Underground engineering structure full-space deformation analysis method based on multi-depth camera array

Abstract

The invention discloses an underground engineering structure total space deformation analysis method based on a multi-depth camera array, and relates to the field of space deformation, and the method comprises the steps: employing the multi-depth camera array to collect a depth image and a color image of an underground engineering structure, and generating an alignment depth image of each depth camera; performing internal reference calibration on the depth camera, and converting an aligned depth image of the depth camera into three-dimensional color point cloud data of the depth camera; performing external parameter calibration on each depth camera to obtain an optimized transformation matrix from each slave camera to the master camera; transforming the three-dimensional color point cloud data of each subordinate camera to a coordinate system of the main camera by adopting an optimized transformation matrix to obtain a three-dimensional model of the underground engineering structure; and determining corresponding point pairs in the three-dimensional model obtained in different periods, and calculating the Euclidean distance between the corresponding point pairs to realize the full-space deformation analysis of the underground engineering structure. According to the invention, low-cost and high-precision underground engineering deformation monitoring is realized.

Description

Technical Field

[0001] This application relates to the field of spatial deformation, and in particular to a method for full-space deformation analysis of underground engineering structures based on a multi-depth camera array. Background Technology

[0002] Deformation monitoring in underground engineering is crucial for ensuring the safety and stability of the structure. Traditional deformation monitoring methods, such as total stations, convergence meters, and laser rangefinders, typically rely on the analysis of on-site cross-sectional data. While these methods offer acceptable accuracy, their efficiency is low, making them unsuitable for the real-time monitoring needs of large-scale underground projects. Three-dimensional laser scanning technology, capable of acquiring structural point cloud data in real-time and rapidly, has gradually become an important tool for deformation monitoring in underground engineering. However, this tool still faces several challenges in practical application. Firstly, the high point cloud density and large data volume place high demands on computer hardware and software for direct analysis and post-processing, raising the application threshold with high-performance and high-storage computers. However, consumer-grade depth cameras, capable of real-time data acquisition and less demanding in terms of lighting conditions, are well-suited for data acquisition in underground engineering, and multi-depth camera arrays can overcome the limitations of a single device's limited working range. In deformation monitoring, two-dimensional cross-sectional monitoring is a conventional method, but cross-sectional analysis only reflects local deformation and cannot comprehensively capture the overall deformation characteristics of the entire structure. In contrast, full-space deformation analysis can not only capture local deformation but also reflect the overall structural deformation characteristics, providing comprehensive data support for engineering safety. Summary of the Invention

[0003] The purpose of this application is to provide a method for full-space deformation analysis of underground engineering structures based on a multi-depth camera array. This method enables accurate 3D reconstruction of underground engineering structures using a multi-depth camera array and allows for full-space deformation analysis of the underground engineering structures under a high-precision 3D model, thereby achieving low-cost, high-precision deformation monitoring of underground engineering projects. To achieve the above objective, this application provides the following solution: Firstly, this application provides a method for full-space deformation analysis of underground engineering structures based on a multi-depth camera array, including: A multi-depth camera array is used to acquire depth and color images of underground engineering structures, and the depth images and color images are aligned to generate an aligned depth image for each depth camera; the multi-depth camera array includes one master camera and multiple slave cameras; For any depth camera, perform intrinsic parameter calibration. Based on the calibrated depth camera, convert the aligned depth image of the depth camera into three-dimensional color point cloud data in the coordinate system of the depth camera itself. For each depth camera, extrinsic parameters are calibrated to obtain the optimized transformation matrix from each slave camera to the master camera; An optimized transformation matrix from each slave camera to the main camera is used to transform the 3D color point cloud data in the coordinate system of each slave camera to the coordinate system of the main camera, and point cloud fusion is performed to obtain the 3D model of the underground engineering structure. Based on the normal vector information of point cloud data, corresponding point pairs in the three-dimensional model acquired at different times are determined, and the full-space deformation analysis of the underground engineering structure is realized by calculating the Euclidean distance between the corresponding points.

