Automatic monitoring system and method for tunnel surrounding rock deformation measurement

Abstract

The invention discloses an automatic monitoring system and method for tunnel surrounding rock deformation measurement, and belongs to the technical field of tunnel construction safety monitoring. The monitoring method comprises the following steps: projecting an infrared structured light pattern to a monitoring section, fusing natural texture information of a surrounding rock surface to construct a virtual target, and obtaining a reference image; arranging a reference terminal, an optical reference point and a reference terminal in a stable area of the tunnel to establish a dynamic space measurement reference; the method comprises the following steps: acquiring a real-time measurement image of a monitored section by using a monitoring terminal, preprocessing the real-time measurement image, and calculating sub-pixel image surface displacement by adopting a digital image correlation algorithm fused with natural texture analysis; and performing dynamic error compensation according to the pose variation fed back by the reference terminal, and finally outputting actual displacement. According to the invention, the whole section of the tunnel surrounding rock monitoring section can be monitored without a solid target, the sub-millimeter level measurement precision is realized, and the method has the advantages of high environmental interference resistance and capability of realizing all-weather automatic monitoring.

Description

Technical Field

[0001] This application relates to an automated monitoring system and method for measuring deformation of surrounding rock in tunnels, belonging to the field of tunnel surrounding rock deformation monitoring technology. Background Technology

[0002] During tunnel construction, high-precision, real-time monitoring of surrounding rock deformation during the initial support stage is crucial for ensuring construction safety. Current mainstream automated monitoring technologies include total station automatic monitoring, 3D laser scanning, and traditional binocular vision measurement. Total station automatic monitoring uses a limited number of reflecting prisms on the surrounding rock surface as targets, automatically aiming at and measuring changes in target coordinates to calculate deformation. However, this method suffers from discrete feature points, leading to blind spots and failing to reflect the continuous deformation of the entire cross-section. 3D laser scanning acquires high-density point cloud data from the tunnel surface for deformation analysis. While it enables surface measurement, the laser signal attenuates significantly in dusty and humid environments, making long-term stable monitoring accuracy difficult. Traditional binocular vision measurement uses two cameras to simulate human eyes, utilizing parallax principles for 3D reconstruction and deformation measurement. However, this method struggles with feature point matching when there is uneven lighting or missing textures within the tunnel, resulting in insufficient measurement stability and reliability.

[0003] The above technologies all rely on physical targets or specific environmental conditions, and are greatly affected by ambient light and dust, resulting in measurement continuity and accuracy that cannot meet the needs of intelligent tunnel construction. Summary of the Invention

[0004] To address the problems existing in the prior art, this application provides an automated monitoring system and method for measuring tunnel surrounding rock deformation that does not rely on physical targets, has strong resistance to environmental interference, and can achieve high-precision continuous monitoring of the entire cross section.

[0005] According to one aspect of this application, an automated monitoring method for tunnel surrounding rock deformation is provided, the monitoring method comprising: The monitoring method includes: S1, a preset infrared structured light pattern is projected onto the monitoring section of the surrounding rock in the area to be monitored to form an infrared structured light band covering the monitoring section. Based on the initial image of the monitoring section collected by the monitoring terminal in the initial stable state, including the infrared structured light band information and the natural texture information of the surrounding rock surface of the monitoring section, the infrared structured light band information and the natural texture information are fused based on the initial image to construct a virtual target of the monitoring section and store it as a reference image. S2, at least one fixed optical reference point and a reference terminal for observing the optical reference point are set up in the stable area of ​​the completed tunnel. Infrared structured light is emitted from the optical reference point to the cross section of the stable area to form an optical reference. A unified measurement coordinate system is established based on the reference terminal. The reference terminal and the monitoring terminal are synchronously collected to obtain the real-time reference image of the optical reference and the real-time measurement image of the virtual target, respectively. S3, preprocess the real-time measurement image acquired in S2 to obtain a high-resolution infrared structured light feature line image; S4. The feature line image obtained in S3 is matched with the reference image. A digital image correlation algorithm based on natural texture analysis is used to obtain the sub-pixel image plane displacement of each feature point in the real-time measurement image relative to the reference image. S5: Based on the position changes of each feature point in the real-time reference image, calculate the pose change of the monitoring terminal itself, and perform geometric correction on the sub-pixel image plane displacement in S4 based on the pose change. Combined with the calibration parameters and measurement distance of the monitoring terminal, convert the corrected sub-pixel image plane displacement into actual displacement and output it. S6 performs time-series analysis on the actual displacement output by S5 to obtain the deformation development trend of the monitored section, and makes judgments and issues warnings based on preset warning rules.

[0006] Preferably, in S1, the infrared structured light pattern is a grid pattern, and the infrared structured light pattern covers the arch, shoulder and sidewall areas of the monitoring section.

[0007] The laser linewidth of the infrared structured light strip is ≤1.5nm, and the linewidth of a single laser on the projection surface is 0.8mm~1.2mm; The diameter of the light spot at the intersection of the infrared structured light bands is ≤2mm.

[0008] Preferably, in S3, the preprocessing of the real-time measurement image includes: image processing of the real-time measurement image based on a deep neural network to segment the light band region of the infrared structured light band from the complex background of the real-time measurement image, and super-resolution reconstruction of the segmented light band region to obtain a high-resolution infrared structured light feature line image.

[0009] Preferably, the aforementioned deep neural network is a PSPNet network structure, wherein the image segmentation and super-resolution reconstruction process is as follows: First, the real-time measurement image is input into the PSPNet network structure, and the feature map is obtained after convolutional downsampling; Then, the feature map is input into the pyramid pooling module, and its output is added to the original feature map; Finally, the segmentation results are upsampled through convolutional layers and bilinear interpolation, and optimized using the Dice Loss function to output a binarized infrared structured light feature line image.

[0010] Preferably, in S4, the calculation process for the sub-pixel image plane displacement is as follows: S401, the feature line image is matched with the reference image, the deformation of the feature point image sub-region is described by the shape function, and the initial value of the integer pixel displacement of the feature point is obtained by the integer pixel search algorithm. S402, Establish a quality assessment model for natural image texture based on feature points to evaluate the similarity of image sub-regions and image quality; S403, based on the Newton-Raphson iterative method, the initial value of the actual displacement of the integer pixel is optimized to obtain the sub-pixel image plane displacement with sub-pixel accuracy.

[0011] Preferably, in the quality evaluation model in S402, the similarity of feature points before and after the deformation of the natural texture image is evaluated using the zero-mean normalized least square distance correlation function C. ZNSSD ; The quality of natural texture images with feature points is evaluated by using the average gray-level gradient of image sub-regions.

[0012] Preferably, in S5, the pose change is the data after outliers are identified and removed using the 3σ criterion; The specific process of converting the corrected sub-pixel image plane displacement into actual displacement includes: First, calculate the scaling factor between the subpixel surface displacement and the actual displacement. K ; Finally, using the scaling factor K The subpixel image plane displacement is converted into actual displacement, which includes the arch settlement displacement and the peripheral convergence displacement.

