Tunnel face three-dimensional reconstruction device and method based on light field photographing technology

Abstract

The invention provides a tunnel face three-dimensional reconstruction device and method based on a light field photographing technology, and relates to the technical field of tunnel construction detection. The device comprises a tracked robot, a folding mechanical arm arranged on the tracked robot, a supporting frame arranged at the top end of the folding mechanical arm, a plurality of sets of light field cameras and a plurality of sets of ultrasonic sensors, wherein the light field cameras and the ultrasonic sensors are arranged on the supporting frame, and the light field cameras are used for collecting light field data of different visual angles of a tunnel face; and a computer electrically connected with the tracked robot, the light field camera and the ultrasonic sensor is arranged on the tracked robot. The defects in the prior art are overcome, high-precision three-dimensional reconstruction of the tunnel face is achieved by collecting light field data and combining an efficient image processing algorithm, the safety, efficiency and monitoring precision of tunnel construction can be improved, and wide application prospects are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of tunnel construction detection, and particularly to a three-dimensional reconstruction device and method for a tunnel face based on light field photography technology. Background Art

[0002] The tunnel face is a crucial part in tunnel construction, and its state directly affects the safety, quality, and efficiency of construction. Due to the complex and changeable geological conditions of the tunnel face, especially in adverse geological environments such as soft rock, faults, or rich groundwater, real-time monitoring and assessment of its morphology and geological characteristics have become an important means to ensure construction safety. However, there are still many deficiencies in the current three-dimensional reconstruction and monitoring technologies for the tunnel face in the industry.

[0003] Currently, the monitoring of the tunnel face mainly relies on traditional measurement methods such as laser scanning, photogrammetry, and total station measurement. Although these methods can provide geometric information of the tunnel face, they have significant limitations. Taking laser scanning as an example, although it has high measurement accuracy, it has high equipment costs, complex data processing, and strong dependence on the lighting environment, and is not suitable for tunnel construction environments with low light or no light conditions. Photogrammetry technology generates three-dimensional models through multi-viewpoint photos, but it is sensitive to shooting angles, scene complexity, and lighting conditions, and is easily affected by the external environment. In addition, the real-time performance and convenience of traditional monitoring methods are insufficient. Usually, it takes a long time to complete data processing, and it is difficult to meet the immediate monitoring requirements of a dynamic construction environment.

[0004] With the increasing requirements for safety and efficiency in modern tunnel construction, the demand for tunnel face monitoring technology is also increasing. During the construction of super-large cross-section tunnels and deep-buried tunnels, stress changes in the tunnel face and surrounding rock masses and small fluctuations in geological conditions may cause serious safety hazards. Therefore, how to achieve more efficient, accurate, and real-time monitoring has become an urgent problem to be solved. Summary of the Invention

[0005] Aiming at the deficiencies of the prior art, the present invention provides a three-dimensional reconstruction device and method for a tunnel face based on light field photography technology. By collecting light field data and combining efficient image processing algorithms, high-precision three-dimensional reconstruction of the tunnel face is achieved, which can improve the safety, efficiency, and monitoring accuracy of tunnel construction and has broad application prospects.

[0006] To achieve the above objectives, the present invention provides the following technical solutions:

[0007] A three-dimensional reconstruction device for tunnel face based on light field photography technology, the device includes a crawler robot and a folding robotic arm arranged above the crawler robot; a support frame is arranged at the top of the folding robotic arm; several light field cameras for collecting light field data of different perspectives of the tunnel face are arranged on the support frame; and several ultrasonic sensors are also arranged on the support frame; a computer is arranged on the crawler robot, and the computer is electrically connected to the crawler robot, the light field cameras, and the ultrasonic sensors.

[0008] Preferably, the support frame is hemispherical; several of the light field cameras and several of the ultrasonic sensors are evenly distributed in a ring on the inner side of the support frame.

[0009] Preferably, several lighting lamps are arranged on the support frame, and several of the lighting lamps are evenly distributed in a ring on the support frame.

[0010] The three-dimensional reconstruction method for tunnel face based on light field photography technology includes the following steps:

[0011] S1. Move the device to the tunnel face detection area through the crawler robot;

[0012] S2. Collect light field data of different perspectives of the tunnel face through several of the light field cameras;

[0013] S3. Preprocess the collected light field data through the computer;

[0014] S4. Extract depth information from the processed light field data and generate a three-dimensional point cloud;

[0015] S5. Reconstruct the three-dimensional point cloud to generate a three-dimensional model;

[0016] S6. Calibrate the accuracy of the three-dimensional model by using the accurate position information provided by several of the ultrasonic sensors;

[0017] S7. Visualize and display the three-dimensional model through the computer.

