A Monocular Polarization Three-Dimensional Imaging Method for Monitoring Deformation of Surrounding Rock in Tunnels

By using monocular polarization 3D imaging technology and a split-focus plane polarization camera to acquire and process images of tunnel surrounding rock, the problems of large-format measurement and construction interference in tunnel surrounding rock deformation monitoring have been solved, and efficient and accurate 3D deformation measurement has been achieved.

CN116182725BActive Publication Date: 2026-11-10BEIJING RES INST OF SPATIAL MECHANICAL & ELECTRICAL TECH
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202211658757.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-12-22
Publication Date
2026-11-10
Estimated Expiration
2042-12-22

AI Technical Summary

Technical Problem

Existing tunnel surrounding rock deformation monitoring technologies suffer from problems such as difficulty in large-format measurements, errors introduced by multiple station setups, significant construction interference, and low accuracy, making it difficult to achieve efficient and accurate three-dimensional deformation measurement.

Method used

The monocular polarization 3D imaging method is adopted. Two phases of tunnel surrounding rock images are acquired by a split-focus plane polarization camera. Preprocessing, stereo matching and monocular polarization 3D reconstruction are performed to calculate the elevation difference to obtain the deformation of the tunnel surrounding rock, avoiding multiple station setups and equipment location requirements.

Benefits of technology

It enables large-format, non-contact, long-distance three-dimensional deformation measurement of tunnel surrounding rock, simplifies operation, improves measurement efficiency and accuracy, reduces construction interference, and obtains real information on tunnel surrounding rock deformation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116182725B_ABST
    Figure CN116182725B_ABST
Patent Text Reader

Abstract

This invention proposes a monocular polarization three-dimensional imaging method for monitoring tunnel surrounding rock deformation, comprising: acquiring two phases of focused planar polarized images of the tunnel surrounding rock in the monitoring area; preprocessing the two phases of focused planar polarized images of the tunnel surrounding rock in the monitoring area; performing stereo matching on the preprocessed two phases of focused planar polarized images of the tunnel surrounding rock in the monitoring area to obtain two phases of three-dimensional models of the tunnel surrounding rock; calculating the elevation difference at each coordinate position based on the obtained three-dimensional models of the tunnel surrounding rock being in the same coordinate system, and obtaining the elevation change at that coordinate position; calibrating the elevation change, and calculating the true value of the tunnel surrounding rock deformation. This invention can achieve large-format simultaneous measurement of the three-dimensional deformation of the tunnel surrounding rock without the need to set up multiple control points, avoiding errors introduced by multiple station setups, and the measurement process is simple and efficient, greatly reducing the impact on construction.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of tunnel surrounding rock deformation monitoring technology, and relates to a monocular polarization three-dimensional imaging method for monitoring tunnel surrounding rock deformation. Background Technology

[0002] The Sichuan-Tibet Railway traverses a region characterized by dramatic topographical variations, complex and variable lithology, widespread deep and active faults, exceptionally active internal and external geological processes, and strong coupling effects from multiple environmental fields, resulting in extremely high engineering difficulty and risk. Monitoring the deformation of deep, high-frequency, and large-scale surrounding rock is a crucial aspect of underground engineering construction. Developing and improving existing monitoring instruments to obtain information most closely resembling actual surrounding rock deformation is a primary concern for engineers.

[0003] Displacement monitoring instruments mainly include traditional convergence meters (commonly used are belt-type and rod-type convergence meters), displacement meters (such as Sinco displacement meters and wire-type displacement meters), and newer monitoring equipment such as total stations, tunnel profile meters, infrared rangefinders, and laser rangefinders. In engineering practice, traditional measuring instruments are widely used, but convergence meters can only measure the convergence value of the tunnel wall and cannot reflect the displacement law of the deep surrounding rock. Among displacement meters, sensor-type displacement meters are expensive and have poor stability, making them unsuitable for long-term monitoring and large-scale use. Mechanical displacement meters mainly use rod-type and wire (cable) multi-point displacement meters. Rod-type displacement meters are prone to problems such as full-hole solidification, which can alter the mechanical properties of the surrounding rock and cause interference between displacements at different depths. Wire (cable) displacement meters are affected by hole collapse and extrusion deformation, leading to distorted measurement results. Total stations are electronic versions of theodolites, integrating horizontal angle, vertical angle, distance (slope distance, horizontal distance), and elevation difference measurement functions. They are widely used in precision engineering surveying or deformation monitoring fields such as large-scale above-ground construction and underground tunnel construction. However, total stations perform single-point mapping, acquiring only single-point coordinates, resulting in low work efficiency and some interference with construction. Furthermore, the use of total stations requires data measurement based on the experience of on-site verification personnel, placing high demands on operator skill and limiting their application. Newer monitoring equipment such as infrared rangefinders and laser rangefinders have not achieved widespread use due to their high cost, significant impact on construction, and high requirements for cleanliness and operator skill. Therefore, there is an urgent need to develop a large-format, non-contact, long-distance, and easy-to-operate measurement method to rapidly acquire large-area, high-precision, and rapid 3D models of tunnel surrounding rock.

