statistical analysis showed that the average absolute error of the uncorrected image was 0.1246 m², the maximum absolute error reached 0.3930 m², and the average percentage error was approximately 17.83%, which was large and unstable, failing to meet the requirements for accurate measurement. After correction, the average absolute error significantly decreased to 0.0096 m², the maximum absolute error was 0.0234 m², and the measured value was highly close to the true value, with an overall accuracy improvement of approximately 14 times. Figure 15 The bar chart shown highlights the significant improvement in accuracy after calibration, further confirming the algorithm's effectiveness in eliminating longitudinal and circumferential distortions. The second set of data, also based on 20 known area samples, performs statistical analysis on the measurement results, such as... Figure 16 As shown in the dot-line diagram, the area measurement difference of this algorithm is within 0.0252 m², the standard deviation of the error is 0.0110 m², and the average percentage error is only 1.25%, which is better than the traditional cylindrical projection method

[48] . These statistical indicators focus on the quantitative analysis of absolute accuracy, indicating that the algorithm has good stability under different cross-sectional shapes, performs well in all cross-sectional shapes, has a uniform error distribution and no systematic deviation, and the area measurement accuracy can reach the centimeter level.

[0173] To further quantify the statistical reliability of the measurement results, this scheme uses the t-distribution (39 degrees of freedom, 95% confidence level) to calculate the mean absolute error and its confidence interval based on all 40 area samples (the first group of 20 known area samples + the second group of 20 statistical analysis samples, covering horseshoe-shaped, irregular, shield circular and square cross-sections, with at least 10 independent samples for each cross-section type). The results are shown in Table 4.

[0174] Table 4. 95% confidence intervals for corrected area measurements (based on t-distribution, 39 degrees of freedom)

[0175] Section type Sample size Mean absolute error (m²) 95% confidence interval (m²) Relative error confidence interval (%) horseshoe shape 10 0.0096 0.0074–0.0118 0.88–1.55 Irregular / unusual 10 0.0099 0.0077–0.0121 0.92–1.60 square 10 0.0105 0.0082–0.0128 1.00–1.78 Circular (shield tunnel) 10 0.0084 0.0065–0.0103 0.78–1.46 overall 40 0.0096 0.0083–0.0109 0.94–1.54

[0176] As shown in Table 4, the mean absolute error confidence intervals for all cross-sectional types remain within a narrow range, and the overall relative error confidence intervals are controlled within [0.94%, 1.54%], indicating that the proposed method has high statistical consistency and robustness in area measurement under various tunnel cross-sectional shapes, with no obvious systematic bias. This result further confirms the effectiveness of longitudinal global smoothing spacing constraint optimization, circumferential TSAM adaptive contour fitting, and distance correction algorithms in eliminating geometric distortion and achieving reliable centimeter-level quantitative assessment.

[0177] Parameter Sensitivity Analysis: To further examine the robustness and parameter stability of the proposed algorithm, this section conducted controlled variable experiments on several key parameters. Experimental data were collected for various cross-sectional shapes, including circular shield tunnels, horseshoe-shaped tunnels, irregular tunnels, and square tunnels. A 25-meter tunnel segment was selected for each shape, covering multiple typical cross-sectional types to ensure representativeness of the results. All analysis results are averages of these different shaped segments to reflect the overall performance of the algorithm under arbitrary tunnel shapes. With other parameters fixed, only a single parameter was adjusted to observe its impact on the RMSE (Real-Time Sequence) of length and the MAE (Maximum Amount Equivalent) of area.

[0178] The physical resolution of the circumferential pixels was set to four values, d, of 3mm, 5mm, 8mm, and 10mm. The results show that the area MAE (Magnitude of Effect) increases monotonically with increasing d: 0.0082m² at 3mm, 0.0099m² at 5mm, 0.0114m² at 8mm, and 0.0131m² at 10mm. The main reason for the increase in area MAE is that the physical spacing of the circumferential pixels becomes coarser with increasing d, resulting in more errors when quantizing the disease boundary, especially noticeable at horseshoe-shaped or irregular cross-sections with significant curvature changes. At the same time, the computational load increases as d decreases. Considering both centimeter-level measurement accuracy and engineering efficiency, d=5mm is the optimal balance point. At this point, the actual physical distance represented by each circumferential pixel is basically consistent with the actual distance represented by the vertical pixel, thus minimizing horizontal and vertical image stretching and maximizing geometric fidelity, making it suitable for disease measurement and visual interpretation.