[0018] Preferably, the specific method for preprocessing light field data in step S3 includes the following steps:

[0019] S3-1. Data denoising: Eliminate noise in the light field data through a filtering algorithm;

[0020] S3-2. Image enhancement: Improve the clarity of the face features by adjusting contrast and brightness;

[0021] S3-3. Viewpoint alignment: Perform spatial alignment on light field data of different perspectives through an image registration algorithm.

[0022] Preferably, in step S4, the method for extracting depth information from the processed light field data is to use a depth extraction algorithm based on ray rearrangement and parallax calculation to generate a high-precision depth map.

[0023] Preferably, the specific method for reconstructing the three-dimensional point cloud in step S5 includes the following steps:

[0024] S5-1. Point cloud optimization: including noise point removal, data smoothing, and surface interpolation to fill holes;

[0025] S5-2. Triangulation meshing: generating a complete three-dimensional surface model through Delaunay triangulation.

[0026] Preferably, the noise point removal in step S5-1 includes:

[0027] Statistical filtering: judging based on the distance mean value between each point and its neighborhood points, and removing the point cloud data points with excessive deviation;

[0028] Radius filtering: retaining the points in the point cloud where the number of neighborhood points meets certain requirements and removing the isolated points.

[0029] Preferably, the specific operation of step S6 includes real-time adjustment and optimization of the collected three-dimensional model data, and precise compensation for equipment errors, deviations caused by movement, and environmental changes.

[0030] Preferably, the three-dimensional model visualization in step S7 includes:

[0031] S7-1. Dynamic visualization: displaying the three-dimensional shape of the tunnel face through real-time rendering software;

[0032] S7-2. Analysis tools: including functions such as crack distribution, deformation measurement, and geological feature annotation.

[0033] The present invention provides a three-dimensional reconstruction device and method for a tunnel face based on light field photography technology. Compared with the prior art, the advantages are as follows:

[0034] (1) By applying the light field photography technology, the present invention can collect richer depth information on the tunnel face, and thus achieve high-precision three-dimensional reconstruction;

[0035] (2) By real-time collecting and processing light field data, the present invention can realize dynamic monitoring of the tunnel face and provide timely feedback for tunnel construction;

[0036] (3) The precise three-dimensional model of the present invention can help construction personnel better understand the state of the tunnel face, contribute to early detection of potential safety hazards, and thus effectively improve construction safety;

[0037] (4) The device and method provided by the present invention adopt automated data processing and visualization technologies, simplifying the operation process of tunnel face monitoring and reducing the need for manual intervention. Description of the Drawings

[0038] Figure 1 It is a three-dimensional structure diagram provided by an embodiment of the present invention;

[0039] Figure 2 It is a three-dimensional structure diagram provided by an embodiment of the present invention;

[0040] Figure 3 It is a three-dimensional implementation structure diagram provided by an embodiment of the present invention;

[0041] Figure 4 It is a schematic diagram of the working state of the light field camera provided by an embodiment of the present invention;

[0042] Figure 5 It is a flowchart of the steps of the three-dimensional reconstruction method provided by an embodiment of the present invention;

[0043] Figure 6 It is a flowchart of the preprocessing of light field data provided by an embodiment of the present invention;

[0044] Figure 7 It is a flowchart of the three-dimensional point cloud reconstruction provided by an embodiment of the present invention;

[0045] In the figure: 1, tracked robot; 2, folding robotic arm; 3, support frame; 4, light field camera; 5, computer; 6, ultrasonic sensor; 7, lighting lamp. Detailed Embodiments

[0046] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the scope of protection of the present invention.

[0047] Embodiment 1:

[0048] A three-dimensional reconstruction device for a tunnel face based on light field photography technology:

[0049] For the specific device, refer to Figure 1-4 , including a tracked robot 1, a folding robotic arm 2, a support frame 3, a light field camera 4, a computer 5, and an ultrasonic sensor 6;

[0050] The tracked robot 1 serves as the mobile platform of the device, with excellent terrain adaptability, capable of operating stably in the complex terrain environment within the tunnel. At the same time, it has a built-in power system, adopts a battery-powered mode, and supports long-term continuous operation. The folding robotic arm 2 is installed on the tracked robot 1, and the support frame 3 is installed at the top of the folding robotic arm 2. The folding robotic arm 2 is connected to the tracked robot 1 and is used to support and adjust the height and angle of the support frame 3 to ensure that the light field camera 4 and the ultrasonic sensor 6 can cover all areas of the tunnel face;

