Triaxial test sample shear deformation process reconstruction method based on visual perception

Abstract

In order to obtain an influence rule of formation and development of a strain localization shear band in a rock-soil body on mechanical properties, the invention aims to provide a triaxial test sample shear deformation process reconstruction method based on visual perception. The method has the advantages that the rock-soil body strain localization shear band forming process and related discrete-continuous medium related motion characteristics can be captured in a non-contact intelligent and lossless mode, and complex microstructure information can be obtained while sample surface displacement and strain are intelligently sensed. According to the technology, real-time shooting is carried out on the rock-soil body sample shearing process through multi-view external load visual identification equipment, threshold segmentation calculation and binarization processing are carried out after shot images are screened, and then the formation of non-uniform deformation of a soil body and the development condition of a non-continuous deformation failure form are intelligently measured. And intelligent calculation of a local concentrated strain distribution area is realized based on a visual algorithm.

Description

Technical Field

[0001] The present invention belongs to the field of information technology and intelligent algorithms, and in particular relates to a method for reconstructing the shear deformation process of a triaxial test specimen based on visual perception. Background Art

[0002] Traditional rock and soil testing technology mainly relies on contact sensors to obtain stress and deformation data. Related equipment includes pressure sensors, displacement sensors and load sensors, which are used to measure the stress-strain relationship of rock and soil specimens under different stress states. These data are helpful for analyzing the shear strength, compressibility and deformation characteristics of rock and soil. However, this method has certain limitations in capturing the local discontinuous deformation development and dynamic changes of failure morphology of the specimen. Since contact sensors can only measure stress and strain at specific points or areas, it is difficult to fully reflect the full-field stress distribution and deformation characteristics of the specimen. In addition, the accuracy and response speed of contact sensors are limited, making it difficult to capture the rapidly changing failure process in real time. Summary of the invention

[0003] In order to solve the above technical problems, the present invention proposes a method for reconstructing the shear deformation process of a triaxial test specimen based on visual perception to solve the problems existing in the above prior art.

[0004] To achieve the above object, the present invention provides a method for reconstructing the shear deformation process of a triaxial test specimen based on visual perception, comprising the following steps:

[0005] Set up an external visual device and use a dot calibration plate to calibrate the camera in the external visual device;

[0006] Using a calibrated camera to capture an experimental image during the shearing experiment, and preprocessing the experimental image to obtain a preprocessed image;

[0007] Segmenting and extracting features from the preprocessed image to obtain a local discontinuous deformation region image;

[0008] Identify the local discontinuous deformation area image, obtain each node of the shear band network, obtain the shear band trunk and the pixels contained in each shear band based on each node of the shear band network; and trace each shear band line along the shear band trunk based on skeletonization to obtain a complete shear band;

[0009] Calculate the displacement gradient and strain tensor of rock and soil specimens during shearing based on the complete shear band;

[0010] The shear process is reconstructed based on the displacement vector and the strain tensor to obtain a visualized rock and soil shear gradient process.

[0011] Preferably, the external visual equipment includes a high-resolution camera and LED lighting equipment;

[0012] The high-resolution camera is used to capture images of the entire test area during the shearing process in real time;

[0013] The LED lighting device is used to provide a stable light source.

[0014] Preferably, the method for calibrating a camera in an external visual device comprises:

[0015] Use the cv2.findChessboardCorners function in OpenCV to detect the corners on the calibration board. When the corners are detected, use the cv2.cornerSubPix function to optimize the sub-pixel accuracy.

[0016] Based on the optimized corner point coordinates, use the cv2.calibrateCamera function to calculate the camera's intrinsic and extrinsic matrix to calibrate the camera.

[0017] Use feature extraction algorithm to extract feature points in the experimental image and calculate its descriptor; perform feature matching on images with different monocular viewpoints to obtain feature point pairs;

[0018] Estimating the feature point pair through relative posture to obtain the relative posture of the camera;

[0019] Based on the relative postures of the cameras, the matched feature point pairs are triangulated to obtain the three-dimensional coordinates of the feature points in the three-dimensional space.

[0020] Preferably, the three-dimensional coordinates are solved by minimizing the reprojection error, and the solution expression is:

[0021] minimize∑∥x i -π(P i ,X)∥ 2

[0022] Among them, X i Represents the observed three-dimensional coordinate point, P i Represents the three-dimensional point that needs to be projected onto the two-dimensional plane, π(P i ,X) represents the projection function, which projects the three-dimensional point X onto the image plane of the i-th view.

[0023] Preferably, the method for preprocessing the experimental image comprises:

[0024] Determine the blur degree of the image through blur detection and remove blurry images;

[0025] Adaptively set the threshold value according to the local features of the local discontinuous deformation in the remaining image;

[0026] The residual image is converted into a grayscale image and smoothed.

[0027] Preferably, the method for performing smoothing processing includes:

[0028] According to the characteristic that the area of ​​the noise points is much smaller than the area of ​​the shear band, the maximum noise point area interval is determined to obtain the optimal area threshold, the noise removal operation is performed using the optimal area threshold, and the image after the noise removal operation is processed using Gaussian filtering.