[0004] Currently, existing domestic literature on monocular polarization 3D imaging technology mainly focuses on theoretical research and experimental verification for simple targets in the laboratory. However, it has problems in reconstructing complex targets and scenes and cannot obtain the true value of the target. Existing domestic patents for tunnel surrounding rock deformation monitoring use wireless communication technology, which requires the deployment of sensor nodes according to the construction situation, and has a certain impact on the construction. Summary of the Invention

[0005] The technical problem solved by this invention is to overcome the shortcomings of the prior art and propose a monocular polarization three-dimensional imaging method for monitoring the deformation of tunnel surrounding rock. This method can achieve large-format simultaneous measurement of the three-dimensional deformation of tunnel surrounding rock without the need to set up multiple control points, thus avoiding errors introduced by multiple station setups. The measurement process is simple and efficient, greatly reducing the impact on construction.

[0006] The technical solution of this invention is: a method for monitoring tunnel surrounding rock deformation using monocular polarization three-dimensional imaging, comprising:

[0007] Two phases of focused plane polarization images of the tunnel surrounding rock were acquired in the monitoring area. The focused plane polarization images of the tunnel surrounding rock in each phase of the monitoring area were obtained by the same focused plane polarization camera.

[0008] Preprocessing was performed on the focal plane polarization images of the surrounding rock of the tunnel in the two monitoring areas, including polarization interpolation, image filtering, histogram equalization, and histogram matching.

[0009] Stereo matching was performed on the preprocessed two-phase monitoring area tunnel surrounding rock focal plane polarization images. The second phase focal plane polarization image was projected onto the coordinate system of the first phase focal plane polarization image to obtain the projected image of the second phase polarization image.

[0010] Monocular polarization 3D reconstruction was performed on the first-phase polarization image and the second-phase polarization image after projection to obtain two-phase tunnel surrounding rock 3D models.

[0011] The three-dimensional model of the tunnel surrounding rock obtained from the first phase of polarized images and the three-dimensional model of the tunnel surrounding rock obtained from the projected image of the second phase of polarized images are in the same coordinate system. The elevation difference at each coordinate position is calculated to obtain the elevation change at that coordinate position. The elevation change is calibrated to calculate the true value of the deformation of the tunnel surrounding rock.

[0012] The placement of the two phases of the focal plane polarization camera does not need to be exactly the same, but it must be ensured that the two measurements cover the same deformation area.

[0013] Stereo matching is performed on the preprocessed two-phase monitoring area tunnel surrounding rock focal plane polarization images, including: taking two overlapping photos; firstly, capturing many feature points based on the principle that feature points have local differences; secondly, optimizing the feature points; thirdly, describing the optimized key feature points; and finally finding an optimal matrix to describe the correspondence between feature points in the two images, thereby determining the relative position of the two images.

[0014] Monocular polarization 3D reconstruction includes:

[0015] First, the polarization degree and polarization angle information of the tunnel surrounding rock surface are obtained using a single polarization image. Second, the zenith angle of the tunnel surrounding rock surface is calculated from the polarization degree information, and the azimuth angle of the tunnel surrounding rock surface is calculated from the polarization angle information. Then, the normal vector of the tunnel surrounding rock surface is represented by the zenith angle and the azimuth angle. Finally, the three-dimensional model of the tunnel surrounding rock is obtained by integrating the normal vector.