[0051] Several groups of light field cameras 4 are provided, and all the several light field cameras 4 are installed on the support frame 3 to collect light field data from different perspectives of the tunnel face. The light field camera 4 can simultaneously collect image information from multiple perspectives and provide depth information. Moreover, the array-type light field camera 4 can capture rich light field information, thereby simultaneously obtaining multi-perspective and depth information, providing high-precision data for 3D model reconstruction. The computer 5 is installed on the tracked robot 1, and the computer 5 is electrically connected to the tracked robot 1 and several light field cameras 4 to process images, generate 3D models, and display images. The computer 5 can be used to process the data collected by the light field camera 4, perform depth calculations and 3D reconstructions, and at the same time control the operation of the device. Several groups of ultrasonic sensors 6 are provided, and all the several ultrasonic sensors 6 are installed on the support frame 3 and electrically connected to the computer 5. They can be used to measure the distance between the device and the tunnel face, provide reference data for the calibration of the 3D model, and the ultrasonic sensors 6 work synchronously with the data collected by the light field camera 4 to ensure the unity of spatial coordinates.

[0052] And the support frame 3 is hemispherical, and several light field cameras 4 and several ultrasonic sensors 6 are evenly distributed in a ring on the inner side of the support frame 3. The ring distribution of the light field camera 4 and the ultrasonic sensor 6 combined with the hemispherical support frame 3 realizes the non-blind-zone coverage of the tunnel face, ensures that the data acquisition devices can be evenly distributed, and improves the comprehensiveness of data acquisition.

[0053] Several lighting lamps 7 are installed on the support frame 3. The lighting lamps 7 can provide uniform illumination for the light field camera 4 and the ultrasonic sensor 6 to ensure normal data acquisition in the low-light environment within the tunnel. The several lighting lamps 7 are evenly distributed in a ring on the support frame 3 to avoid light blind zones.

[0054] Embodiment 2:

[0055] Using the device set in the above Embodiment 1, the following method is referred to for 3D reconstruction of the tunnel face:

[0056] S1. Move the device to the tunnel face detection area through the tracked robot 1;

[0057] The position of the light field camera 4 can be adjusted by the crawler robot 1, and the height and shooting angle of the light field camera 4 can be adjusted by cooperating with the folding robotic arm 2. Meanwhile, the light source is supplemented for the light field camera 4 by the lighting lamp 7;

[0058] S2. Collect the light field data of different perspectives of the tunnel face through a number of light field cameras 4;

[0059] The layout of the light field camera 4 array and the collection of light field information are completed through the following mechanism:

[0060] 1. Multi-perspective collection:

[0061] Through multiple light field cameras 4 distributed in a hemispherical shape, each light field camera 4 can be responsible for collecting and recording the images within its own perspective range, including the intensity and color of the light. Furthermore, different light field cameras 4 collect light from different angles. The perspective differences of different light field cameras 4 photographing the same target point provide the integrity of the light direction. The superposition of these images can record the spatial distribution and direction change of the light, avoiding blind spots. The coverage areas of multiple light field cameras 4 intersect with the light direction, making the light field data of the entire scene complete and realizing the complete collection of light field information;

[0062] 2. Parallax information:

[0063] The light field information captured by the camera array depends on the parallax phenomenon: the position of the target point is different in the images of different cameras;

[0064] Using the internal and external parameters of the camera (focal length, baseline length, etc.), the parallax can be used to calculate the depth;

[0065] Parallax formula:

[0066]

[0067] d: Depth of the target point;

[0068] f: Focal length of the camera;

[0069] B: Baseline (camera spacing);

[0070] Δx: Parallax (pixel offset of the target point in the images of different perspectives);

[0071] Through parallax and geometric calculations, the depth coordinates of each ray are obtained.

[0072] 3. Light direction:

[0073] The direction of each ray is determined by the position of the camera and the pixel coordinates of the target point;

[0074] In the images of multiple cameras, the position change of the target point provides the angular information (θ, φ) of the light propagation.

[0075] 4. Multi-view Fusion:

[0076] The multi-view images recorded by the camera array are fused into discrete samples of light field data through feature matching and geometric alignment, and a continuous light field is formed through an interpolation algorithm during reconstruction.