[0179] 20. Vertical Target Image Height Pixel Count: The original data used in the experiment consisted of 25-meter tunnel sections of various cross-sectional shapes. The total number of original scan sections varied slightly due to shape and acquisition density, but all were within the range of approximately 6000 to 7500 per section. Based on this, three settings were tested for vertical target image height: 4000, 5000, and 6000 pixels. The results showed that the length RMSE variation was less than 0.4 cm, with a maximum fluctuation of 0.38 cm, and the area MAE variation was less than 0.001 m², indicating that the vertical calibration module has strong robustness to vertical resolution and is insensitive to image height. The vertical calibration module achieves uniform distribution of scan lines and correction of mileage errors through sleeper or reference point benchmarks, automated edge detection, global smoothing spacing constraint optimization, and cubic spline interpolation, maintaining stability even when the pixel count varies within a certain range.

[0180] The appropriate value for the vertical target image height essentially depends on the length of the original data and the number of scan sections. The longer the original data, the longer the tunnel section being acquired, and the more original scan sections there are, the greater the number of vertical pixels that can be supported. This is because more original scan lines provide richer geometric information, supporting higher-resolution uniform resampling. If the original data is short, the number of pixels should be reduced accordingly to avoid over-interpolation. Therefore, it is not recommended that the number of vertical pixels exceed the number of original scan sections, i.e., the total number of original scan lines. If the manually set number of vertical pixels is significantly greater than the number of original scan sections, it constitutes oversampling, and the algorithm cannot create additional realistic geometric details from the limited original data, potentially leading to interpolation artifacts or loss of local details, thus affecting measurement accuracy.

[0181] To avoid inconsistent stretching and distortion in the image, the physical resolution 'd' of the horizontal circumferential pixels and the actual physical distance represented by each vertical pixel should be kept as equal as possible; that is, the actual distances in centimeters corresponding to the pixel spacing should be nearly identical. If the horizontal and vertical pixel settings result in a significant difference in the actual distances represented by these two values, it will cause overall image stretching or compression, affecting visual continuity and the geometric accuracy of lesion measurements. It is recommended that when selecting parameters, the mileage length of the original data segment and the total number of scan sections be calculated first, and then the horizontal and vertical pixel settings be coordinated accordingly to ensure that the actual physical distance represented by each pixel is as balanced as possible.

[0182] This solution addresses key challenges in generating orthophotos of tunnels of arbitrary shapes from mobile laser scanning point clouds, including non-uniform scanning lines, projection distortion, and poor adaptability to complex cross-sections. It proposes a complete high-precision reconstruction and calibration algorithm framework. Through the organic synergy of data preprocessing, longitudinal global calibration, and circumferential adaptive calibration, this framework achieves distortion-free, geometrically consistent orthophoto generation and supports reliable measurement of lesion length and area at the centimeter level.

[0183] This invention proposes a unified longitudinal calibration method applicable to both sleeper-equipped and sleeperless tunnels. Using automatic sleeper edge detection or high-precision manual reference points as the geometric benchmark, it integrates global smoothing spacing constraint optimization, natural cubic spline interpolation, and hard constraint mechanisms to effectively eliminate mileage accumulation errors and non-uniform scan line distribution, achieving high-precision uniform mapping of longitudinal pixel physical distances. Furthermore, it proposes a complete circumferential calibration algorithm and innovatively develops a Tunnel Cross-Section Adaptive Fitting Method (TSAM) and a contour optimization-based distance correction algorithm. TSAM robustly extracts high-quality polygonal contours suitable for various cross-sections, including circular, horseshoe, irregular, and square shapes, through feature-priority dynamic threshold layer-by-layer refinement and area-shape consistency-driven vertex smoothing optimization. The distance correction algorithm further achieves strict equidistant sampling on each side and introduces density-curvature adaptive grayscale weight mapping to effectively compensate for intensity distortion caused by incident angle and local geometric changes, ultimately constructing a distortion-free orthophoto grayscale image where pixels accurately correspond to the true circumferential physical distance.