[0013] According to another aspect of this application, an automated monitoring system for tunnel surrounding rock deformation is provided, the monitoring system being used in the aforementioned automated monitoring method for measuring tunnel surrounding rock deformation, the monitoring system comprising: The virtual target and optical reference construction unit includes at least one virtual target construction unit that projects an infrared structured light pattern onto the tunnel monitoring section to form a virtual target, and an optical reference point that projects infrared structured light onto the section of a completed stable area of ​​the tunnel to form an optical reference. The virtual target construction unit is installed on the sidewall of the area to be monitored, and the optical reference point is installed on the sidewall of the completed stable area. The virtual target construction unit and the optical reference point each integrate a coaxial visible light source for auxiliary aiming during equipment installation and debugging. The acquisition unit includes at least one monitoring terminal and at least one reference terminal. The monitoring terminal and the reference terminal are respectively equipped with filters that only transmit infrared structured light bands. The monitoring terminal is installed in the area to be monitored in the tunnel and is used to acquire real-time measurement images of the virtual target on the monitoring section. The reference terminal is installed in the completed and stable area of ​​the tunnel and is used to acquire real-time reference images of the optical reference point. A data processing and control unit is connected to the monitoring terminal, the reference terminal, the target construction unit, and the optical reference point, respectively. The data processing and control unit is configured as follows: The virtual target construction unit and the optical reference point are controlled to form a virtual target in the monitoring area and an optical reference in the stable area, respectively. The real-time measurement image is preprocessed, and a digital image correlation algorithm based on natural texture analysis is executed to calculate the sub-pixel image plane displacement of each feature point in the real-time measurement image relative to the reference image. The sub-pixel image plane displacement is geometrically corrected, and the corrected sub-pixel image plane displacement is converted into actual displacement and output. The analysis and early warning module is connected to the data processing and control unit. The analysis and early warning module obtains the deformation development trend of the monitoring section based on the time sequence analysis of the converted actual displacement, and makes judgments and issues early warnings based on preset early warning rules.

[0014] The beneficial effects that this application can produce include: This application emits infrared structured light bands to the monitoring section of the tunnel monitoring area. The virtual target, constructed by fusing the infrared structured light band information with the natural texture information of the surrounding rock surface, completely eliminates the need for physical targets. This provides stable and identifiable optical features for subsequent monitoring. The virtual target covers the entire monitoring section, eliminating blind spots in discrete point monitoring. Combined with narrow-bandpass filters configured at the monitoring and reference terminals, only valid infrared signals are received, ensuring high quality and stability of the input image from the source. This guarantees high quality and stability of the input image signal in complex tunnel environments, laying a foundation for all subsequent high-precision processing. By employing image super-resolution segmentation technology based on the PSPNet network, real-time measurement image details are enhanced. Combined with a digital image correlation algorithm based on natural texture analysis, displacement measurement accuracy is improved to the sub-pixel level, ultimately achieving sub-millimeter-level actual displacement measurement accuracy.

[0015] By deploying reference terminals in stable areas and fixed optical reference points, a dynamic spatial reference is provided for the entire monitoring process. At the same time, by calculating and dynamically compensating for the pose drift of the monitoring terminal caused by factors such as vibration and temperature changes in real time, the attitude drift of the monitoring terminal is eliminated. This solves the core problem of reference inaccuracy in visual measurement systems in dynamic environments and ensures the accuracy and reliability of long-term monitoring data. Attached Figure Description

[0016] Figure 1 In this embodiment of the invention, the optical reference planes are located in the monitoring area of ​​the tunnel and the completed tunnel, respectively. A schematic diagram illustrating the principle of the virtual target and optical reference formed by the stable region; Figure 2 The automated monitoring system for measuring deformation of surrounding rock in tunnels, as described in this embodiment of the invention, is used within the tunnel. A schematic diagram of the site layout; Figure 3 In this embodiment of the invention, the Newton-Raphson method is used for subpixel displacement measurement. A schematic diagram; Figure 4 The present invention employs the 3σ criterion to remove outliers from pose changes in this embodiment. Fruit image; Figure 5 The settlement observation curves of the monitoring system and total station described in this application in the embodiment of the present invention are shown in section DK37+820-A. Figure 6 This is the peripheral convergence observation curve of the monitoring system and total station described in this application in the DK37+820-S1 segment in an embodiment of the present invention; Figure 7 The settlement observation curves of the monitoring system and total station described in this application in the embodiment of the present invention are shown in section DK37+825-A. Figure 8 This is the peripheral convergence observation curve of the monitoring system and total station described in this application in the DK37+825-S1 segment in an embodiment of the present invention; Figure 9 The settlement observation curves of the monitoring system and total station described in this application in the embodiment of the present invention are shown below in section DK37+841-A. Figure 10 This is the peripheral convergence observation curve of the monitoring system and total station described in this application in the DK37+841-S1 segment in this embodiment of the invention; List of components and reference numerals: 1. Reference terminal; 2. Monitoring terminal; 3. Virtual target; 4. Virtual target construction unit; 5. Optical reference point; 6. Optical reference. Detailed Implementation

[0017] According to one embodiment of this application, an automated monitoring method for measuring deformation of surrounding rock in tunnels is provided, the monitoring method comprising: S1. Infrared structure transmitters and monitoring terminals are deployed in the tunnel monitoring area. The infrared structure transmitters project a preset infrared structured light pattern onto the monitoring section of the surrounding rock in the monitoring area to form an infrared structured light band covering the monitoring section. Based on the initial image of the monitoring section, which includes the infrared structured light band and the natural texture information of the surrounding rock surface in the initial stable state, collected by the monitoring terminals deployed in the tunnel monitoring area, a virtual target of the monitoring section is constructed based on the initial image and stored as a reference image. The target used in this application is a virtual target. Compared to physical reflective targets or prism targets installed on the surrounding rock surface, this avoids monitoring failures caused by target damage, obstruction, or installation difficulties. Furthermore, the infrared structured light band of the virtual target can cover dense feature point data across the entire tunnel monitoring section, thus eliminating blind spots in discrete point monitoring. The virtual target integrates infrared structured light band information with the natural texture information of the surrounding rock surface of the monitoring section, enabling it to obtain relatively clear image signals even in harsh construction environments. This achieves not only full-section measurement but also high-quality real-time measurement images. In this application, the infrared structured light is a near-infrared invisible laser of a specific wavelength. This characteristic avoids interference from strong construction light and has better penetration ability through tunnel dust than visible light, thus ensuring the continuity of monitoring.

[0018] In this application, the wavelength of the infrared structured light is any one of 850nm, 910nm, and 915nm. This is because infrared structured light in this band can be actively emitted without being affected by environmental factors such as visible light changes, tunnel dust, and smoke, and can establish optical feature lines of the tunnel monitoring section. Furthermore, infrared structured light in this band can also project structured light patterns, and the feedback signal from the natural texture of the tunnel inner wall surface is sufficient to form clear and identifiable optical features. In one embodiment of this application, infrared structured light in the 915±5nm band is preferred, as it conforms to the stable transmission characteristics of the near-infrared band, and the wavelength tolerance is controlled within ±5nm, which can effectively avoid distortion of the structured light pattern caused by wavelength drift.