[0029] Preferably, the method for segmenting and extracting features from the preprocessed image comprises:

[0030] Segment the local non-continuous deformation area image based on the adaptively set threshold;

[0031] The edge of the image in the local non-continuous deformation area is obtained through Canny edge detection;

[0032] The shear band is bridged, and the noise in the bridged image is removed by morphological operations, and the clear outline of the shear band is extracted;

[0033] The scan line algorithm is used to update the intersection information.

[0034] Preferably, the method for obtaining a complete shear band comprises:

[0035] In the skeletonized image, starting from a node, the search is extended along the trunk, and the trunk pixels found are recorded until another node is found. The trunk of a shear band is determined based on the two nodes, and the remaining non-trunk shear band pixels are merged into the nearest connected shear band trunk through a merging algorithm to form a complete shear band.

[0036] Preferably, the calculation expression of the displacement gradient is:

[0037]

[0038] Among them, u x and u y are the displacements in the x and y directions, respectively, are the displacement gradients of the subset in the x and y directions.

[0039] Preferably, the method for calculating the strain tensor includes:

[0040] When the deformation is small, the calculation is performed by the linear strain tensor;

[0041] When large deformations occur the Green-Lagrange strain tensor is used for calculations.

[0042] Compared with the prior art, the present invention has the following advantages and technical effects:

[0043] By introducing computer vision and image processing technology, the present invention provides a new intelligent identification method for rock and soil shear tests, which can capture the development and failure form of local discontinuous deformation inside the soil in real time and accurately. The implementation of the entire system includes equipment installation and configuration, real-time image acquisition and processing during the test, local discontinuous deformation feature analysis, multi-viewpoint fusion and three-dimensional reconstruction, as well as experimental report generation and data storage. By applying advanced image processing algorithms and machine learning techniques, the system can provide detailed and intuitive test data, providing strong support for geotechnical test research. In the future, with the further development of computer vision technology, this method can be applied in a wider range of engineering fields, providing a more accurate and scientific basis for engineering safety and design. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0045] Figure 1 is a flow chart of a method according to an embodiment of the present invention;

[0046] Figure 2 is a schematic diagram of a target according to an embodiment of the present invention;

[0047] Figure 3 A schematic diagram of a target recognition process according to an embodiment of the present invention;

[0048] Figure 4 is a measurement image of a sample according to an embodiment of the present invention;

[0049] Figure 5 A schematic diagram of a visual perception device according to an embodiment of the present invention;

[0050] Figure 6 This is a schematic diagram of the visual imaging principle of an embodiment of the present invention;

[0051] Figure 7 It is a schematic diagram of strain calculation according to an embodiment of the present invention;

[0052] Figure 8 A schematic diagram of an image screening process according to an embodiment of the present invention;

[0053] Fig. 9 A schematic diagram of a shear band trunk extraction process according to an embodiment of the present invention;

[0054] Fig.10 3D reconstruction flow chart of an embodiment of the present invention. DETAILED DESCRIPTION

[0055] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0056] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0057] Embodiment 1

[0058] like Figure 1 As shown, this embodiment provides a method for reconstructing the shear deformation process of a triaxial test specimen based on visual perception, comprising the following steps:

[0059] Set up an external visual device and use a dot calibration plate to calibrate the camera in the external visual device;

[0060] Using a calibrated camera to capture an experimental image during the shearing experiment, and preprocessing the experimental image to obtain a preprocessed image;

[0061] Segmenting and extracting features from the preprocessed image to obtain a local discontinuous deformation region image;

[0062] Identify the local discontinuous deformation area image, obtain each node of the shear band network, obtain the shear band trunk and the pixels contained in each shear band based on each node of the shear band network; and trace each shear band line along the shear band trunk based on skeletonization to obtain a complete shear band;

[0063] Calculate the displacement gradient and strain tensor of rock and soil specimens during shearing based on the complete shear band;

[0064] The shear process is reconstructed based on the displacement vector and the strain tensor to obtain a visualized rock and soil shear gradient process.

[0065] The specific method is as follows:

[0066] S1. First install the equipment as follows:

[0067] S1.1 installs external visual recognition equipment around the unit shear test equipment, including two high-resolution cameras and LED lighting equipment. One is located in front of the equipment and the other is located on the side to ensure that the entire test area can be covered. The camera is responsible for real-time image capture during the shearing process, and the LED lighting equipment provides a stable light source to ensure clear images.

[0068] S1.2 Connect the camera to the image acquisition module via a high-speed data cable, and connect the image acquisition module to the data processing system. Ensure that the data cable connection is stable and does not affect the normal operation of the test equipment. Install the image processing software in the data processing system and configure the relevant parameters, including the image resolution, shooting frequency and processing algorithm parameters. The camera transmits the captured images to the data processing system in real time through the image acquisition module.