[0004] According to the specific embodiments provided in this application, this application has the following technical effects: This application provides a method for full-space deformation analysis of underground engineering structures based on a multi-depth camera array. By using a multi-depth camera array, depth and color images of the underground engineering structure are acquired from multiple angles and orientations, laying the foundation for obtaining a clear and concise 3D model of the underground engineering structure. Furthermore, multi-consumer-grade depth cameras are less expensive than 3D laser scanners, and the multi-camera array solves the problem of insufficient working range of a single camera, making it more suitable for the illumination conditions of underground engineering sites than ordinary high-resolution cameras. In addition, the 3D color point cloud data from each subordinate camera's own coordinate system is transformed to the coordinate system of the main camera, facilitating subsequent fusion, processing, and analysis. Point cloud fusion from the main viewpoint can cover a wider range of scenes, reduce occlusion, and improve the integrity of the 3D reconstruction. The method for accurately locating point pairs corresponding to point clouds at different times enables full-space deformation analysis of underground engineering structures. Attached Figure Description

[0005] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0006] Figure 1 A flowchart illustrating a method for full-space deformation analysis of underground engineering structures based on a multi-depth camera array, provided as an embodiment of this application; Figure 2 This is a schematic diagram illustrating the process of acquiring three-dimensional color point cloud data for a method of full-space deformation analysis of underground engineering structures based on a multi-depth camera array. Figure 3(1) is a schematic diagram of the horizontal stitching of the three-dimensional reconstruction of the underground engineering structure based on a multi-depth camera array for full-space deformation analysis; Figure 3(2) is a schematic diagram of the vertical stitching after the horizontal stitching of the three-dimensional reconstruction of the underground engineering structure. Figure 4 This is a flowchart of a full-space deformation analysis method for underground engineering structures based on a multi-depth camera array. Figure 5 The results of full-space deformation analysis of underground engineering structures are presented in a method for full-space deformation analysis of underground engineering structures based on a multi-depth camera array. Figure 6 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0007] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0008] To make the objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0009] In one exemplary embodiment, such as Figure 1 As shown, a method for full-space deformation analysis of underground engineering structures based on a multi-depth camera array is provided, which includes the following steps 101 to 105: Step 101 involves acquiring depth and color images of the underground engineering structure using a multi-depth camera array, aligning the depth and color images to generate an aligned depth image for each depth camera. The multi-depth camera array includes one master camera and multiple slave cameras. The consumer-grade depth camera used in this application is Azure Kinect, and the multi-depth camera array is connected via a star or daisy-chain connection, using a 3mm audio cable for data synchronization between devices.

[0010] Furthermore, during multi-depth camera array acquisition, to prevent interference between infrared lasers, the offset time for each camera was set to 160µs. The Software Development Kit (SDK) was used to provide the parameters `depth_delay_off_color_usec` or `subsidiary_delay_of_master_usec` to ensure that each infrared laser triggered within the 160µs limit. Then, the cameras were opened sequentially using `k4a_device_open` and `k4a_device_start_cameras` from the SDK, starting with the slave cameras and then the master camera. Depth and color images of the underground engineering structure were captured, and `k4a_transformation_Depth_image_to_clor_camera` was used to align the depth and color images, generating an aligned depth image for each depth camera.

[0011] Here, depth_delay_off_color_usec is the capture time delay of the depth camera relative to the color camera, subsidiary_delay_of_master_usec is the delay time (in microseconds) of the slave camera relative to the master camera, k4a_device_open is to open the Kinect device, k4a_device_start_cameras is to start the camera, and k4a_transformation_Depth_image_to_clor_camera is to transform the depth image to the color camera coordinate system.