[0077] The light field camera 4 can collect images of multiple perspectives of the tunnel face at one time through array distribution, with a large coverage range, avoiding the disadvantages of traditional cameras that require multiple moves for shooting. In the monitoring of the tunnel face, this characteristic enables it to quickly capture the overall view and refine local features. Moreover, by recording the direction information of light rays, the light field camera 4 can directly calculate the depth map in the scene, providing high-precision data for the 3D reconstruction of the tunnel face. Compared with traditional technologies such as laser scanning, the depth extraction process of the light field camera 4 is more straightforward and does not require additional hardware support. At the same time, through the processing of light field data, multi-view simulation of the scene can be achieved, and scene images of other perspectives can be reconstructed even under restricted shooting angles. The light field camera 4 array captures complete light field information from both spatial and angular aspects through the collaborative work of multiple views and multiple directions, providing rich basic data for 3D scene reconstruction.

[0078] S3. Preprocess the collected light field data through the computer 5;

[0079] The preprocessing of light field data includes the following steps:

[0080] S3-1. Data denoising, eliminating noise in the light field data through a filtering algorithm;

[0081] 1. Gaussian filtering: Smooth the light field image and reduce high-frequency noise;

[0082]

[0083] I′(x,y): Pixel value of the filtered image;

[0084] I(u,v): Pixel value of the original image;

[0085] σ: Standard deviation of the Gaussian kernel, used to control the filtering intensity.

[0086] 2. Median filtering: Take the median of the pixel values within a local window to remove salt-and-pepper noise;

[0087] I′(x, y) = median{I(x + i, y + j)|i, j ∈ [-k, k]}

[0088] I′(x,y): Filtered pixel value;

[0089] k: Window radius.

[0090] S3-2. Image enhancement to improve the clarity of the tunnel face features by adjusting contrast and brightness;

[0091] Contrast enhancement: Use linear stretching method to expand the pixel value distribution range;

[0092]

[0093] I′(x,y): The enhanced pixel value;

[0094] I min ,I max : The minimum and maximum pixel values of the original image;

[0095] L min ,L max : The target brightness range.

[0096] Brightness adjustment: Automatically adjust the brightness according to the ambient light conditions;

[0097] I′(x,y) = I(x,y) + ΔL

[0098] ΔL: The brightness compensation value, dynamically calculated according to the ambient light intensity.

[0099] S3-3. Viewpoint alignment to spatially align the light field data of different viewpoints through an image registration algorithm;

[0100] 1. Feature point detection and matching:

[0101] Use the SIFT (Scale-Invariant Feature Transform) algorithm to detect key points in the image;

[0102] D(x,y,σ) = L(x,y,kσ) - L(x,y,σ)

[0103] D(x,y,σ): The Difference of Gaussian image;

[0104] L(x,y,σ): The Gaussian blur result of the image in the scale space σ.

[0105] Calculate the matching degree of feature points through the Euclidean distance:

[0106]

[0107] f i1 ,f i2 : The descriptors of two feature points.

[0108] 2. Geometric transformation and registration: Estimate the geometric transformation matrix (such as the affine transformation matrix) between images;

[0109]

[0110] x′, y′: Coordinates after registration;

[0111] a, b, c, d, e, f: Affine transformation parameters.

[0112] 3. Error optimization: Optimize the registration result by minimizing pixel differences;

[0113]

[0114] E: Registration error;

[0115] I 1 , I 2 : Gray values of two images.

[0116] In this way, through data denoising and image enhancement, the signal-to-noise ratio and clarity of the light field data can be significantly improved, providing high-quality input for subsequent processing. Moreover, the view alignment algorithm precisely fuses multi-view data, ensuring the consistency of the point cloud data during the 3D reconstruction process. At the same time, each algorithm has been optimized to support real-time processing of light field data, meeting the dynamic detection requirements of the tunnel construction site.

[0117] S4. Extract depth information from the processed light field data and generate a 3D point cloud;

[0118] When extracting depth information, a depth extraction algorithm based on ray rearrangement and disparity calculation is used to generate a high-precision depth map;

[0119] The ray rearrangement algorithm changes the position of the virtual focus plane to maximize the clarity of a specific depth area in the light field data, thereby deriving the depth value of that area. The disparity calculation algorithm uses the pixel offset relationship between images from different viewpoints in the light field data and combines camera parameters to calculate depth information. The extracted depth map, combined with the internal and external camera parameters, generates a 3D point cloud.