[0019] In this application, the laser linewidth of the infrared structured light strip is controlled to be ≤1.5nm, with a single strip linewidth of 0.8mm~1.2mm, ensuring clear identification at a projection distance of 50m, balancing accuracy and cost, and avoiding the spectral broadening effect caused by tunnel vibration. The focal spot of the infrared structured light is ≤2mm, and the beam divergence angle is ≤4mrad, achieving sub-millimeter-level monitoring accuracy at a distance of 50m and maintaining pattern distortion-free operation within 100m, making it suitable for long tunnel monitoring. Furthermore, the single strip length of the infrared structured light is 3m~5m, which can match common tunnel cross-sectional dimensions (3m wide sidewall, 5m high arch) to cover key areas from the arch shoulder to the sidewall; the projection distance is 10m~100m, meeting the needs of conventional 30m~60m monitoring cross-section spacing and extra-long tunnels. Thus, by precisely controlling the laser linewidth, spot size, and divergence angle of the infrared structured light, high-precision, long-distance structured light pattern projection is achieved in complex tunnel environments, providing a reliable optical feature basis for surrounding rock deformation monitoring.

[0020] In this application, the infrared structured light pattern is a grid-like pattern formed by the intersection of horizontal and vertical structured light, so as to respectively achieve horizontal coverage of the tunnel horizontal monitoring section for convergence monitoring, and vertical coverage of the vertical monitoring section for settlement monitoring; thereby ensuring that traditional feature points such as the arch crown, arch shoulder, and sidewalls are all within the monitoring range.

[0021] In this application, the monitoring terminal and the reference terminal are monocular machines equipped with infrared band filters, that is, specific infrared band filters are added to the monitoring terminal and the reference terminal so that the monitoring terminal and the reference terminal only observe the optical feature lines of the monitoring section.

[0022] The natural texture information of the surrounding rock surface includes: structural information, natural texture information, and deformation information of the surrounding rock cross-section surface.

[0023] This application is based on the measurement principle of monocular machine vision structured light. By installing virtual target construction units on the sidewall of the initial support area, a preset infrared structured light pattern is actively projected onto each monitoring section of the surrounding rock monitoring area, forming multiple complete structured light infrared bands on each monitoring section of the tunnel surrounding rock. The infrared structured light band information and the natural texture information of the surrounding rock surface are fused to construct a virtual cursor, and the unique position of the infrared structured light band is ensured by a structural positioning device. A depth image super-resolution algorithm is used to perform non-contact intelligent analysis and measurement of the deformation of the infrared structured light band. Compared to the 3-7 discrete feature points typically set by total stations in existing technologies, this technology enables the monitoring of deformation of dense feature points across the entire cross section and the synchronous intelligent real-time monitoring of deformation across multiple cross sections.

[0024] S2, at least one fixed optical reference point and a reference terminal for observing the optical reference point are set up in the stable area of ​​the completed tunnel. A unified measurement coordinate system is established based on the reference terminal. The reference terminal and the monitoring terminal are synchronously collected to obtain the real-time reference image of the optical reference and the real-time measurement image of the virtual target on the monitoring section, respectively. It should be noted that in this application, the reference terminal is deployed in the stable zone of the secondary lining, and a fixed reference that does not move with the deformation of the surrounding rock is established for the entire measurement system based on the monitoring terminal, so that all deformation measurement values ​​have a unified and traceable coordinate reference, providing key input data for the subsequent dynamic compensation algorithm, and enabling the monitoring terminal to sense and quantify its own position and attitude changes.

[0025] The monitoring terminal and the reference terminal acquire data synchronously at a preset frequency. The monitoring terminal acquires measurement images of the virtual target on the monitoring section, and the reference terminal acquires reference images of the optical reference point.

[0026] The specific process for constructing a unified measurement coordinate system is as follows: Under initial stability: the reference terminal is taken as the spatial reference origin of the entire measurement system, and the initial relative position between the reference terminal and the monitoring terminal is taken as the reference position; The initial coordinates of each feature point in the initial reference image containing the optical reference point in the back-view reference section (stable area) are simultaneously acquired by the reference terminal when the monitoring terminal acquires the initial image. These coordinates are then established as the initial reference reference point for the entire measurement system. During real-time monitoring, the monitoring terminal and the reference terminal synchronously trigger the acquisition command. The acquired data is as follows: the monitoring terminal obtains a real-time measurement image containing the virtual target of the current monitoring section, and the reference terminal obtains a real-time reference image containing fixed reference points. Based on the changes in each feature point in the real-time reference image relative to the initial reference reference point, and the relative position changes of the reference terminal and the monitoring terminal, the pose change of the monitoring terminal itself is calculated. In one embodiment of this application, in order to reduce calculation, the reference terminal and the monitoring terminal are coaxially fixed together, and their fields of view are respectively facing the stable region and the monitoring region. At this time, changes in the monitoring terminal will cause the reference terminal to change synchronously. Since the optical reference in the stable region does not change, when the reference terminal changes synchronously due to the pose change of the monitoring terminal, the real-time reference image will change compared to the initial reference image. Therefore, when calculating the pose change of the monitoring terminal itself, it is only necessary to calculate the pose change of the monitoring terminal during the monitoring process by considering the changes in each feature point in the real-time reference image relative to the initial reference reference point.

[0027] S3, preprocess the real-time measurement image acquired in S2 to obtain a high-resolution infrared structured light feature line image; In this application, the preprocessing of the real-time measurement image includes: Image processing based on real-time measurement images using deep neural networks enables the segmentation of the infrared structured light band region from a complex background. Super-resolution reconstruction of the segmented light band region is then performed to obtain a high-resolution infrared structured light feature line image.

[0028] Among them, the deep neural network is the PSPNet network structure.

[0029] The specific process of image segmentation and super-resolution reconstruction is as follows: First, the real-time measurement image is input into the PSPNet network structure, and the feature map is obtained after convolutional downsampling; Then, the feature map is input into the pyramid pooling module, and its output is added to the original feature map; Finally, the image is upsampled by 8 times through convolutional layers and bilinear interpolation. The Dice Loss function is used to analyze the pixel labels of the real segmented image and the pixel categories predicted by the model. The overlap between the predicted and real results is calculated to obtain clear second image data. Finally, a high-resolution binarized infrared structured light feature line image with tens of millions of pixels is output.

[0030] In this application, image preprocessing is based on PSPNet's super-resolution segmentation, which directly and accurately segments the structured light stripe region from a complex and noisy tunnel background, effectively overcoming interference from dust adhesion, water stains, and local reflections on feature recognition. Through super-resolution technology, image details are enhanced without replacing high-resolution hardware, providing higher-quality input data for sub-pixel-level displacement calculations.

[0031] S4. The feature line image obtained in S3 is matched with the reference image. A digital image correlation algorithm based on natural texture analysis is used to obtain the sub-pixel image plane displacement of each feature point in the real-time measurement image relative to the reference image. In this application, the calculation process for sub-pixel image plane displacement is as follows: S401, the feature line image is matched with the reference image, the deformation of the feature point image sub-region is described by the shape function, and the initial value of the integer pixel displacement of the feature point is obtained by the integer pixel search algorithm. S402, Establish a quality assessment model for natural image texture based on feature points to evaluate the similarity of image sub-regions and image quality; In the quality assessment model, the similarity between feature points and natural texture images before and after deformation is evaluated using the zero-mean normalized least square distance correlation function C. ZNSSD ; The quality of natural texture images with feature points is evaluated by using the average gray-level gradient of image sub-regions.