[0012] In addition, appropriate operating modes for the depth and color cameras need to be selected. The WFOV mode for the depth camera corresponds to a depth image resolution of 1024×1024; the RGB camera corresponds to a resolution of 4096×3072. Initially, the devices are warmed up, with each device collecting 100 frames beforehand to stabilize performance at the current room temperature. In WFOV mode, a fisheye camera is used, which causes distortion in the depth image at the edges, resulting in errors in the depth values. Therefore, a depth image within a 600×600 pixel area is used for point cloud conversion.

[0013] Step 102: Perform intrinsic parameter calibration on any depth camera. Based on the calibrated depth camera, convert the aligned depth image of the depth camera into 3D color point cloud data in the depth camera's own coordinate system. For example... Figure 2The diagram illustrates the process of acquiring 3D color point cloud data. The Azure Kinect SDK integrates the OpenCV library for parameter calibration of a single depth camera. First, the calibration board is photographed from different angles using the Azure Kinect color camera, resulting in a total of 30 images. These images are then imported into the SDK for single-camera calibration, using OpenCV's corner detection to obtain the intrinsic parameters and distortion coefficients of the single depth camera. Azure Kinect integrates both a depth camera and a color camera.

[0014] Step 103: Perform extrinsic parameter calibration on each depth camera to obtain the optimized transformation matrix from each slave camera to the master camera.

[0015] Step 104: Using the optimized transformation matrix from each slave camera to the main camera, the 3D color point cloud data in the coordinate system of each slave camera is transformed to the coordinate system of the main camera, and point cloud fusion is performed to obtain the 3D model of the underground engineering structure. The underground engineering structure in this application is a tunnel. Four Azure Kinect arrays are used for tunnel reconstruction. The optimized transformation matrix of the four cameras is used for point cloud fusion. The horizontal stitching forms a tunnel ring structure, as shown in Figure 3(1). In the longitudinal direction of the tunnel, artificially set feature points are used to assist the Iterative Closest Point (ICP) algorithm to stitch adjacent ring point cloud segments vertically, as shown in Figure 3(2). Among them, artificially set features include landmarks, targets, specific patterns, etc.

[0016] Step 105: Based on the normal vector information of the point cloud data, determine the corresponding point pairs in the three-dimensional model acquired at different times, and realize the full-space deformation analysis of the underground engineering structure by calculating the Euclidean distance between the corresponding points.

[0017] By implementing steps 101 to 105 above, this application has reconstructed the underground engineering structure with high accuracy and conducted full-space deformation analysis based on the accurate three-dimensional point cloud model.

[0018] In another exemplary embodiment of this application, step 103 is replaced by steps 201-202: Step 201: Using a common-view calibration board, perform extrinsic parameter calibration on each depth camera to obtain the initial transformation matrix of each slave camera relative to the master camera.

[0019] Step 202: Using the actual scene 3D color point cloud data collected by each depth camera, and based on the geometric and color features of the actual scene 3D color point cloud data, optimize the initial transformation matrix of each slave camera relative to the master camera to obtain the optimized transformation matrix of each slave camera to the master camera.

[0020] In another exemplary embodiment of this application, step 201 is replaced by the following steps 301 to 303: Step 301: For any slave camera, use the slave camera and the master camera to respectively acquire color images of the calibration plate in the common viewing area.

[0021] Step 302: Based on the calibration board color image, the intrinsic parameters of the slave camera, the intrinsic parameters of the master camera, and the size parameters of the calibration board, the relative poses of the calibration board and the slave camera, as well as the relative poses of the calibration board and the master camera, are obtained using OpenCV functions and the PnP algorithm.

[0022] Step 303: Based on the relative poses of the calibration board and the slave camera and the relative poses of the calibration board and the master camera, obtain the initial transformation matrix of the slave camera relative to the master camera.