[0016] The advantages of this invention compared to the prior art are:

[0017] Monitoring the deformation of deep, high-frequency, and large-scale surrounding rock is a crucial aspect of underground engineering construction. Developing and improving existing monitoring instruments to obtain information most closely resembling actual rock deformation is a primary concern for engineers. There is an urgent need to develop a large-format, non-contact, long-distance, and easy-to-operate measurement method to rapidly acquire large-area, high-precision, and rapid 3D models of tunnel surrounding rock. This invention overcomes the shortcomings of existing technologies by providing a monocular polarization 3D imaging method for monitoring tunnel surrounding rock deformation. This method enables simultaneous, large-format measurement of the 3D deformation of tunnel surrounding rock without the need for multiple control points, avoiding errors introduced by repeated station setups. The measurement process is simple and efficient, significantly reducing the impact on construction.

[0018] (1) In engineering practice, total stations are widely used for deformation monitoring. However, the use of total stations requires the placement of reflectors at the measurement points, which limits the placement location. This invention adopts a new monocular polarization three-dimensional imaging technology, which uses a single camera to instantly obtain a three-dimensional model of the tunnel surrounding rock. Compared with traditional binocular and multi-view three-dimensional imaging technologies, it does not require obtaining the equipment location or on-site calibration, thus minimizing interference with construction and ensuring measurement accuracy while simplifying the measurement operation. In addition, compared with existing single-point mapping methods such as total stations, infrared rangefinders, and laser rangefinders, this invention uses optical imaging to obtain regional deformation, which greatly improves efficiency.

[0019] (2) Based on the classic monocular polarization three-dimensional imaging technology, targeted improvements were made for tunnel surrounding rock deformation monitoring. By using a feature-based matching algorithm, the two phase images were projected onto the same coordinate system to solve the problem of inconsistent positions of the two phase measurement devices. The problem that the classic monocular polarization three-dimensional imaging technology could not obtain the true value of the three-dimensional deformation was solved by using pre-calibrated experimental equipment. Attached Figure Description

[0020] Figure 1 This is a flowchart of the method of the present invention.

[0021] Figure 2 This is a schematic diagram of the monocular polarization three-dimensional imaging principle according to an embodiment of the present invention. Detailed Implementation

[0022] The method of this invention is as follows:

[0023] S1. Collect two phases of focused plane polarization images of the surrounding rock of the tunnel in the monitoring area. The focused plane polarization images of the surrounding rock of the tunnel in the monitoring area in each phase are obtained by the same focused plane polarization camera.

[0024] The placement of the two phases of the focal plane polarization camera does not need to be exactly the same, but it must be ensured that the two measurements cover the same deformation area.

[0025] S2. Preprocess the focused planar polarization images of the tunnel surrounding rock in the two monitoring areas, including polarization interpolation, image filtering, histogram equalization, and histogram matching.

[0026] S3. Perform stereo matching on the preprocessed two-phase monitoring area tunnel surrounding rock focal plane polarization images, project the second phase focal plane polarization image onto the coordinate system of the first phase focal plane polarization image, and obtain the image after the second phase polarization image is projected.

[0027] The purpose of stereo matching the preprocessed two-phase monitoring area tunnel surrounding rock focal plane polarization images is to find the transformation relationship between the coordinate systems of the two images. The specific process is as follows:

[0028] To capture two photos of overlapping scenes, we first capture many feature points based on the principle that feature points have local differences. Then, we optimize the feature points, describe the optimized key feature points, and finally find an optimal matrix to describe the correspondence between the feature points of the two images, thereby determining the relative positions of the two images.

[0029] S4. Perform monocular polarization 3D reconstruction on the first-phase polarization image and the second-phase polarization image after projection to obtain two-phase tunnel surrounding rock 3D models.

[0030] The specific process of monocular polarization 3D reconstruction:

[0031] First, the polarization degree and polarization angle information of the tunnel surrounding rock surface are obtained using a single polarization image. Second, the zenith angle of the tunnel surrounding rock surface is calculated from the polarization degree information, and the azimuth angle of the tunnel surrounding rock surface is calculated from the polarization angle information. Then, the normal vector of the tunnel surrounding rock surface is represented by the zenith angle and the azimuth angle. Finally, the three-dimensional model of the tunnel surrounding rock is obtained by integrating the normal vector.

[0032] S5. The three-dimensional model of the tunnel surrounding rock obtained from the first phase polarized image and the three-dimensional model of the tunnel surrounding rock obtained from the projected image of the second phase polarized image are in the same coordinate system. By calculating the elevation difference at each coordinate position, the elevation change at that coordinate position can be determined. By calibrating the elevation change, the true value of the tunnel surrounding rock deformation can be calculated.