[0120] S5. Reconstruct the 3D point cloud to generate a 3D model;

[0121] The 3D point cloud reconstruction includes the following steps:

[0122] S5-1. Point cloud optimization, including noise point removal, data smoothing, and surface interpolation to fill holes;

[0123] Point cloud optimization is a basic step in 3D reconstruction, used to improve the quality, integrity, and smoothness of the point cloud data. This step includes the following key technologies:

[0124] 1. Noise point removal: Remove noise points and isolated points in the point cloud through a filtering algorithm to ensure the purity of the point cloud;

[0125] Statistical filtering formula:

[0126]

[0127] If d i > σ, then consider the point p i as a noise point and remove it;

[0128] d i : The average distance between the point p i and its neighborhood points;

[0129] k: The number of neighborhood points;

[0130] σ: The distance threshold, determined by experiment or preset.

[0131] Noise point removal includes:

[0132] Statistical filtering: Based on the average distance between each point and its neighborhood points, judge and remove the point cloud data points with excessive deviation;

[0133]

[0134] Where:

[0135] d i : The average distance between the point p i and its neighborhood points;

[0136] k: The number of neighborhood points;

[0137] σ: The set distance threshold.

[0138] Radius filtering: Retain the points in the point cloud whose number of neighborhood points meets certain requirements, and remove isolated points;

[0139] N(p i , r) ≥ N min

[0140] Where:

[0141] N(p i , r): The number of neighborhood points of the point p i within the radius r;

[0142] N min : The set minimum number of neighborhood points.

[0143] 2. Data smoothing processing: Reduce the roughness of the point cloud through a smoothing algorithm and enhance the surface smoothness;

[0144] Bilateral filtering formula:

[0145]

[0146] p′ i : The point cloud point after smoothing processing;

[0147] G: Spatial distance weight, Gaussian function;

[0148] G r : Eigenvalue weight, such as point cloud color or depth.

[0149] Moving Least Squares (MLS): Adjust the data based on the local fitting surface of points to make the point cloud smoother;

[0150] z = ax 2 + by 2 + cxy + dx + ey + f

[0151] Use the least squares method to solve the fitting parameters a, b, c, d, e, f and optimize the local surface.

[0152] 3. Surface interpolation to fill holes: For the data missing areas in the point cloud, use the interpolation algorithm to fill the holes and generate a continuous surface;

[0153] Surface interpolation formula:

[0154] z = ax 2 + by 2 + cxy + dx + ey + f

[0155] z: The height of the interpolation point;

[0156] a, b, c, d, e, f: Surface fitting parameters, calculated by the least squares method.

[0157] The interpolation method ensures the smoothness and continuity of the point cloud surface in the missing area, thus improving the integrity of the 3D model.

[0158] S5-2. Triangulation-based modeling, generating a complete 3D surface model through Delaunay triangulation;

[0159] After optimizing the point cloud data, generate a complete 3D surface model through triangulation, which specifically includes the following steps:

[0160] 1. Delaunay triangulation: Use the Delaunay triangulation algorithm to perform triangulation modeling on the point cloud data and generate a dense grid;

[0161] Delaunay condition: For any triangle, there are no other point cloud points inside its circumcircle:

[0162]

[0163] p i ,p j ,p m : Points that form the triangle;

[0164] p: The point to be tested.

[0165] The triangulation that satisfies the Delaunay condition can maximize the angles of triangles, avoid generating slender triangles, and thus ensure the uniformity of the mesh.

[0166] 2. Normal vector calculation: Calculate the normal vector for each triangle, which is used for subsequent lighting rendering and texture mapping.

[0167] Normal vector calculation formula:

[0168]

[0169] n: The unit normal vector of the triangle;

[0170] p 1 , p 2 , p 3 : The vertex coordinates of the triangle.

[0171] 3. Mesh simplification: To reduce the model complexity while maintaining details, use the mesh simplification algorithm to optimize the triangular mesh;

[0172] Error contraction formula (Quadric Error Metrics):

[0173] Q(v) = v T Qv

[0174] Q: The error matrix of the vertex;

[0175] v: The coordinates of the vertex.

[0176] Select the vertices to be removed through the error accumulation matrix to optimize the mesh structure.

[0177] 4. Texture mapping: Use the color data of the light field camera to perform texture mapping on the triangular mesh to restore the true appearance of the heading face;

[0178] Texture mapping formula:

[0179]

[0180] (u, v): Texture coordinates;

[0181] (x, y, z): The coordinates of the point cloud in three-dimensional space.