[0023] By using the pairwise relative positions of any slave camera and the master camera, the data acquired by any slave camera in its own coordinate system is ultimately unified into the coordinate system of the master camera. First, the calibration board is photographed with either slave camera or the master camera from an overlapping viewpoint. Then, the color images of the calibration board acquired by the two cameras, the YAML files of the intrinsic parameters of the two cameras, and the JSON file of the calibration board are imported into the SDK for dual-camera calibration. The initial transformation matrix of any slave camera relative to the master camera is obtained through the OpenCV functions integrated in the SDK.

[0024] In another exemplary embodiment of this application, step 201 can only obtain the initial transformation matrix between cameras, which still has a certain error compared to accurate transformation matrix estimation. Step 202 specifically includes the following steps 401 to 408: Step 401: For any slave camera, transform the actual scene 3D color point cloud data collected by the slave camera in its own coordinate system to the coordinate system of the main camera using the initial transformation matrix of the slave camera relative to the main camera.

[0025] Step 402: In the coordinate system of the main camera, for any point cloud in the actual scene 3D color point cloud data collected by the slave camera and the actual scene 3D color point cloud data collected by the main camera, determine the flatness coefficient and the color change coefficient of the point cloud.

[0026] Step 403: Based on a set threshold and the flatness coefficient of each point cloud, filter the actual scene 3D color point cloud data collected by the slave camera and the actual scene 3D color point cloud data collected by the main camera to obtain the feature points of the slave camera and the main camera. When the flatness coefficient of any point is greater than the set threshold, the point is a feature point, indicating that it has some undulations compared to the surrounding point set.

[0027] Step 404: For any feature point among the feature points of the slave camera and the main camera, obtain the descriptor of the feature point based on the FPFH descriptor and the color change coefficient of the feature point. For any feature point, construct an FPFH descriptor to characterize the spatial geometric features of the point, and add a color change coefficient b to characterize the color change, constructing a descriptor. Then, the descriptor of the m-th feature point... for: .in, Let f(m) be the FPFH descriptor for the m-th feature point. Let be the color change coefficient of the m-th feature point.

[0028] Step 405: Calculate the similarity between the descriptors of the feature points of the slave camera and the descriptors of the feature points of the master camera, and determine multiple matching point pairs based on the similarity.

[0029] Step 406: Using SVD to solve the pose relationship based on the spatial coordinates of any matching point pair, the rotation matrix and translation vector of the matching point pair are obtained.

[0030] Step 407: Determine the root mean square error of the matching point pair based on the rotation matrix and translation vector of the matching point pair.

[0031] Step 408: Determine the minimum root mean square error based on the root mean square error of all matching point pairs, and determine the optimized transformation matrix from the slave camera to the master camera based on the rotation matrix corresponding to the minimum root mean square error.

[0032] In another exemplary embodiment of this application, step 402, determining the flatness coefficient and the color change coefficient of the point cloud, specifically includes the following steps 501 to 503: Step 501: Determine the k nearest neighbor points of the point cloud to construct the neighborhood point set of the point cloud. ,in, For the neighborhood point set, For the first The nearest neighbor point, The first nearest neighbor. It is the second nearest neighbor.

[0033] Step 502: Calculate the centroid of the neighborhood point set, and determine the first covariance matrix based on the centroid and the k nearest neighbors. Perform eigenvalue decomposition on the first covariance matrix to obtain multiple eigenvalues, and determine the flatness coefficient of the point cloud based on the multiple eigenvalues. The flatness coefficient is obtained based on the ratio of the sum of the multiple eigenvalues ​​to the smallest eigenvalue among the multiple eigenvalues.

[0034] Step 503: Construct a second covariance matrix based on the color values ​​of the k nearest neighbors. Perform eigenvalue decomposition on the second covariance matrix to obtain multiple eigenvalues. Determine the color change coefficients of the point cloud based on these multiple eigenvalues. The color values ​​of the nearest neighbors are RGB information.

[0035] In another exemplary embodiment of this application, in step 502, the first covariance matrix is ​​determined using the following formula: .