[0033] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0034] This embodiment utilizes a single polarization camera to acquire two images of the tunnel surrounding rock before and after deformation. The two images are preprocessed and stereo matched. The second polarization image is projected onto the coordinate system of the first polarization image. Monocular polarization 3D reconstruction is performed on both images to obtain two 3D models of the tunnel surrounding rock. The 3D deformation of the tunnel surrounding rock is then calculated, and the actual deformation value is adjusted. The main components include:

[0035] (1) Collect focal plane polarization images of the surrounding rock of the tunnel in the two monitoring periods.

[0036] The focused planar polarization images of the tunnel surrounding rock in each monitoring period are obtained by the same focused planar polarization camera; the placement of the focused planar polarization camera in the two periods does not need to be exactly the same, but it is necessary to ensure that the two measurements cover the same deformation area.

[0037] A focal plane polarization camera, model MER-502-79U3M POL, from Daheng Imaging Company was selected. The detector size is 2048×2448, and its focal plane is simultaneously marked with 0°, 45°, 90°, and 135° polarization gratings, enabling it to output a two-dimensional mosaic image containing polarization information at 0°, 45°, 90°, and 135°. This polarization camera was used to perform two-stage imaging of the tunnel surrounding rock, obtaining one focal plane polarization image I1 and I2 in each stage.

[0038] (2) Preprocessing the focused planar polarization images of the surrounding rock of the tunnel in the two monitoring periods.

[0039] Preprocessing was performed on the focal plane polarization images of the surrounding rock of the tunnel in the two monitoring areas, including polarization interpolation, image filtering, histogram equalization, and histogram matching.

[0040] ① Polarization interpolation

[0041] To address the issue of image intensity discontinuities along a single polarization direction, the polarization interpolation method employs a classic demosaic algorithm. After polarization interpolation, the acquired I... j Four polarization images were calculated from the image, denoted as I. j_0 I j_45 I j_90 I j_135 Where j = 1, 2, represents the measurement period number, and the specific method is as follows:

[0042] The entire focal plane polarization image with a size of 2M×2N is divided into 2×2 units. However, except for the outermost row and column, all other pixels are used more than 4 times, contributing to the calculation of the 4 superpixels.

[0043] Starting with the first 2×2 polarization unit consisting of pixels (1,1), (1,2), (2,1), and (2,2), the polarization focal plane array is translated across the entire focal plane array with a row-wise step size of a and a column-wise step size of a, dividing the polarization focal plane array into (2M-1)×(2N-1) 2×2 units, with overlap between units. The resulting four polarization images are all (2M-1)×(2N-1) in size, ensuring that the resolution of the polarization images is almost not reduced.

[0044] ②Image filtering

[0045] Image filtering employs an adaptive median filtering method to remove noise from the image, implemented using a classic adaptive filter.

[0046] ③ Histogram equalization

[0047] Histogram equalization uses the classic histogram equalization algorithm to adjust the contrast between two images to meet the matching requirements of the two images.

[0048] ④ Histogram matching

[0049] Histogram matching uses the classic histogram matching algorithm to balance the brightness difference between two images to meet the matching requirements of the two images.

[0050] (3) Perform stereo matching on the preprocessed two-phase monitoring area tunnel surrounding rock focal plane polarization images.

[0051] Stereo matching was performed on the preprocessed two-phase monitoring area tunnel surrounding rock focal plane polarization images. The second phase focal plane polarization image was projected onto the coordinate system of the first phase focal plane polarization image to obtain the projected image of the second phase polarization image.

[0052] The purpose of stereo matching the preprocessed two-phase monitoring area tunnel surrounding rock focal plane polarization images is to find the transformation relationship between the coordinate systems of the two images. The specific process is as follows:

[0053] To capture two photos of overlapping scenes, we first capture many feature points based on the principle that feature points have local differences. Then, we optimize the feature points, describe the optimized key feature points, and finally find an optimal matrix to describe the correspondence between the feature points of the two images, thereby determining the relative positions of the two images.

[0054] The purpose of stereo matching of two polarization images is to obtain the geometric transformation relationship between two images located in the same space but with different scenes. This invention employs a feature-based method, which is highly resistant to rotation, perspective, and grayscale changes, and possesses feature scale invariance, rotation invariance, and illumination invariance. The specific process is as follows:

[0055] ① Feature point capturing

[0056] Take two photos with overlapping scenes. To ensure a sufficient number of common feature points, the overlap degree shall be 30% or more; establish an image pyramid to find key points at different scales. Based on the principle that feature points have local distinctiveness, take a window centered at each point, for example, a 5×5 or 7×7 pixel window, to describe the surrounding environment of the feature point. Move the window in any direction, and if the surrounding environment changes significantly, it can be deduced that the point has local distinctiveness.