[0032] S403, based on the Newton-Raphson iterative method, the initial value of the actual displacement of the integer pixel is optimized to obtain the sub-pixel image plane displacement with sub-pixel accuracy.

[0033] S5: Based on the position change of the optical reference point in the real-time reference image, calculate the pose change of the monitoring terminal itself, and perform geometric correction on the sub-pixel image plane displacement in S4 based on the pose change. Combined with the calibration parameters and measurement distance of the monitoring terminal, convert the corrected sub-pixel image plane displacement into actual displacement and output it. Among them, the pose change is the data after outliers have been identified and removed using the 3σ criterion; The pose change is used to perform geometric correction on the sub-pixel image plane displacement obtained in step S4, so as to eliminate the systematic error caused by the displacement or vibration of the monitoring terminal. Combining the calibration parameters of the monitoring terminal and the measurement distance, the specific process of converting the corrected sub-pixel image plane displacement into the actual displacement is as follows: Based on the inherent properties of the monitoring terminal, as well as its pose and distance relative to the monitoring section, a scaling factor between image pixels and actual dimensions is calculated. This scaling factor is then used to convert sub-pixel image plane displacement into arch settlement displacement and peripheral convergence displacement. This allows for real-time calculation and compensation of the monitoring terminal's pose changes caused by factors such as support loosening, construction vibration, and temperature deformation, based on data from a reference terminal. This ensures the stability and accuracy of long-term monitoring data and solves the core problem of reference drift in visual measurement systems operating in dynamic environments.

[0034] S6 performs time-series analysis on the actual displacement output by S5 to obtain the deformation development trend of the monitored section. Based on preset early warning rules, it determines whether to issue an early warning and generates a monitoring report. Specifically: By performing time-series analysis on the actual displacement output by S5, the deformation development trend is obtained; based on the deformation development trend, the deformation amount is predicted, and then the predicted deformation amount is compared with the preset multi-level management threshold. If the limit is exceeded, the corresponding level of warning is triggered and a monitoring report is generated.

[0035] According to another embodiment of this application, an automated monitoring system for measuring deformation of surrounding rock in tunnels is provided, the monitoring system comprising: The virtual target and optical reference construction unit includes at least one virtual target construction unit that projects an infrared structured light pattern onto the tunnel monitoring section to form a virtual target, and an optical reference point that projects infrared structured light onto the section of a completed stable area of ​​the tunnel to form an optical reference. The virtual target construction unit is installed on the sidewall of the area to be monitored, and the optical reference point is installed on the sidewall of the completed stable area. The virtual target construction unit and the optical reference point each integrate a coaxial visible light source for auxiliary aiming during equipment installation and debugging. The acquisition unit includes at least one monitoring terminal and at least one reference terminal. The monitoring terminal and the reference terminal are respectively equipped with filters that transmit only infrared structured light bands. The monitoring terminal is installed in the area to be monitored in the tunnel and is used to acquire real-time measurement images of virtual targets on the monitoring section. The reference terminal is installed in a stable area of ​​the completed tunnel and is used to acquire real-time reference images of fixed optical reference points, which are infrared reflective markers. A data processing and control unit is connected to the monitoring terminal, the reference terminal, the target construction unit, and the optical reference point, respectively. The data processing and control unit is configured as follows: The virtual target construction unit and the optical reference point are controlled to form a virtual target in the monitoring area and an optical reference in the stable area, respectively. The real-time measurement image is preprocessed, and a digital image correlation algorithm based on natural texture analysis is executed to calculate the sub-pixel image plane displacement of each feature point in the real-time measurement image relative to the reference image. The sub-pixel image plane displacement is geometrically corrected, and the sub-pixel image plane displacement is converted into actual displacement by solving the pose change of the monitoring terminal. The analysis and early warning module is connected to the data processing and control unit. Based on the time-series analysis of the converted actual displacement, the analysis and early warning module obtains the deformation trend analysis of the monitoring section, the preset threshold comparison and early warning issuance, and feeds the results back to the superior unit.

[0036] In this application, both the target construction unit and the optical reference point are infrared structured light emitting terminals. The infrared structured light emitting terminal integrates a coaxial visible light source for auxiliary aiming during equipment installation and debugging.

[0037] The present application is described in detail below with reference to the embodiments, but the present application is not limited to these embodiments.

[0038] Example: This embodiment provides a specific example of the deployment of the automated monitoring system for tunnel surrounding rock deformation described in this application and the monitoring process of the automated monitoring method for tunnel surrounding rock deformation.

[0039] In this embodiment, the area to be monitored in the tunnel is the initial support area, and the stable area is the completed and structurally stable secondary lining area or the closed invert arch area. Both monitoring terminal 2 and reference terminal 1 are 12-megapixel industrial cameras, each equipped with a protective cover and a 915nm narrowband pass filter. The 915nm narrowband pass filter is installed inside the protective covers of the monitoring terminal and the reference terminal. In this embodiment, as... Figure 2 As shown, monitoring terminal 2 and reference terminal 1 are coaxially fixed together, and their fields of view are respectively oriented towards the monitoring area (i.e., Figure 1 The initial support region and the stable region (i.e., the initial support region shown) and the stable region (i.e., the stable region shown) Figure 1 (As shown in the diagram). Optical reference point 5 and virtual target construction unit 4 are infrared structured light emitting terminals, respectively. The data processing and control unit is a PLC, which has a built-in control unit for synchronously acquiring data from the monitoring terminal and the reference terminal, an image processing algorithm integrating core algorithms such as PSPNet segmentation and sub-pixel DIC (including ZNSSD and Newton-Raphson iteration), and a dynamic reference compensation algorithm for solving the pose drift of the system (specifically the monitoring terminal) and compensating for errors. The data processing and control unit is also equipped with a data management and communication module for storing data acquired by the monitoring terminal and the reference terminal, communicating with the upper-level platform via wired (Ethernet) or wireless (bridge) means, and is also equipped with a power supply to provide stable power to the field equipment. The analysis and early warning module is a unit built into the data processing and control unit or deployed on a cloud server. It provides a web or client interface for visualization of monitoring data, curve analysis, preset threshold setting, early warning release, and report management functions. The infrared structured light emitted by the infrared structured light transmitting terminal has a wavelength of 915nm. The width of a single light strip on the monitoring section is designed to be about 1mm at a distance of 20 meters. The infrared structured light pattern is formed by the intersection of 6 horizontal and 4 vertical light strips to form a grid. The output power is 200mW, and it integrates a coaxial visible light LED to assist in visual observation of the light spot position during installation and debugging. It has an IP67 protection rating to adapt to the tunnel environment.