[0036] .

[0037] .

[0038] in, Let the centroid of the neighborhood point set be . The number of nearest neighbors. The index of the nearest neighbor. For the first The nearest neighbor point, For variables and variables Covariance between , , Let i be the coordinate components of the i-th nearest neighbor. , and The coordinate components of the centroid, Let be the first covariance matrix. For variables and variables Covariance between For variables and variables Covariance between For variables and variables Covariance between For variables and variables Covariance between For variables and variables The covariance between them.

[0039] In another exemplary embodiment of this application, in step 405, the number of slave cameras is calculated using the following formula. The descriptor of the first feature point and the first feature point of the main camera Similarity of descriptors for each feature point: .

[0040] in, For the slave camera's first The descriptor of the first feature point and the first feature point of the main camera Similarity of descriptors for each feature point The dimension of the FPFH descriptor. For the first The FPFH descriptor of the nth feature point dimension, For the first The FPFH descriptor of the nth feature point dimension, For the first Color change coefficient of each feature point For the first Color change coefficient of each feature point As the first weighting factor, The first and second weighting factors are used to control the importance of geometric and color information in the descriptor. The first weighting factor is 0.5, and the second weighting factor is 0.5.

[0041] In another exemplary embodiment of this application, the different periods include a first period and a second period, wherein the point cloud data corresponding to the 3D model acquired in the first period is used as the source point cloud data, and the point cloud data corresponding to the 3D model acquired in the second period is used as the target point cloud data. Figure 4 As shown, step 105 specifically includes the following steps 601 to 606: Step 601: Downsample the source point cloud data to obtain downsampled point cloud data.

[0042] The source point cloud data is traversed using a spatial sampling window, with half the length of the spatial sampling window used as the sampling step size, based on feature points. Replace all points within the window. Feature points. The calculation formula is as follows: .

[0043] .

[0044] .

[0045] in, The three-dimensional coordinates of the feature point Let x be the x-coordinate of the u-th point in the spatial sampling window. Let be the y-coordinate of the u-th point in the spatial sampling window, and E be the total number of points in the spatial sampling window. The first sampling coefficient, The second sampling coefficient, The third sampling coefficient, The fourth sampling coefficient, The values ​​are constant. Downsampling was performed using the x and y coordinates of the centroids of all points within the spatial sampling window and their Z-axis coordinates on the quadratic surface that satisfies the fit of all points as feature points.

[0046] Step 602: For any point in the downsampled point cloud data, determine the normal vector of the point; based on the normal vector and the three-dimensional coordinates of the point, establish the spatial straight line equation of the point: ;in, These are the three-dimensional coordinates of the corresponding points on a straight line in space. The three-dimensional coordinates of the point are: Let be the normal vector of the point. These are intermediate parameters.

[0047] Step 603: Calculate the distance of each point in the target point cloud data to the spatial straight line from the point, and filter the target point cloud data according to the distance and a set number to obtain multiple filtered point clouds.

[0048] Step 604: Establish a spatial plane equation based on the selected point clouds, obtain intermediate parameters based on the spatial line equation of the points and the spatial plane equation, and determine the three-dimensional coordinates of the corresponding points on the spatial line based on the intermediate parameters.

[0049] Step 605: Determine the Euclidean distance between the corresponding point on the spatial line and the point based on the three-dimensional coordinates of the corresponding point on the spatial line and the three-dimensional coordinates of the point.

[0050] Step 606: Use the Euclidean distance as the spatial deformation value of the point, and perform full-space deformation analysis of the underground engineering structure based on the spatial deformation value of each point in the downsampled point cloud data.

[0051] In another exemplary embodiment of this application, in step 603, the first point in the target point cloud data is calculated using the following formula. The distance from each point to the straight line in space: .

[0052] .

[0053] .

[0054] .

[0055] .