[0184] Furthermore, field tests on a mixed mining and shield tunneling section of Chongqing Metro Line 16 and a typical shield tunneling section of Shenzhen Metro Line 17 demonstrate that the proposed method exhibits excellent adaptability and stability for various complex cross-sections. After calibration, image geometric distortion is significantly eliminated, the RMSE for length measurement is reduced to 2.62 cm, the MAE for area measurement is reduced to 0.0099 m², and the average percentage error is controlled within 1.3%. Compared with traditional circle fitting, ellipse fitting, and harmonic mapping methods, this method achieves significant improvements in adaptability to arbitrary shapes, longitudinal stability, and quantification accuracy, meeting the centimeter-level accuracy requirements for intelligent identification of tunnel defects, statistical analysis of ancillary facilities, and structural health assessment.

[0185] The proposed algorithm significantly improves the efficiency of integrating and utilizing the appearance and geometric information of the tunnel inner wall, providing reliable technical support for accurate evaluation of tunnel service performance, maintenance decision-making, and digital operation and maintenance.

[0186] To achieve the above objectives, the present invention also provides a tunnel orthophoto generation and processing system based on global constraint optimization and contour refinement distance determination, such as... Figure 19 As shown, the system is used to implement the tunnel orthophoto generation and processing method based on global constraint optimization and contour refinement distance determination. The system specifically includes:

[0187] A data generation and acquisition unit is used to generate and acquire first data corresponding to the tunnel; wherein, the first data is the original point cloud data inside the tunnel, and the original point cloud data includes spatial coordinate information data and reflection intensity information data;

[0188] The first data processing and generation unit is used to longitudinally calibrate the first data based on a longitudinal geometric benchmark with known physical spacing and in combination with a global constraint optimization method, and generate second data corresponding to the first data; wherein, the second data is a point cloud cross-section sequence with uniformly distributed scan lines and accurate mileage in the longitudinal direction of the tunnel.

[0189] The second data processing and generation unit is used for contour refinement and distance determination processing of each section in the second data, and to generate third data corresponding to the second data; wherein, the third data is tunnel orthophoto data without geometric distortion; the contour refinement and distance determination processing includes extracting the main lining contour polygon of each section, and based on the main lining contour polygon, establishing a mapping relationship between pixels and circumferential real physical distance through a distance equal division method.

[0190] The longitudinal geometric reference includes the sleeper centerline automatically detected from the initial grayscale image generated from the original point cloud using image processing technology, or artificial markers with known arc length spacing laid out by external measurement means.

[0191] The first data processing and generation unit further includes: a first construction module, configured to construct a continuous first function corresponding to the scan line number as the independent variable and the corrected mileage as the dependent variable; wherein the first function is a mapping function; wherein the first function is a natural cubic smooth spline function; a second construction module, configured to construct a corresponding second function using the known physical spacing of the longitudinal geometric reference as a hard constraint condition and introducing a smoothing regularization term; wherein the second function is an objective function; the smoothing regularization term is an integral penalty term on the second derivative of the spline function; and a first generation module, configured to calculate and generate the corresponding first function based on the second function, and map the pseudo mileage of each original scan line to the corrected real physical mileage.

[0192] The second data processing and generation unit further includes: a third construction module, used to sort and process the cross-sectional point cloud and construct the corresponding initial polygonal contour; a first processing module, used to combine a feature-priority retention dynamic threshold refinement strategy to insert new vertices layer by layer on the basis of the initial polygonal contour; and a second processing module, used to optimize the vertex position of the polygonal contour after the vertices are inserted by minimizing an energy function that includes an area fidelity term and a side length smoothing term, and generate the corresponding main lining contour polygon.

[0193] The dynamic threshold refinement strategy that prioritizes feature retention includes prioritizing points that are most far from the current edge, have a large curvature change, or have a significant azimuth change when determining whether to insert a new vertex.