[0057] Calculate the brightness gradient of the image in the x and y directions using the Sobel operator, and smooth the gradient with a Gaussian function of σ=1.5 to reduce the influence of noise. It is easy to find that when moving the window in a region with gentle image changes, the change of the accumulated brightness value of a small area is not obvious; a slight movement of the window near a key point will strongly change the accumulated brightness value.

[0058] The brightness variation value can be calculated by the following formula:

[0059] E(u,v)=∑ x,y w(x,y)[I(x+u,y+v)-I(x,y)] 2

[0060] Wherein, w(x,y) is the weight of the Gaussian function, and I(x,y) is the brightness gradient at this point.

[0061] During calculation, this formula can be approximated as:

[0062]

[0063] By comparing the eigenvalues I1 and I2 of the matrix , the state of the point can be determined. If I1>>I2 or I1<<I2, it indicates that the point is located at a longitudinal or transverse edge; if I1 and I2 are approximate and have very small values, it indicates that the point is located in a smooth region; if I1 and I2 are approximate but have large values, it indicates that the point is a key point.

[0064] ② Adaptive non-maximum suppression

[0065] Since there are many key points captured in the previous step, it is necessary to select some preferred feature points. Set the initial value of a radius r to infinity. When r decreases continuously, retain the key points whose R values of other key points within the radius r are all smaller than the R value of the center point, add them to the queue, and stop the search after the number of key points in the queue reaches a preset value.

[0066]

[0067] x i is the two-dimensional coordinate of the key point, crobust =0.9.

[0068] ③ Key point description

[0069] The above operations have obtained the location coordinates, scale information, and orientation of the keypoints. A direction function needs to be specified for the feature points to ensure the extracted features are rotationally invariant. The basic idea of ​​descriptors is to divide the region surrounding the keypoint into blocks, calculate the gradient histogram within each block, and generate a unique vector. This vector includes not only the keypoint but also the surrounding pixels. An 8×8 neighborhood is taken around the keypoint, where each small cell represents a pixel in the scale space of the feature point's neighborhood. Every 2×2 pixels form a sub-region, and each sub-region generates an eight-directional gradient histogram. The sum of each gradient direction forms a seed point, and a feature point is composed of information from 4×4 seed points. Each sub-region generates a descriptor, and a descriptor involves 8 directions. Each keypoint has 4×4×8 = 128 dimensions.

[0070] Keypoint descriptor generation steps:

[0071] First, determine the image region required for calculating the descriptor. The formula for calculating the radius of the image region is:

[0072]

[0073] Where, σ oct It is the intra-group scale of the key point group, d=4.

[0074] Next, move the coordinates to the main direction of the key point, and rotate them to get the new coordinates:

[0075]

[0076] The region is then divided into 4×4 sub-blocks. For each sub-block, histogram calculations are performed in eight directions to obtain the gradient magnitude in each direction, resulting in a total of 128-dimensional descriptive vectors. Within the image radius, the gradient magnitude and direction of each pixel are calculated. Each gradient magnitude is multiplied by a Gaussian weight parameter to generate a direction histogram. The descriptor gradient direction histogram is calculated from the blurred image at the scale of the keypoint. The direction histogram is calculated as follows:

[0077]

[0078] x k : Column distance between this point and the key point;

[0079] y k The line distance between this point and the key point

[0080] σ w: equal to half of the width of the description sub-window, 3σ * the number of histogram columns (taken as 4);

[0081] weight: represents the value of a bin in the directional histogram.

[0082] Calculate gradient direction histograms in eight directions within a 2×2 window area, and plot the accumulated value for each gradient direction to form a seed point. Then, perform histogram statistics in the next 2×2 area to form the next seed point, generating a total of 16 seed points.

[0083] The resulting 128 descriptor vectors are:

[0084] W = (w1, w2, ..., w 128 )

[0085] Standardization:

[0086]

[0087] Normalized vector:

[0088] L=(l1,l2,...,l 128 )

[0089] ④ Key point matching

[0090] Homography can be simply understood as a transformation used to describe the positional mapping between an object in the world coordinate system and the pixel coordinate system. The corresponding transformation matrix is ​​called the homography matrix. The homography matrix can be represented as:

[0091]

[0092] Keypoint matching employs the Random Sample Consensus (RANSAC) algorithm. RANSAC shares some similarities with least squares, but the difference lies in the fact that least squares tries to adapt to all points, including outliers, while RANSAC can produce a model calculated using only inliers with sufficiently high probabilities; however, careful parameter configuration is essential. Experiments have shown that for datasets containing 80% error, RANSAC significantly outperforms least squares.