[0040] (1) S1, The deployment of the automated monitoring system for tunnel surrounding rock deformation is as follows: like Figure 1 and Figure 2 As shown, a series of monitoring sections are selected in the initial support area of ​​the tunnel. Infrared structured light transmitting terminals are installed on the sidewall of each monitoring section as virtual target construction units 4. Infrared structured light transmitting terminals are installed on sections in the stable region as optical reference points 5. Infrared structured light is emitted to the corresponding sections through the virtual target construction units 4 and the optical reference points 5, respectively. Then, as shown... Figure 1The infrared structured light patterns shown will form on the monitoring section and the section of the stable region, respectively, covering the arch, shoulder and sidewall of the monitoring section. A monitoring terminal 2 is installed on a stable sidewall support about 15-30 meters behind the monitoring section. A reference terminal 1 is installed in the stable area of ​​the completed tunnel, i.e., the secondary lining area. The reference terminal 1 is coaxially connected with the monitoring terminal 2. The field of view of the reference terminal 1 is aligned with the optical reference 6 fixed on the secondary lining surface. The monitoring terminal 2, the reference terminal 1, the virtual target construction unit 4, and the optical reference point 5 are respectively connected to the data processing and control unit.

[0041] (2) The process of the automated monitoring method for tunnel surrounding rock deformation is as follows: 1) Construct virtual target 3 and perform system initialization and calibration: At the moment when the surrounding rock is considered initially stable, such as in the early stages of support completion, the data processing and control unit controls the virtual target construction unit 4 and optical reference point 5 to project an infrared structured light pattern consisting of a mesh structure of 5-8 transverse and 3-5 longitudinal light bands onto the monitoring section and the stable area section, respectively. This pattern completely covers the key areas of the arch, shoulder, and sidewalls of the monitoring section and the stable area section. At a projection distance of 50 meters, the width of a single light band on the section is designed to be approximately 1 mm, and the diameter of the intersection spot is ≤2 mm. Figure 1 As shown, a ring of light spots is projected onto the tunnel surrounding rock wall to indicate the location of various features of the monitoring section and optical reference. Simultaneously, the data processing and control unit controls monitoring terminal 2 and reference terminal 1 to synchronously acquire the following images: monitoring terminal 2 acquires a clear initial image of the monitoring section, and reference terminal 1 acquires an initial reference image of the stable region section including optical reference 6. The initial image includes natural texture information of the surrounding rock surface under infrared structured light pattern illumination and structured light band information of the infrared structured light. This initial image is used as the virtual target 3 of the monitoring section and stored as a reference image for subsequent comparison by monitoring terminals. Using reference terminal 1 as the physical reference origin, the initial reference image serves as the calibration reference for the entire measurement system, constructing a unified measurement coordinate system; thus completing the construction of the virtual target 3 and the system initialization calibration.

[0042] 2) During real-time monitoring, the data processing and control unit controls monitoring terminal 2 and reference terminal 1 to synchronously trigger acquisition commands. The acquired data includes: monitoring terminal 2 acquiring a real-time measurement image containing the virtual target 3 of the current monitoring section, and reference terminal 1 acquiring a real-time reference image containing fixed reference points. The acquired data is then transmitted to the data processing and control unit via a wireless local area network. The data processing and control unit is the local intelligent measurement and monitoring platform server. The data processing and control unit is responsible for receiving the real-time measurement images of the monitoring section of the tunnel surrounding rock or the real-time reference image of the back-view reference section; calculating the deformation of each feature point of the monitoring section and the corresponding feature points in the real-time reference image; and finally summarizing the deformation of all feature points of all monitoring sections and sending it to the upper-level platform (i.e., the railway engineering management platform) to achieve online automated monitoring and early warning of tunnel surrounding rock deformation measurement.

[0043] 3) The data processing and control unit uses a deep learning model based on the PSPNet (Pyramid Scene Resolution Network) network structure to preprocess the real-time measurement images. The specific process is as follows: First, the real-time measurement images are input into the network, and feature maps are obtained by downsampling through convolutional layers; The feature map is then fed into the pyramid pooling module (PPM) to capture multi-scale contextual information. Next, the PPM output is fused with the original feature map, and then upsampled by 8 times through convolution and bilinear interpolation to restore the original image size. The Dice Loss function is used to analyze and obtain the pixel labels of the real segmented image and the pixel categories of the model-predicted segmented image. Based on the obtained pixel labels of the real segmented image and the pixel categories of the model-predicted segmented image, the overlap between the predicted and real results is calculated, resulting in a clear binary segmentation map. In the final binary segmentation map output by the PSPNet network structure, the region identified as an infrared structured light band is the foreground (white), and the background is black.

[0044] Obtain the pixel coordinates of the super-resolution infrared structured light image based on the clear binary segmentation map: ; In the formula, lu , lv These represent the actual convergence displacement in the horizontal direction and the actual settlement displacement in the vertical direction, respectively, in mm; lu' and lv' These represent the horizontal and vertical pixel coordinates after deformation, in pixels. δ This indicates a systematic error.

[0045] The acquired infrared structured light image pixel coordinates are transmitted to the surrounding rock structure deformation judgment module to analyze the changes in the structured light image within the tunnel surrounding rock structure area of ​​the real-time super-resolution segmentation image.

[0046] The super-resolution infrared structured light image is compared with the tunnel surrounding rock structure section to obtain the left and right rotation angle αc and the pitch angle βc between the monitoring terminal and the tunnel surrounding rock structure section. Based on the errors in the left-right rotation angle αc and pitch angle βc of the monitoring terminal, the installation angles of the monitoring terminal and the reference terminal are adjusted respectively, and the image pixel coordinates of the infrared structured light after angle correction are obtained to ensure the accuracy of the monitoring image data. Specifically: Methods for distortion correction in monitoring terminals: The calibrated monitoring terminal's intrinsic and distortion parameters are obtained, and the relationship between the coordinates of the spatial 3D reference point M and the pixel coordinates of the monitoring terminal is obtained based on the pinhole principle. This allows for the acquisition of the monitoring terminal's planar pixel coordinates. lx' and ly' ; The focal length in the parameters of the monitoring terminal will be monitored. fx , fy Optical center coordinates cx , c Substitute y into the pixel coordinates for calculation, and transform the pixel coordinates into normalized image coordinates before distortion correction. lx' and ly' : ; The radial distortion parameter among the calibrated distortion parameters k 1 , k 2 and k 3 Tangential distortion parameters p 1 and p 2 The coordinates of the measurement and monitoring terminal are distorted to obtain the distorted coordinates of the monitoring terminal. x'' and y'' : ; In the formula, r 2 : The square of the distance from the pixel to the optical center r 2 = lx' 2 + ly' 2 .

[0047] Installation error correction of monitoring terminal: The super-resolution infrared structured light image is compared with the tunnel cross section to obtain the left and right rotation angle αc=α1c+α2c and the pitch angle βc=β1c+β2c that exist between them. In the formula, αc represents the horizontal angular deviation of the monitoring terminal; α1c represents the error of the horizontal installation angle of the monitoring terminal itself; α2c is the horizontal installation rotation error of the reference terminal; βc is the vertical angular deviation; β1c is the error of the vertical installation angle of the monitoring terminal itself; and β2c is the vertical installation deviation of the reference terminal.

[0048] Based on the left and right rotation angle α of the monitoring terminal c The pitch angle βc error was adjusted by changing the installation angles of the measurement and monitoring terminal and the reference terminal to obtain the corrected pixel coordinates of the angle-corrected infrared structured light image. u'' , v'' They are respectively: .