[0182] S6. Use the accurate position information provided by several ultrasonic sensors 6 to calibrate the accuracy of the three-dimensional model;

[0183] Calibrating the three-dimensional model through the ultrasonic sensor 6 includes making real-time adjustments and optimizations to the collected three-dimensional model data, and precisely compensating for equipment errors, deviations caused by movement, and environmental changes.

[0184] S7. Visualize the three-dimensional model through the computer 5;

[0185] The visualization of the three-dimensional model includes: dynamic visualization, which shows the three-dimensional form of the tunnel face through real-time rendering software, and analysis tools, including functions such as crack distribution, deformation measurement, and geological feature annotation. In this way, the collected and processed three-dimensional data can be presented to the user in an intuitive and dynamic manner, enabling the user to intuitively and quickly understand the overall form and local details of the tunnel face, and providing support for construction monitoring, analysis, and decision-making.

[0186] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A three-dimensional reconstruction device for a tunnel face based on light field photography technology, characterized in that: The device comprises a crawler robot (1) and a folding mechanical arm (2) arranged above the crawler robot (1); a support frame (3) is arranged at the top of the folding mechanical arm (2); The support frame (3) is provided with a plurality of light field cameras (4) for collecting light field data of different viewing angles of the tunnel face; and the support frame (3) is also provided with a plurality of ultrasonic sensors (6); The crawler robot (1) is provided with a computer (5), and the computer (5) is electrically connected to the crawler robot (1), the light field camera (4), and the ultrasonic sensor (6).

2. The device according to claim 1, characterized in that: The support frame (3) is hemispherical; the plurality of light field cameras (4) and the plurality of ultrasonic sensors (6) are evenly distributed in a ring shape on the inner side of the support frame (3).

3. The device according to claim 2, characterized in that: A plurality of lighting lamps (7) are arranged on the support frame (3), and the plurality of lighting lamps (7) are evenly distributed on the support frame (3) in a ring shape.

4. A method for three-dimensional reconstruction of a tunnel face based on light field photography technology, characterized in that: The method is to use any device of claims 1 to 3 to perform processing, and the matrix reconstruction method includes the following steps: S1, using a crawler robot (1) to move the device to a tunnel face detection area; S2, collecting light field data of different viewing angles of the tunnel face by using a plurality of light field cameras (4); S3, preprocessing the collected light field data through a computer (5); S4, extracting depth information from the processed light field data and generating a three-dimensional point cloud; S5, reconstructing the three-dimensional point cloud to generate a three-dimensional model; S6, using the precise position information provided by the plurality of ultrasonic sensors (6) to accurately calibrate the three-dimensional model; S7. Visually displaying the three-dimensional model through the computer (5).

5. The three-dimensional reconstruction method according to claim 4, characterized in that: The specific method for preprocessing the light field data in step S3 comprises the following steps: S3-1, data denoising: eliminate noise in light field data through filtering algorithms; S3-2, Image enhancement: Improve the clarity of face features by adjusting contrast and brightness; S3-3. Perspective alignment: The light field data from different perspectives are spatially aligned through the image registration algorithm.

6. The three-dimensional reconstruction method according to claim 4, characterized in that: The method of extracting depth information from the processed light field data in step S4 is to use a depth extraction algorithm based on light rearrangement and parallax calculation to generate a high-precision depth map.

7. The three-dimensional reconstruction method according to claim 4, characterized in that: The specific method of reconstructing the three-dimensional point cloud in step S5 includes the following steps: S5-1. Point cloud optimization: including noise point removal, data smoothing and surface interpolation and hole filling; S5-2. Triangulated mesh modeling: Generate a complete three-dimensional surface model through Delaunay triangulation.

8. The three-dimensional reconstruction method according to claim 7, characterized in that: The noise point removal in step S5-1 includes: Statistical filtering: Based on the mean distance between each point and its neighboring points, the point cloud data points with excessive deviation are eliminated; Radius filtering: retain points in the point cloud whose number of neighborhood points meets certain requirements and remove isolated points.

9. The three-dimensional reconstruction method according to claim 4, characterized in that: The specific operation of step S6 includes real-time adjustment and optimization of the collected three-dimensional model data, and accurate compensation for equipment errors, deviations caused by movement, and environmental changes.

10. The three-dimensional reconstruction method according to claim 4, characterized in that: The three-dimensional model visualization display in step S7 includes: S7-1. Dynamic visualization: Display the three-dimensional shape of the tunnel face through real-time rendering software; S7-2. Analysis tools: including crack distribution, deformation measurement and geological feature annotation functions.

Images