[0093] The RANSAC algorithm is used to find an optimal homography matrix H that maximizes the number of data points satisfying this matrix. Typically, h is set to... 33 =1 to normalize the matrix. Since the homography matrix has 8 unknown parameters, at least 8 linear equations need to be solved. A pair of points can be used to derive two equations, so there are at least 4 pairs of matching points.

[0094]

[0095] Where (x,y) represents the corner position of the first image, (x',y') represents the corner position of the second image, and s is the scale parameter.

[0096] The RANSAC algorithm randomly selects four samples from the matching dataset, ensuring that these four samples are not collinear, calculates the homography matrix, and then uses this model to test all data. It calculates the number of data points that satisfy this model and the projection error (i.e., the cost function). If this model is optimal, then the corresponding cost function is minimized. The cost function is as follows:

[0097]

[0098] Furthermore, in step S3, the second polarization image is projected onto the coordinate system of the first polarization image. First, a new blank canvas is created, the upper and lower boundaries of the two-dimensional coordinates of the two images after projection are compared, the maximum value of each direction boundary is selected as the size of the new image, and the intersection area of ​​the two images is calculated.

[0099] After processing, the processed image I'2 is obtained. The superscript "'" indicates the image after the second polarization image is projected. At this time, I1 and I'2 are in the same coordinate system.

[0100] (4) Monocular polarization 3D reconstruction was performed on the first phase polarization image and the second phase polarization image after projection to obtain the two phase tunnel surrounding rock 3D models.

[0101] First, the polarization degree and polarization angle information of the tunnel surrounding rock surface are obtained using a single polarization image. Second, the zenith angle of the tunnel surrounding rock surface is calculated from the polarization degree information, and the azimuth angle of the tunnel surrounding rock surface is calculated from the polarization angle information. Then, the normal vector of the tunnel surrounding rock surface is represented by the zenith angle and the azimuth angle. Finally, the three-dimensional model of the tunnel surrounding rock is obtained by integrating the normal vector.

[0102] The specific process of monocular polarization 3D reconstruction:

[0103] ① Obtain the target surface normal

[0104] According to the law of reflection, the angle of incidence and the angle of reflection of the incident light are equal, both equal to angle θ in the diagram. The reflected light propagates along the z-axis, and the detector lies in the xoy plane. (Target surface normal) The projection onto the xoy plane is The angle with the x-axis is This indicates the incident azimuth angle of the incident light.

[0105] ② Calculate the degree of polarization and polarization angle from the polarization image.

[0106] The four polarization images obtained from step S2 are denoted as I... j_0 I j_45 I j_90 I j_135 In this context, j = 1, 2 represents the measurement period. Using Stokes parameters to represent polarized light, the Stokes parameters can be expressed as:

[0107]

[0108] S0 = I j_0° +I j_90°

[0109] S1=I j_0° -I j_90°

[0110] S2=I j_45° -I j_135°

[0111] S3 = I j_L -I j_R

[0112] Among them, I j_0 I j_90° I represents the light intensity in the polarization directions of 0° and 90°, respectively. j_45 I j_135° I represents the light intensity in polarization directions of 45° and 135°. j_L I j_R S0 and S2 represent the light intensities of left-handed and right-handed circularly polarized light, respectively. Therefore, S0 is proportional to the ordinary light intensity, S1 and S2 represent the difference in linear polarization intensity in different directions, and S3 represents the difference between left-handed and right-handed circularly polarized light.

[0113] The degree of polarization represents the proportion of polarized light intensity to the total light intensity, and is expressed by the Stokes parameter as follows:

[0114]

[0115] Circularly polarized light is rare in nature, and S3 can generally be ignored; therefore, the degree of linear polarization can be obtained:

[0116]

[0117] ψ is the angle between the major axis and the x-axis of the ellipse, also known as the polarization angle, which plays an important role in polarization imaging detection. It is represented by the Stokes parameter as follows:

[0118]

[0119] ③ Calculate the zenith angle and azimuth angle of the normal to the target surface.