[0056] in, For the target point cloud data, the first The distance between each point and the spatial straight line from the given point. The parameters of the first spatial line are... For the parameters of the second space line, For the parameters of the line in the third space, For the target point cloud data, the first The three-dimensional coordinates of the points The index of a point in the target point cloud data. These are the parameters of the line in the fourth space.

[0057] In another exemplary embodiment of this application, in step 604, the spatial plane equation is established using the following formula: .

[0058] .

[0059] .

[0060] .

[0061] .

[0062] in, For the parameters of the first spatial plane, For the second spatial plane parameters, For the second spatial plane parameters, For the parameters of the first spatial plane, The 3D coordinates of the first point cloud after filtering. The three-dimensional coordinates of the second point cloud after filtering. The three-dimensional coordinates of the third point cloud after filtering.

[0063] In step 604, the intermediate parameter t is determined using the following formula: .

[0064] In one exemplary embodiment, such as Figure 5 As shown, this application demonstrates deformation testing of an underground tunnel structure. First, data is acquired and fused using a multi-depth camera. Then, deformation label boxes are attached as the actual deformation values, with box heights of 20mm, 160mm, and 200mm. Full-space deformation analysis is then performed. Statistical analysis shows that the deformation error for a 20mm deformation can converge to 2mm, with a minimum deformation error of only 0.4mm. For larger deformations of 160mm and 200mm, the error results are concentrated and converged within 20mm, with a minimum deformation deviation of only 3mm. The deformation measurement results demonstrate the accuracy of the method applied in this application for full-space deformation analysis.

[0065] This application utilizes a consumer-grade depth camera to reduce hardware costs and employs a multi-camera array to overcome the limitations of a single camera's working range. Furthermore, the calibration process yields a camera transformation matrix, avoiding the cumulative errors from stitching together multiple point clouds. High-precision reconstruction of underground engineering structures was achieved, and full-space deformation analysis was performed based on the accurate 3D point cloud model.

[0066] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 6 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media. The database stores depth and color images. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When executed by the processor, the computer program implements a method for full-space deformation analysis of underground engineering structures based on a multi-depth camera array.

[0067] Those skilled in the art will understand that Figure 6 The structures shown are merely block diagrams of some structures related to the present application and do not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0068] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0069] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for full-space deformation analysis of underground engineering structures based on multi-depth camera arrays, characterized in that, The method includes: A multi-depth camera array is used to acquire depth and color images of underground engineering structures, and the depth images and color images are aligned to generate an aligned depth image for each depth camera; the multi-depth camera array includes one master camera and multiple slave cameras; For any depth camera, perform intrinsic parameter calibration. Based on the calibrated depth camera, convert the aligned depth image of the depth camera into three-dimensional color point cloud data in the coordinate system of the depth camera itself. For each depth camera, extrinsic parameters are calibrated to obtain the optimized transformation matrix from each slave camera to the master camera; An optimized transformation matrix from each slave camera to the main camera is used to transform the 3D color point cloud data in the coordinate system of each slave camera to the coordinate system of the main camera, and point cloud fusion is performed to obtain the 3D model of the underground engineering structure. Based on the normal vector information of point cloud data, corresponding point pairs in the three-dimensional model acquired at different times are determined, and the full-space deformation analysis of the underground engineering structure is realized by calculating the Euclidean distance between the corresponding points.

2. The method for full-space deformation analysis of underground engineering structures based on multi-depth camera arrays according to claim 1, characterized in that, For each depth camera, extrinsic parameter calibration is performed to obtain the optimized transformation matrix from each slave camera to the master camera, specifically including: Using a common-view calibration board, the extrinsic parameters of each depth camera are calibrated to obtain the initial transformation matrix of each slave camera relative to the master camera. Using the actual scene 3D color point cloud data collected by each depth camera, and based on the geometric and color features of the actual scene 3D color point cloud data, the initial transformation matrix of each slave camera relative to the master camera is optimized to obtain the optimized transformation matrix of each slave camera to the master camera.