[0194] And / or, the second data processing generation unit further includes: a third processing module, used to identify and delete the contour segments corresponding to the local protruding auxiliary structures on the main lining contour polygon, and create a corresponding simplified main lining contour; a second generation module, used to perform fixed-distance equal-division sampling on each side of the simplified main lining contour according to a preset target circumferential resolution, and generate a series of corresponding equal-division points; a numerical assignment module, used to assign corresponding grayscale values ​​to the equal-division points based on the correspondence between the equal-division points and the original point cloud points, combined with adaptive weights; wherein, the adaptive weights are dynamically adjusted based on the local point density and / or curvature of the original point cloud points.

[0195] And / or, the third processing module further includes: an identification and removal module, used to identify and remove a continuous sequence of vertices corresponding to the auxiliary structure based on the radial distance abrupt change of the polygon vertex relative to the centroid of the cross section and / or the curvature abrupt change at the vertex.

[0196] In the system embodiment of the present invention, the specific details of the method steps involved in the generation and processing of tunnel orthophotos based on global constraint optimization and contour refinement distance have been described above. That is to say, the functional modules in the system are used to implement the steps or sub-steps in the above method embodiment, which will not be repeated here.

[0197] To achieve the above objectives, the present invention also provides a tunnel orthophoto generation and processing platform based on global constraint optimization and contour refinement ranging, such as... Figure 20 As shown, the system includes a processor, a memory, and a control program for a tunnel orthophoto generation and processing platform based on global constraint optimization and contour refinement ranging. The processor executes the control program, which is stored in the memory. This control program implements the steps of the tunnel orthophoto generation and processing method based on global constraint optimization and contour refinement ranging. The specific details of these steps have been described above and will not be repeated here.

[0198] In this embodiment of the invention, the tunnel orthophoto generation and processing platform based on global constraint optimization and contour refinement distance has a built-in processor, which can be composed of integrated circuits. For example, it can be composed of a single packaged integrated circuit, or multiple integrated circuits with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor connects to various components using various interfaces and lines, and executes programs or units stored in the memory, as well as calling data stored in the memory, to perform various functions and process data for tunnel orthophoto generation and processing based on global constraint optimization and contour refinement distance. The memory is used to store program code and various data, is installed in the tunnel orthophoto generation and processing platform based on global constraint optimization and contour refinement distance, and achieves high-speed and automatic access to programs or data during operation. The memory includes read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0199] To achieve the above objectives, the present invention also provides a computer-readable storage medium, such as... Figure 21As shown, the computer-readable storage medium stores a control program for a tunnel orthophoto generation and processing platform based on global constraint optimization and contour refinement distance determination. This control program implements the steps of the tunnel orthophoto generation and processing method based on global constraint optimization and contour refinement distance determination. In other words, a computer-readable storage medium is provided, storing a computer program that, when executed by a processor, implements the steps of the tunnel orthophoto generation and processing method based on global constraint optimization and contour refinement distance determination.

[0200] Specifically, the storage medium is a non-transitory computer-readable medium, which may be solid-state flash memory, SD card, or hard disk, and has a shock-resistant design to adapt to the tunnel vibration environment. The storage medium stores an executable instruction set, which includes: image data preprocessing code, laser intensity correction code, longitudinal and circumferential calibration code, and equal division processing code. When the instruction set is loaded and executed by the processor, steps S001-S002 can be implemented sequentially, and configuration parameters can be adjusted via INI file. The storage medium is compatible with embedded operating systems and has an error handling mechanism to ensure reliability, facilitating engineering replication and maintenance.

[0201] In the description of embodiments of the present invention, it should be noted that any process or method description in the flowcharts or otherwise described herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of the preferred embodiments of the present invention includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order according to the functions involved, as should be understood by those skilled in the art to which the embodiments of the present invention pertain.