[0049] The calculations for infrared structured light, when it is clearly visible, include the following steps: Infrared structured light is extracted from real-time reference images or real-time measurement images, and the highest point of the center line of the infrared structured light is taken as the apex of the arch. Multiple sidewall convergence regions are divided on both sides of the apex of the arch. The infrared structured light in the real-time reference image or real-time measurement image is matched with the infrared structured light of the stored initial reference infrared structured light or reference image to analyze the positional deformation of feature points located on the arch and / or sides; the positional deformation is a variable in both vertical and horizontal directions; the stability and safety of the tunnel surrounding rock structure and support system are determined in real time to provide support for tunnel structure deformation prediction; and the deformation of the arch height and sidewall feature points in the tunnel surrounding rock structure is obtained through analysis.

[0050] The Dice Loss function is used to optimize segmentation accuracy during network training. The formula for the Dice Loss function is as follows:

[0051] Where Y represents the set of pixel labels in the real segmented image, and represents the actual infrared structured light band. The pixel region it is located in; The set of pixel categories in the image segmentation predicted by the model represents the set of infrared nodes identified by the algorithm. Light-forming pixel area; This represents the total number of pixels in the predicted segmented image, i.e., the number of pixels covered by the infrared structured light band identified by the algorithm; |Y| represents the total number of pixels in the actual segmented image, that is, the number of pixels actually covered by the infrared structured light strip.

[0052] This process not only enables robust segmentation of infrared structured light stripes from complex backgrounds, but also enhances the clarity of stripe edges through super-resolution technology, outputting a high-resolution image of infrared structured light feature lines, laying the foundation for subsequent high-precision displacement calculations.

[0053] 4) Sub-pixel image plane displacement calculation, the specific process is as follows: ① Feature point selection: In the reference image stored in 1), a series of feature points are selected along the center line of the infrared structured light band. With each feature point as the center, a rectangular image sub-region is taken as the template. ② Perform integer pixel search: In the infrared structured light feature line image obtained after preprocessing in step 3), using the initial position of the aforementioned feature points as the center, a fast algorithm such as the three-step search method is used to find the region most similar to the template grayscale distribution, thus obtaining the initial integer pixel displacement value of the feature points. u 0 ,v 0 ); ③ Construct a quality assessment model for the natural texture of high-quality images. To accurately describe deformation, a first-order shape function is used to model the displacement and deformation of image sub-regions. The expression for the first-order shape function is as follows:

[0054] In the formula: , for The first-order partial derivative; , for The first-order partial derivative.

[0055] in, x, y The coordinates of feature points within the reference sub-region; x', y' These are the coordinates of the feature points after deformation; u、v For translation components; u x 、u y 、v x 、v y These are the displacement gradient components; x 0 , y 0 These are the pixel coordinates of a selected reference feature point within a sub-region of the reference image; ∆ x The horizontal pixel coordinates of any pixel in the reference image sub-region x Horizontal coordinates of the reference point x0 The difference reflects the lateral offset of the point relative to the reference point (unit: pixels); ∆ y The vertical pixel coordinates of any pixel in the reference image sub-region y Vertical coordinates of the reference point y 0 The difference reflects the longitudinal offset of the point relative to the reference point (unit: pixels); the deformation parameter vector to be solved is... .

[0056] To evaluate matching quality and image usability, zero-mean normalized least square distance correlation was used. function( C ZNSSD As a similarity measure, it evaluates the similarity of feature points in natural texture images before and after deformation, and is insensitive to linear changes in illumination. The average gray-level gradient of image sub-regions is used to evaluate the quality of the image itself, expressed as:

[0057] in: in, , ; In the formula: f(x,y) Indicates the reference image sub-region in x,y grayscale field; g ( x', y' ) is the target image sub-region in the corresponding x'、y' grayscale value at that location M This represents the total number of pixels in the reference / target image sub-region. f m This represents the average gray value of a sub-region of the reference image; g m This represents the average gray value of a sub-region of the target image after deformation.

[0058] Zero-mean normalized least square distance correlation function The range of values ​​is Through simple mathematical derivation, we can obtain... .

[0059] ④ Optimize sub-pixel image plane displacement: using initial integer pixel displacement values As the starting point of the iteration, the Newton-Raphson iterative method is used to optimize the solution of deformation parameters with sub-pixel accuracy. and They are respectively and The integer pixel search results. By minimizing C ZNSSD The function value is iteratively solved until convergence, ultimately yielding a high-precision sub-pixel image plane displacement. du ,d v ).

[0060] like Figure 3 As shown, reference image sub-region f(x, y) With target image sub-region g'(x',y') The coordinate correspondence satisfies: ; and Indicates horizontal displacement u Displacement gradients in the x and y directions; and Represents vertical displacement v Displacement gradients in the x and y directions; 1) Construct the vector to be solved .

[0061] 2) Establish relevant functions: 3) When the deformation parameters to be determined The objective function reaches its minimum value when the gray levels of sub-regions in the image before and after deformation are most similar, meaning the gray level gradient is 0.

[0062] in, This is the first-order gradient vector of the correlation function. Let be the second-order gradient matrix of the correlation function. This is the initial value for the iteration. Calculate... First-order partial derivatives And set it to zero, that is: 4) Solve using the Newton-Raphson iterative method. The initial value for the iteration is... ,in and They are respectively and The search results are integer pixels. During the iteration process, it is also necessary to calculate... Second-order partial derivatives , Approximate values ​​can also be used:

[0063] It should be noted that during the iteration process, the gray level and gray level gradient value of the sub-pixel position of the image sub-region need to be calculated through the above (2).

[0064] The iterative calculation formula is:

[0065] Through formula Convergence is achieved when the value is less than a set threshold ℇ, ultimately yielding high-precision sub-pixel image plane displacement for each feature point. d u , d v ), where: P i P represents the deformation parameter value in the i-th iteration. i+1 This represents the deformation parameter value for the (i+1)th iteration, with an accuracy of 0.1 pixels or even higher.

[0066] 5) Dynamic error compensation and actual displacement conversion, specifically: Because the support of monitoring terminal 2 may experience slight displacement or rotation due to construction vibration, temperature changes, etc., its own pose change may be misjudged as surrounding rock deformation. This step uses data from reference terminal 1 to eliminate this error; thereby eliminating the drift error of the monitoring system itself and converting the sub-pixel image plane displacement into the true physical displacement.

[0067] a. Process the real-time reference image acquired by reference terminal 1. Track the coordinate change (Δ) of optical reference point 5 in the image using a sub-pixel algorithm. u b , Δ v b Since the reference terminal and the optical reference point are fixed, the change of the reference terminal 1 itself can represent the pose change of the entire measurement system. For example, translation and rotation, the change of the coordinates of the real-time reference image of the reference terminal 1 directly reflects the change of the reference terminal itself. Therefore, by combining the intrinsic parameters of the reference terminal and the known distance of the optical reference point, the small displacement and rotation changes of the reference terminal itself in three-dimensional space, i.e., the pose change ΔT, can be calculated.