3. The method for full-space deformation analysis of underground engineering structures based on multi-depth camera arrays according to claim 2, characterized in that, Using a common-view calibration board, extrinsic parameters are calibrated for each depth camera to obtain the initial transformation matrix of each slave camera relative to the master camera, specifically including: For any slave camera, the slave camera and the master camera respectively acquire color images of the calibration board in the common field of view area; Based on the calibration board color image, the intrinsic parameters of the slave camera, the intrinsic parameters of the master camera, and the size parameters of the calibration board, the relative poses of the calibration board and the slave camera, as well as the relative poses of the calibration board and the master camera, are obtained using OpenCV functions and the PnP algorithm. Based on the relative poses of the calibration board and the slave camera, and the relative poses of the calibration board and the master camera, the initial transformation matrix of the slave camera relative to the master camera is obtained.

4. The method for full-space deformation analysis of underground engineering structures based on multi-depth camera arrays according to claim 2, characterized in that, Using real-world 3D color point cloud data acquired by each depth camera, and based on the geometric and color features of the real-world 3D color point cloud data, the initial transformation matrix of each slave camera relative to the master camera is optimized to obtain the optimized transformation matrix from each slave camera to the master camera. Specifically, this includes: For any slave camera, the actual scene 3D color point cloud data collected by the slave camera in its own coordinate system is transformed to the coordinate system of the main camera through the initial transformation matrix of the slave camera relative to the main camera. In the coordinate system of the main camera, for any point cloud in the actual scene 3D color point cloud data collected by the slave camera and the actual scene 3D color point cloud data collected by the main camera, determine the flatness coefficient and the color change coefficient of the point cloud. Based on the set threshold and the flatness coefficient of each point cloud, the actual scene 3D color point cloud data collected by the slave camera and the actual scene 3D color point cloud data collected by the main camera are filtered to obtain the feature points of the slave camera and the feature points of the main camera. For any feature point among the feature points of the slave camera and the feature points of the master camera, a descriptor for the feature point is obtained based on the FPFH descriptor and the color change coefficient of the feature point. Calculate the similarity between the descriptors of the feature points of the slave camera and the descriptors of the feature points of the master camera, and determine multiple matching point pairs based on the similarity. The pose relationship is solved using SVD based on the spatial coordinates of any matching point pair, and the rotation matrix and translation vector of the matching point pair are obtained. The root mean square error of the matching point pair is determined based on the rotation matrix and translation vector of the matching point pair. Based on the root mean square error of all matching point pairs, determine the minimum root mean square error, and based on the rotation matrix corresponding to the minimum root mean square error, determine the optimized transformation matrix from the slave camera to the master camera.

5. The method for full-space deformation analysis of underground engineering structures based on multi-depth camera arrays according to claim 4, characterized in that, Determining the flatness coefficient and color variation coefficient of the point cloud specifically includes: Determine the k nearest neighbors of the point cloud to construct a neighborhood point set of the point cloud; Calculate the centroid of the neighborhood point set, and based on the centroid and the k nearest neighbors, determine the first covariance matrix and perform eigenvalue decomposition on the first covariance matrix to obtain multiple eigenvalues. Determine the flatness coefficient of the point cloud based on the multiple eigenvalues; wherein the flatness coefficient is obtained based on the ratio of the sum of the multiple eigenvalues ​​to the smallest eigenvalue among the multiple eigenvalues. Based on the color values ​​of the k nearest neighbors, a second covariance matrix is ​​constructed. The second covariance matrix is ​​then subjected to eigenvalue decomposition to obtain multiple eigenvalues. The color change coefficients of the point cloud are determined based on these multiple eigenvalues.