[0202] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processing module, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, a “computer-readable medium” can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, the computer-readable medium can even be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0203] This invention generates and acquires first data corresponding to a tunnel through a method; wherein the first data is raw point cloud data inside the tunnel, the raw point cloud data including spatial coordinate information data and reflection intensity information data; based on a longitudinal geometric reference with known physical spacing, and combined with a global constraint optimization method, the first data is longitudinally calibrated and processed to generate second data corresponding to the first data; wherein the second data is a sequence of point cloud cross-sections with uniformly distributed scan lines and accurate mileage along the longitudinal direction of the tunnel; each cross-section in the second data is refined and distance-fixed, and a third data corresponding to the second data is generated; wherein the third data is orthophoto data of the tunnel without geometric distortion; the contour refinement and distance-fixing processing includes... This method includes extracting the main lining contour polygon of each cross section, and establishing a mapping relationship between pixels and the actual physical distance in the circumferential direction based on the main lining contour polygon through a fixed-distance equal division method. It also includes the corresponding system, platform, and storage medium. This method can overcome the limitations of local correction in dynamic filtering, achieve global optimal compensation for longitudinal mileage errors, and improve the geometric consistency of long-distance tunnel images. Furthermore, it accurately extracts the main lining contour in complex cross sections, effectively eliminating interference from auxiliary structures such as pipelines and cable troughs, ensuring that the fixed-distance equal division is based on the actual tunnel wall. Based on the fixed-distance equal division, it achieves adaptive grayscale equalization, compensating for intensity unevenness caused by changes in laser incident angle, point density, and curvature, thereby improving the visual quality and detail recognition of the image.

[0204] In other words, based on the adaptive fitting and fixed-distance equal division framework, this invention replaces traditional dynamic filtering with a global constraint optimization method in the longitudinal direction. It uses known physical distances as hard constraints and combines them with a smoothing regularization term to achieve globally optimal compensation for nonlinear velocity fluctuations, improving the longitudinal geometric consistency of long-distance tunnel images. In the circumferential tunnel cross-section adaptive fitting method (TSAM), feature-priority retention dynamic threshold refinement and area-shape consistency vertex optimization, combined with automatic identification and removal technology for auxiliary structures, effectively solve the problem of interference from pipelines, cable troughs, etc., on the main lining contour under complex cross-sections, improving the robustness and accuracy of contour extraction. Simultaneously, it introduces a density-curvature adaptive grayscale weight mapping mechanism, dynamically adjusting weights based on local point density and curvature to effectively compensate for intensity unevenness caused by laser incident angle and local geometric changes, resulting in natural grayscale transitions and clear details in the image. Compared with existing technologies, the distortion-free orthophotos generated by this invention achieve comprehensive optimization in three dimensions: geometric accuracy, contour fidelity, and visual quality, providing a high-fidelity data foundation for intelligent tunnel defect identification, auxiliary facility statistics, and digital operation and maintenance.

[0205] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for generating and processing tunnel orthophotos based on global constraint optimization and contour refinement, characterized in that, The method includes: Generate and acquire first data corresponding to the tunnel; wherein, the first data is the original point cloud data inside the tunnel, and the original point cloud data includes spatial coordinate information data and reflection intensity information data; Based on a longitudinal geometric datum with known physical spacing, and combined with a global constraint optimization method, the first data is longitudinally calibrated and the second data corresponding to the first data is generated; wherein, the second data is a point cloud cross-section sequence with uniformly distributed scan lines and accurate mileage in the longitudinal direction of the tunnel. The contour refinement and distance determination process is performed on each section in the second data to generate third data corresponding to the second data; wherein, the third data is tunnel orthophoto data without geometric distortion; the contour refinement and distance determination process includes extracting the main lining contour polygon of each section, and based on the main lining contour polygon, establishing a mapping relationship between pixels and circumferential real physical distance through a distance equal division method.

2. The method for generating and processing tunnel orthophotos based on global constraint optimization and contour refinement distance determination according to claim 1, characterized in that, The longitudinal geometric reference includes the sleeper centerline automatically detected from the initial grayscale image generated from the original point cloud using image processing technology, or artificial markers with known arc length spacing laid out by external measurement means.

3. A method for generating and processing tunnel orthophotos based on global constraint optimization and contour refinement distance determination according to claim 1 or 2, characterized in that, The process of longitudinally calibrating the first data based on a longitudinal geometric datum with known physical spacing, and combining it with a global constraint optimization method, and generating second data corresponding to the first data, further includes: Construct a continuous first function corresponding to the scan line number as the independent variable and the corrected mileage as the dependent variable; wherein, the first function is a mapping function; wherein, the first function is a natural cubic smooth spline function; Using the known physical spacing of the longitudinal geometric datum as a hard constraint, and introducing a smoothing regularization term, a corresponding second function is constructed; wherein, the second function is the objective function; and the smoothing regularization term is an integral penalty term on the second derivative of the spline function; Based on the second function, the corresponding first function is calculated and generated, and the pseudo mileage of each original scan line is mapped to the corrected real physical mileage.