[0068] b) The pose change sequence over a period of time is filtered using the 3σ criterion. Outliers caused by strong instantaneous disturbances are identified, and the results are as follows: Figure 4 As shown, the change curve of the data sequence after removing outliers by 3σ does not show obvious abrupt changes compared to the original data sequence, and the overall change trend is relatively smooth.

[0069] c, using the calculated system pose change ΔT Construct a reverse coordinate transformation to apply the sub-pixel image plane displacement to all feature points calculated above. d u ,d v ) Perform reverse geometric correction to obtain the image plane displacement of the corrected displacement caused purely by the deformation of the surrounding rock. du',dv' ); d, the corrected image plane displacement (du',dv') The actual displacement converted to the tunnel cross-section is as follows: After obtaining the image plane displacement of each feature point (du',dv') It is also necessary to calculate the ratio of image pixels to actual size, i.e., the calibration coefficient. K Only then can the actual displacement values ​​of each feature point be calculated. Based on the geometric model of imaging, the scaling factor is derived. K It can be represented as:

[0070] In the formula: Calibration factor, mm / pixel; f Lens focal length, mm; The pixel size of the image sensor is in μm; au 0, bv 0) represents the pixel coordinates of the camera's principal point; au , bv () represents the pixel coordinates of the feature point; D denoted as , which is the distance from the feature point to monitoring terminal 2, expressed in meters (m).

[0071] Calculate the pixel resolution S (mm / pixel) of the real-time measured image at the monitored section: In a two-dimensional digital image correlation measurement system, the pixel size p, sensor size s, and lens focal length f of monitoring terminal 2 are fixed known parameters. Based on the vertical elevation angle α and horizontal angle β of monitoring terminal 2, the imaging position (x, y) of the feature point in the real-time measurement image, and the actual distance d between the feature point and monitoring terminal 2, the scaling factor of the feature point on the image plane and the object plane can be calculated. K(X,Y)=k(p,s,f,α,β,(x,y),d) Under these conditions, the image plane displacement [u(x,y),v(x,y)] obtained by digital image correlation methods is clearly linearly proportional to the object surface displacement [U(X,Y),V(X,Y)], expressed as: u(x,y)=K U (X,Y); v(x,y)=K V (X,Y); Therefore, the pixel resolution of the real-time measured image of the tunnel surrounding rock monitoring section at the feature points can be calculated based on the real-time measured image. p ( X (Unit: mm / pixel); its calculation formula is:

[0072] In the formula: x 1.x 2 represents the actual coordinates of the two feature points in the horizontal direction on the object surface; y 1. y 2 represents the actual coordinates of the two feature points in the direction perpendicular to the object surface; z 1. z 2 represents the actual distance (depth coordinates of the object surface) from the two feature points to the monitoring terminal. m 1. m 2 represents the pixel coordinates of the two feature points in the horizontal direction of the image; n 1. n 2 represents the pixel coordinates of the two feature points in the vertical direction of the image; Actual settlement of tunnel surrounding rock monitoring section H(∆) for:

[0073] In the formula, m (t1) represents the vertical pixel coordinate of the arch feature point at the initial time t1; m (t2) represents the vertical pixel coordinate of the arch feature point at monitoring time t2; P ( x ) represents the settlement correction factor.

[0074] The horizontal convergence of the tunnel surrounding rock monitoring section is L(∆) for: ; In the formula: m (a) m (b) represents the lateral pixel coordinates of the convergent feature points on the left and right sides; P ( x ) represents the convergence correction factor (same as the settlement correction factor, calibrated on site); n (a) n (b) represents the vertical pixel coordinates of the convergent feature points on the left and right sides; L (0) represents the horizontal reference length of the cross section at the initial moment, in mm.

[0075] This embodiment establishes a three-level early warning mechanism: normal, warning, and alarm. Preset thresholds are set based on the surrounding rock grade or design requirements. Real-time deformation rate and cumulative deformation are automatically compared with the corresponding preset thresholds. For example, when the deformation rate continuously exceeds the set value, or the cumulative deformation reaches 70% of the managed value, an early warning is triggered (yellow indicator, SMS notification); when the managed value is reached or exceeded, an alarm is triggered (red flashing, audible and visual alarm, notification to responsible personnel).

[0076] It can also be uploaded to the superior project management platform via network interface.

[0077] In this embodiment, a non-visible light source combined with natural texture information of the tunnel surrounding rock is used to replace the traditional physical target, achieving sub-millimeter-level measurement across the entire cross-section, thus improving image quality and measurement coverage. Based on traditional digital image correlation algorithms, the limitations of hardware CCD resolution are overcome, improving the theoretical accuracy of long-distance measurements. Simultaneously, a quality evaluation model based on natural texture analysis is established. By unifying the sub-millimeter-level spatiotemporal reference and dynamic spatial position compensation, the error accumulation problem caused by camera distortion and reference changes in multi-system, multi-target, and multi-time-period operations is solved. Combining image preprocessing and surrounding rock structure deformation judgment algorithms, super-resolution processing is performed on poor-quality observation data, improving the accuracy and reliability of measurement calculations.

[0078] To verify the effectiveness and accuracy of this system, synchronous comparative monitoring was conducted between the monitoring system described in this application and a Leica TS16 total station at 63 cross-sections, including the Mangkangshan Tunnel and Honglashan Tunnel. The results are shown in Table 1. Table 1

[0079] As shown in Table 1, the correlation coefficient between the two monitoring data is ≥0.99, and the difference in cumulative deformation is ≤1.5mm, which meets the accuracy requirements of the "Technical Specification for Monitoring and Measurement of Railway Tunnels" (Q / CR 9218-2015). Under high dust conditions such as tunnel blasting and slag removal, the monitoring system described in this application maintains stable image acquisition and high-precision measurement results by means of infrared light source and filter.

[0080] This application compares the results of the test system described in this application with those of the total station measurement at the right entrance of the #Huajiaoqing No. 2 Tunnel on the Yunyong Expressway, with a frequency of 30 seconds / time to 10 minutes / item. The results are as follows: Figures 5-10 As shown, by Figures 5-10 It can be seen that: The settlement curves of the arch crown and the peripheral convergence curves of all monitoring sections show that the deformation trends measured by the monitoring system and the total station are completely consistent. The curves fluctuate synchronously without any reverse deviation or trend divergence, confirming that the correlation coefficient between the two measurement data is ≥0.99. The cumulative difference of the monitoring system described in this application is ≤1.5mm, meeting the accuracy requirement of ≤2mm cumulative deformation measurement error in the "Technical Specification for Monitoring and Measurement of Railway Tunnels". Regarding the peripheral convergence, the cumulative convergence difference is ≤1mm and the standard deviation is ≤1.0mm, which is far below the error threshold allowed by the specification, demonstrating the sub-millimeter measurement accuracy advantage of the monitoring system described in this application.

[0081] And by appendix Figure 5The continuous monitoring over the past two months shows that the curve is smooth and without abrupt changes, which confirms that the dynamic reference compensation technology effectively suppresses the positional drift of the monitoring terminal and ensures the reliability of long-term monitoring data. At the same time, compared with the discrete point measurement of a total station, the system of this application achieves continuous monitoring of the entire cross section without reducing the measurement accuracy. The above description is merely a few embodiments of this application and is not intended to limit this application in any way. Although this application discloses preferred embodiments as described above, it is not intended to limit this application. Any changes or modifications made by those skilled in the art without departing from the scope of the technical solution of this application using the disclosed technical content are equivalent to equivalent implementation cases and fall within the scope of the technical solution.