6. The method for full-space deformation analysis of underground engineering structures based on multi-depth camera arrays according to claim 5, characterized in that, The first covariance matrix is ​​determined using the following formula: ； ； ； in, Let the centroid of the neighborhood point set be . The number of nearest neighbors. The index of the nearest neighbor. For the first The nearest neighbor point, For variables and variables Covariance between , , Let i be the coordinate components of the i-th nearest neighbor. , and The coordinate components of the centroid, Let be the first covariance matrix. For variables and variables Covariance between For variables and variables Covariance between For variables and variables Covariance between For variables and variables Covariance between For variables and variables The covariance between them.

7. The method for full-space deformation analysis of underground engineering structures based on multi-depth camera arrays according to claim 5, characterized in that, The slave camera's first... The descriptor of the first feature point and the first feature point of the main camera Similarity of descriptors for each feature point: ； in, The slave camera's first The descriptor of the first feature point and the first feature point of the main camera Similarity of descriptors for each feature point The dimension of the FPFH descriptor. For the first The FPFH descriptor of the nth feature point dimension, For the first The FPFH descriptor of the nth feature point dimension, For the first Color change coefficient of each feature point For the first Color change coefficient of each feature point As the first weighting factor, This is the second weighting factor.

8. The method for full-space deformation analysis of underground engineering structures based on multi-depth camera arrays according to claim 1, characterized in that, The different periods include a first period and a second period. The point cloud data corresponding to the three-dimensional model obtained in the first period is used as the source point cloud data, and the point cloud data corresponding to the three-dimensional model obtained in the second period is used as the target point cloud data. Based on the normal vector information of point cloud data, corresponding point pairs in the 3D model acquired at different times are determined. By calculating the Euclidean distance between corresponding point pairs, the full-space deformation analysis of the underground engineering structure is realized, specifically including: Downsampling is performed on the source point cloud data to obtain downsampled point cloud data; For any point in the downsampled point cloud data, determine the normal vector of the point, and based on the normal vector and the three-dimensional coordinates of the point, establish the spatial straight line equation of the point: ;in, These are the three-dimensional coordinates of the corresponding points on a straight line in space. The three-dimensional coordinates of the point are: Let be the normal vector of the point. For intermediate parameters; Calculate the distance from each point in the target point cloud data to the spatial straight line from the point, and filter the target point cloud data according to the distance and a set number to obtain multiple filtered point clouds; A spatial plane equation is established based on the selected point clouds. Intermediate parameters are obtained based on the spatial line equations of the points and the spatial plane equations. The three-dimensional coordinates of the corresponding points on the spatial line are determined based on the intermediate parameters. Determine the Euclidean distance between the corresponding point on the spatial line and the point based on the three-dimensional coordinates of the point on the spatial line and the three-dimensional coordinates of the point. Using the Euclidean distance as the spatial deformation value of the point, and based on the spatial deformation value of each point in the downsampled point cloud data, the full-space deformation analysis of the underground engineering structure is realized.

9. The method for full-space deformation analysis of underground engineering structures based on multi-depth camera arrays according to claim 8, characterized in that, The following formula is used to calculate the first point in the target point cloud data. The distance from each point to the straight line in space: ； ； ； ； ； in, For the target point cloud data, the first The distance between each point and the straight line in space from the given point. The parameters of the first spatial line are... For the parameters of the second space line, For the parameters of the third space line, For the target point cloud data, the first The three-dimensional coordinates of the points The index of a point in the target point cloud data. These are the parameters of the line in the fourth space.

10. The method for full-space deformation analysis of underground engineering structures based on multi-depth camera arrays according to claim 8, characterized in that, The equation of the space plane is established using the following formula: ； ； ； ； ； in, For the parameters of the first spatial plane, For the second spatial plane parameters, For the second spatial plane parameters, For the parameters of the first spatial plane, The 3D coordinates of the first point cloud after filtering. The 3D coordinates of the second point cloud after filtering. The three-dimensional coordinates of the third point cloud after filtering.

Images