4. The method for generating and processing tunnel orthophotos based on global constraint optimization and contour refinement distance determination according to claim 1, characterized in that, The contour refinement and grading process for each section in the second data, and the generation of third data corresponding to the second data, further includes: The cross-sectional point cloud is sorted and processed, and the corresponding initial polygonal contours are constructed. Combining a dynamic threshold refinement strategy that prioritizes feature preservation, new vertices are inserted layer by layer based on the initial polygonal contour. By minimizing an energy function that includes area fidelity terms and side length smoothing terms, the vertex positions of the polygonal contour after vertices are inserted are optimized, and the corresponding main lining contour polygon is generated.

5. The method for generating and processing tunnel orthophotos based on global constraint optimization and contour refinement distance determination according to claim 4, characterized in that, The dynamic threshold refinement strategy that prioritizes feature retention includes prioritizing points that are most far from the current edge, have a large curvature change, or have a significant azimuth change when determining whether to insert a new vertex.

6. A method for generating and processing tunnel orthophotos based on global constraint optimization and contour refinement distance determination according to claim 1 or 4, characterized in that, The contour refinement and grading process for each section in the second data, and the generation of third data corresponding to the second data, further includes: Identify and delete the contour segments corresponding to the local protruding auxiliary structures on the main lining contour polygon, and create a corresponding simplified main lining contour. Based on the preset target circumferential resolution, each edge of the simplified main lining contour is sampled at fixed intervals and a series of corresponding division points are generated. Based on the correspondence between the equally divided points and the original point cloud points, and combined with adaptive weights, the equally divided points are assigned corresponding gray values; wherein, the adaptive weights are dynamically adjusted based on the local point density and / or curvature of the original point cloud points.

7. The method for generating and processing tunnel orthophotos based on global constraint optimization and contour refinement distance determination according to claim 6, characterized in that, The process of identifying and deleting contour segments corresponding to local protruding auxiliary structures on the main lining contour polygon, and creating corresponding simplified main lining contours, further includes: Based on the radial distance abrupt change of polygon vertices relative to the centroid of the cross section and / or the curvature abrupt change at the vertices, the continuous vertex sequence corresponding to the attached structure is identified and removed.

8. A tunnel orthophoto generation and processing system based on global constraint optimization and contour refinement, characterized in that, The system is used to implement the tunnel orthophoto generation and processing method based on global constraint optimization and contour refinement distance as described in any one of claims 1 to 7, the system comprising: A data generation and acquisition unit is used to generate and acquire first data corresponding to the tunnel; wherein, the first data is the original point cloud data inside the tunnel, and the original point cloud data includes spatial coordinate information data and reflection intensity information data; The first data processing and generation unit is used to longitudinally calibrate the first data based on a longitudinal geometric benchmark with known physical spacing and in combination with a global constraint optimization method, and generate second data corresponding to the first data; wherein, the second data is a point cloud cross-section sequence with uniformly distributed scan lines and accurate mileage in the longitudinal direction of the tunnel. The second data processing and generation unit is used for contour refinement and distance determination processing of each section in the second data, and to generate third data corresponding to the second data; wherein, the third data is tunnel orthophoto data without geometric distortion; the contour refinement and distance determination processing includes extracting the main lining contour polygon of each section, and based on the main lining contour polygon, establishing a mapping relationship between pixels and circumferential real physical distance through a distance equal division method.

9. A tunnel orthophoto generation and processing platform based on global constraint optimization and contour refinement, characterized in that, The system includes a processor, a memory, and a control program for a tunnel orthophoto generation and processing platform based on global constraint optimization and contour refinement distance determination. The processor executes the control program, which is stored in the memory. This control program implements the tunnel orthophoto generation and processing method based on global constraint optimization and contour refinement distance determination as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a control program for a tunnel orthophoto generation and processing platform based on global constraint optimization and contour refinement distance determination. The control program for the tunnel orthophoto generation and processing platform based on global constraint optimization and contour refinement distance determination implements the tunnel orthophoto generation and processing method based on global constraint optimization and contour refinement distance determination as described in any one of claims 1 to 7.

Images