[0120] To more intuitively represent the surface normal of the target to be reconstructed, the incident angle θ, and the incident azimuth angle... The relationship between the normals is expressed in polar coordinates. It is expressed as follows, where the incident angle θ is called the zenith angle of the normal, and the incident azimuth angle is called the azimuth angle of the normal.

[0121] Based on the definition of polarization degree and the Fresnel reflectance formula, the relationship between polarization degree P and zenith angle θ (normal N and detector line of sight) can be derived:

[0122]

[0123]

[0124]

[0125] The normal N is represented by the following formula:

[0126]

[0127]

[0128]

[0129] The azimuth angle φ can be obtained from the Stokes vector:

[0130] φ = ψ + (0 or π)

[0131]

[0132] To address the uncertainty in azimuth angle, boundary conditions are introduced. Specifically, φ = ψ or φ = ψ + π is directly set at the occluding boundary. Then, φ is selected at adjacent points to ensure the normal direction aligns with the normal direction at the boundary. Finally, the remaining uncertain φ is iterated to the last azimuth angle according to the aforementioned conditions and rules.

[0133] The zenith angle and azimuth angle together determine the normal to the target surface, and its expression is:

[0134]

[0135] Where n x n y n z Let θ represent the components of the normal along the x-axis, y-axis, and z-axis, respectively, where θ∈[0°,90°]. Therefore, the target surface normal can be determined after obtaining the zenith angle and azimuth angle of the normal.

[0136] ④ Solve the target 3D model by integrating the normals.

[0137] Assuming the target surface is continuously integrable, and using the function Z(x,y) to represent the height of the target surface, we obtain the following expression:

[0138]

[0139]

[0140] The above expression defines the two partial derivatives of the surface function Z(x,y) as p and q, respectively. For such integrable normal gradients, the target surface can be recovered using local or global integration methods.

[0141] After obtaining the micro-surface element normals by utilizing the diffuse reflection polarization characteristics of object surfaces in natural scenes, it is not possible to directly reconstruct the 3D surface through integration due to the unique nature of solving 3D surfaces. Therefore, a 3D surface restoration function based on the Frankot-Chellappa algorithm is established. Its surface function z(x,y) can be represented as a series of orthogonal basis functions. A linear combination, w = (w x ,w y The function z(x,y) is in two-dimensional form. Therefore, through a linear combination of orthogonal basis functions, its surface function z(x,y) can be expressed as shown in the following equation:

[0142]

[0143] C(w) represents the expansion coefficients of the function z(x,y). The gradient field of the surface function is expressed in the orthogonal function expansion form shown below:

[0144]

[0145]

[0146] and Let represent the optimal sets of coefficients in the x and y directions, respectively. To simplify the algorithm implementation, Fourier coefficients with a complete orthogonal basis are chosen for expansion, and their basis functions can be represented as follows:

[0147]

[0148] In the formula, M and N represent the dimensions of the two-dimensional image, respectively. Then, by performing a Fourier transform on the basis functions and combining it with the following formula, the gradient field of the surface function can be obtained:

[0149]

[0150] Therefore, we can obtain the surface function z(x, y) of the object with respect to the gradient field (p). g ,qg The representation of ) is shown in the following formula:

[0151]

[0152] F{} and F -1 {} represent the Discrete Fourier Transform and the Inverse Fourier Transform, respectively. This model establishes a connection between the discrete gradient field information and the surface function information in the spatial domain in the Fourier domain through the decomposition of the Fourier orthogonal basis to reconstruct the surface function z(x,y), and then solves for the target surface height information dataset.

[0153] The three-dimensional model is calculated using the surface function formulas described above, and the three-dimensional result is expressed in Z-shape. j This indicates that j = 1, 2 represents the test period number.

[0154] (5) Calculation of three-dimensional deformation of tunnel surrounding rock

[0155] The 3D model of the tunnel surrounding rock obtained from the first phase of polarized images and the 3D model of the tunnel surrounding rock obtained from the projected image of the second phase of polarized images are in the same coordinate system. By calculating the elevation difference at each coordinate position, the elevation change at that coordinate position can be determined. By calibrating the elevation change, the true value of the tunnel surrounding rock deformation can be calculated.

[0156] The relative value of deformation Z is obtained by calculating the difference between the two three-dimensional models of the surrounding rock of the tunnel. The formula for calculating the relative value of deformation Z is as follows:

[0157] Z = Z2 - Z1

[0158] Z1 represents the three-dimensional result obtained from the first phase of image inversion, and Z2 represents the three-dimensional result obtained from the first phase of image inversion.