Claims

1. An automated monitoring method for measuring deformation of surrounding rock in tunnels, characterized in that, The monitoring method includes: S1, a preset infrared structured light pattern is projected onto the monitoring section of the surrounding rock in the area to be monitored to form an infrared structured light band covering the monitoring section. Based on the initial image of the monitoring section collected by the monitoring terminal in the initial stable state, including the infrared structured light band information and the natural texture information of the surrounding rock surface of the monitoring section, the infrared structured light band information and the natural texture information are fused based on the initial image to construct a virtual target of the monitoring section and store it as a reference image. S2, at least one fixed optical reference point and a reference terminal for observing the optical reference point are set up in the stable area of ​​the completed tunnel. Infrared structured light is emitted from the optical reference point to the cross section of the stable area to form an optical reference. A unified measurement coordinate system is established based on the reference terminal. The reference terminal and the monitoring terminal are synchronously collected to obtain the real-time reference image of the optical reference and the real-time measurement image of the virtual target, respectively. S3, preprocess the real-time measurement image acquired in S2 to obtain a high-resolution infrared structured light feature line image; S4. The feature line image obtained in S3 is matched with the reference image. A digital image correlation algorithm based on natural texture analysis is used to obtain the sub-pixel image plane displacement of each feature point in the real-time measurement image relative to the reference image. S5: Based on the position changes of each feature point in the real-time reference image, calculate the pose change of the monitoring terminal itself, and perform geometric correction on the sub-pixel image plane displacement in S4 based on the pose change. Combined with the calibration parameters and measurement distance of the monitoring terminal, convert the corrected sub-pixel image plane displacement into actual displacement and output it. S6 performs time-series analysis on the actual displacement output by S5 to obtain the deformation development trend of the monitored section, and makes judgments and issues warnings based on preset warning rules.

2. The automated monitoring method for measuring deformation of surrounding rock in tunnels according to claim 1, characterized in that, In S1, the infrared structured light pattern is a grid pattern, and the infrared structured light pattern covers the arch, shoulder and side wall areas of the monitoring section.

3. The automated monitoring method for measuring deformation of surrounding rock in tunnels according to claim 1, characterized in that, In S1, the infrared structured light is a near-infrared invisible laser, and the wavelength of the near-infrared invisible laser is any one of 850nm, 910nm and 915nm. The laser linewidth of the infrared structured light strip is ≤1.5nm, and the linewidth of a single laser on the projection surface is 0.8mm~1.2mm; The diameter of the light spot at the intersection of the infrared structured light bands is ≤2mm.

4. The automated monitoring method for measuring deformation of surrounding rock in tunnels according to claim 1, characterized in that, In S3, the preprocessing of the real-time measurement image includes: image processing of the real-time measurement image based on a deep neural network to segment the light band region of the infrared structured light band from the complex background of the real-time measurement image, and super-resolution reconstruction of the segmented light band region to obtain a high-resolution infrared structured light feature line image.

5. The automated monitoring method for measuring deformation of surrounding rock in tunnels according to claim 4, characterized in that, The deep neural network is a PSPNet network structure, in which the image segmentation and super-resolution reconstruction process is as follows: First, the real-time measurement image is input into the PSPNet network structure, and the feature map is obtained after convolutional downsampling; Then, the feature map is input into the pyramid pooling module, and its output is added to the original feature map; Finally, the segmentation results are upsampled through convolutional layers and bilinear interpolation, and optimized using the Dice Loss function to output a binarized infrared structured light feature line image.

6. The automated monitoring method for measuring deformation of surrounding rock in tunnels according to claim 1, characterized in that, In S4, the calculation process for sub-pixel image plane displacement is as follows: S401, the feature line image is matched with the reference image, the deformation of the feature point image sub-region is described by the shape function, and the initial value of the integer pixel displacement of the feature point is obtained by the integer pixel search algorithm. S402, Establish a quality assessment model for natural image texture based on feature points to evaluate the similarity of image sub-regions and image quality; S403, based on the Newton-Raphson iterative method, the initial value of the actual displacement of the integer pixel is optimized to obtain the sub-pixel image plane displacement with sub-pixel accuracy.

7. The automated monitoring method for measuring deformation of surrounding rock in tunnels according to claim 6, characterized in that, In the quality assessment model of S402, the similarity of feature points in the natural texture image before and after deformation is evaluated using the zero-mean normalized least square distance correlation function C. ZNSSD ; The quality of natural texture images with feature points is evaluated by using the average gray-level gradient of image sub-regions.

8. The automated monitoring method for measuring deformation of surrounding rock in tunnels according to claim 1, characterized in that, In S5, the pose change is the data after outliers are identified and removed using the 3σ criterion; The specific process of converting the corrected sub-pixel image plane displacement into actual displacement includes: First, calculate the scaling factor between the subpixel surface displacement and the actual displacement. K ; Finally, using the scaling factor K The subpixel image plane displacement is converted into actual displacement, which includes the arch settlement displacement and the peripheral convergence displacement.

9. An automated monitoring system for measuring deformation of surrounding rock in tunnels, characterized in that, For implementing the automated monitoring method for measuring tunnel surrounding rock deformation according to any one of claims 1 to 8, the monitoring system comprises: The virtual target and optical reference construction unit includes at least one virtual target construction unit that projects an infrared structured light pattern onto the tunnel monitoring section to form a virtual target, and an optical reference point that projects infrared structured light onto the section of a completed stable area of ​​the tunnel to form an optical reference. The virtual target construction unit is installed on the sidewall of the area to be monitored, and the optical reference point is installed on the sidewall of the completed stable area. The virtual target construction unit and the optical reference point each integrate a coaxial visible light source for auxiliary aiming during equipment installation and debugging. The acquisition unit includes at least one monitoring terminal and at least one reference terminal. The monitoring terminal and the reference terminal are respectively equipped with filters that only transmit infrared structured light bands. The monitoring terminal is installed in the area to be monitored in the tunnel and is used to acquire real-time measurement images of the virtual target on the monitoring section. The reference terminal is installed in the completed and stable area of ​​the tunnel and is used to acquire real-time reference images of the optical reference point. A data processing and control unit is connected to the monitoring terminal, the reference terminal, the target construction unit, and the optical reference point, respectively. The data processing and control unit is configured as follows: The virtual target construction unit and the optical reference point are controlled to form a virtual target in the monitoring area and an optical reference in the stable area, respectively. The real-time measurement image is preprocessed, and a digital image correlation algorithm based on natural texture analysis is executed to calculate the sub-pixel image plane displacement of each feature point in the real-time measurement image relative to the reference image. The sub-pixel image plane displacement is geometrically corrected, and the corrected sub-pixel image plane displacement is converted into actual displacement and output. The analysis and early warning module is connected to the data processing and control unit. The analysis and early warning module obtains the deformation development trend of the monitoring section based on the time sequence analysis of the converted actual displacement, and makes judgments and issues early warnings based on preset early warning rules.

Images