[0159] If the actual length of a line segment in the area to be measured is 20mm, and its pixel count in the image is 169, then the conversion factor is:

[0160]

[0161] The true value of the deformation can be calculated using the following formula:

[0162] Z_true=c·Z

[0163] The measured deformation of a tunnel under construction is as follows:

[0164] Actual value (mm) Measured value (mm) Error (mm) Wide cracks in tunnel wall 3.65 4.04 0.39 Narrow cracks in the tunnel wall 0.9 1.4 0.5 Tunnel surrounding rock deformation 13.73 12 1.73

[0165] The contents not described in detail in this specification are common knowledge to those skilled in the art.

Claims

1. A method for monitoring tunnel surrounding rock deformation using monocular polarization three-dimensional imaging, characterized in that, include: Two phases of focused plane polarization images of the tunnel surrounding rock were acquired in the monitoring area. The focused plane polarization images of the tunnel surrounding rock in each phase of the monitoring area were obtained by the same focused plane polarization camera. Preprocessing was performed on the focal plane polarization images of the surrounding rock of the tunnel in the two monitoring areas, including polarization interpolation, image filtering, histogram equalization, and histogram matching. The polarization interpolation process includes: dividing the entire focal plane polarization image of size 2M×2N into 2×2 units as the smallest unit; except for the outermost row and column, all other pixels are used more than 4 times to contribute to the calculation of the 4 superpixels. Starting with the first 2×2 polarization unit consisting of pixels (1,1), (1,2), (2,1) and (2,2), the polarization unit is translated across the entire focal plane array with a row step size a and a column step size a, dividing the polarization focal plane array into (2M-1)×(2N-1) 2×2 units, with overlap between units; Stereo matching was performed on the preprocessed two-phase monitoring area tunnel surrounding rock focal plane polarization images. The second phase focal plane polarization image was projected onto the coordinate system of the first phase focal plane polarization image to obtain the projected image of the second phase polarization image. Monocular polarization 3D reconstruction was performed on the first-phase polarization image and the second-phase polarization image after projection to obtain two-phase tunnel surrounding rock 3D models. The three-dimensional model of the tunnel surrounding rock obtained from the first phase polarized image and the three-dimensional model of the tunnel surrounding rock obtained from the image projected from the second phase polarized image are in the same coordinate system. The elevation difference at each coordinate position is calculated to obtain the elevation change at that coordinate position. The elevation change is calibrated to calculate the true value of the tunnel surrounding rock deformation. The placement of the two phases of the focal plane polarization camera does not need to be exactly the same, but it must be ensured that the two measurements cover the same deformation area. Stereo matching is performed on the preprocessed two-phase monitoring area tunnel surrounding rock focal plane polarization images, including: taking two photos with overlapping scenes, first capturing multiple feature points based on the principle that feature points have local differences, then optimizing the feature points, then describing the optimized key feature points, and finally finding an optimal matrix to describe the correspondence between feature points in the two images, thereby determining the relative position of the two images. The description of the selected key feature points includes: By dividing the region surrounding the keypoint into blocks, calculating the gradient histogram within each block, and generating a unique vector, the vector includes not only the keypoint but also the surrounding pixels. An 8×8 neighborhood is taken around each keypoint, with each small cell representing a pixel in the scale space of the feature point's neighborhood. Every 2×2 pixels form a sub-region, and each sub-region generates an eight-directional gradient histogram. The sum of each gradient direction forms a seed point, and a feature point is composed of information from 4×4 seed points. Each sub-region generates a descriptor, which involves 8 directions, resulting in 4×4×8=128 dimensions for each keypoint. Monocular polarization 3D reconstruction includes: First, the polarization degree and polarization angle information of the tunnel surrounding rock surface are obtained using a single polarization image. Second, the zenith angle of the tunnel surrounding rock surface is calculated from the polarization degree information, and the azimuth angle of the tunnel surrounding rock surface is calculated from the polarization angle information. Then, the normal vector of the tunnel surrounding rock surface is represented by the zenith angle and the azimuth angle. Finally, the three-dimensional model of the tunnel surrounding rock is obtained by integrating the normal vector.

Citation Information

Patent Citations

  • Base-line-based three-dimensional deformation monitoring method of subway tunnel

    CN108020196A

  • Instantaneous passive area array three-dimensional imaging method of time-varying ground object

    CN115184953A