[0069] S1.3 Configure the data processing system. Install the image processing software on a high-performance computer and set the parameters for image screening, threshold segmentation, and binarization. Through these processing steps, the required information can be extracted from the original image, and the image can be enhanced and optimized.

[0070] S2. After the equipment is installed, use the dot calibration plate to calibrate the camera:

[0071] During the S2.1 calibration process, the cv2.findChessboardCorners function in OpenCV is used to detect the corner points on the calibration board. If corner points are detected, the cv2.cornerSubPix function is used to optimize the sub-pixel accuracy.

[0072] S2.2 Assume that the coordinates of the corner points on the calibration board in the world coordinate system are known. Usually, the z coordinate of the corner point can be set to 0 to obtain the 2D plane coordinates. The detected corner point coordinates are stored in an array as the corner point coordinates in the image coordinate system. Then, the cv2.calibrateCamera function is used to calculate the camera's intrinsic parameter matrix K and extrinsic parameter matrix [R|T] to calibrate the camera.

[0073]

[0074] Among them, f x and f y is the focal length, C x and C y are the optical center coordinates.

[0075] P=K[R|T]

[0076] Where R is the rotation matrix and T is the translation vector.

[0077] S2.3 For each monocular camera, use the feature extraction algorithm to intelligently extract feature points in the image and calculate its descriptor. Then perform feature matching between images from different monocular perspectives.

[0078] The KNN matching algorithm is used to match the extracted features and calculate the Euclidean distance between its feature descriptor and all feature descriptors in another image:

[0079]

[0080] Where p and q are two 128-dimensional feature vectors.

[0081] For each feature point, find the k nearest neighbors in the other image (usually k = 2), and determine whether it is a correct match based on the nearest neighbor distance ratio. The nearest neighbor distance ratio can be defined as:

[0082]

[0083] The threshold is set to 0.75, and the match is considered reliable only when the Ratio is less than this threshold.

[0084] Calculate the model's error (such as reprojection error) for all points, and consider points with errors less than a preset threshold as inliers. For the basic matrix F, the reprojection error can be expressed as:

[0085]

[0086] Where x1 and x2 are the homogeneous coordinates of the matching feature point pair.

[0087] S2.4 Relative pose estimation From the point pairs obtained from feature matching, the five-point method is used to calculate the relative pose between the cameras (i.e., the relative rotation matrix R and the displacement vector t).

[0088] The essential matrix E satisfies the following constraints:

[0089]

[0090] The essential matrix can be decomposed into the relative rotation matrix R and the antisymmetric matrix of the translation vector t:

[0091] E=[t] × R

[0092]

[0093] For each pair of corresponding points (x1, x2), according to the essential matrix constraint A linear equation can be obtained. After expansion, the following form is obtained:

[0094]

[0095] Where E ijis the element of the essential matrix E. For five pairs of points, a 5x 9 equation matrix can be obtained, and then the real solution is extracted by singular value decomposition (SVD) decomposition and polynomial constraints. For each solution, it is substituted into the calculated E = αE1 + βE2 and singular value decomposition is performed. The solution whose singular value meets the constraints is selected, and the relative rotation R and translation t of the camera are calculated.

[0096] The final relationship between the basic matrix F and the camera intrinsic parameter matrix K is as follows:

[0097]

[0098] Among them, E is the essential matrix, and the relative posture between cameras can be obtained by decomposing the essential matrix.

[0099] S2.5 triangulates the matched feature point pairs through the relative pose of the camera to restore the positions of these feature points in three-dimensional space. For points x1 and x2 of two perspectives, their three-dimensional coordinates can be solved by minimizing the reprojection error:

[0100] minimize∑∥x i -π(P i ,X)∥ 2 ;

[0101] Among them, x i represents the observed three-dimensional coordinate point, P i Represents the three-dimensional point that needs to be projected onto the two-dimensional plane, π(P i ,X) is the projection function, which projects the three-dimensional point X onto the image plane of the i-th view.

[0102] S3. In the data processing system, the image processing software preprocesses the received image. The preprocessing steps include image screening, threshold segmentation and binarization.

[0103] S3.1 The purpose of image screening is to remove blurry or invalid images, so as to ensure the quality of the image in subsequent processing. Blur detection is used to determine the blurriness of the image by calculating the gradient or Laplacian operator of the image.

[0104] Gradient detection formula:

[0105] Laplacian operator:

[0106] By calculating the contrast of the image, it is determined whether the image is valid.

[0107]

[0108] Among them, I max and Imin are the maximum and minimum grayscale values ​​of the image respectively.

[0109] S3.2 Adaptive threshold segmentation after screening

[0110] Adaptively setting the threshold value according to the local features of the local discontinuous deformation in the image;

[0111] Assuming the image size is M×N, first use the global threshold method to segment the image and divide the image into n (M / n)×(N / n) 2 regions, where n is the common divisor of M and N;

[0112] After local threshold binarization, the intersection of multiple consecutively shot images f1, f2 and f3 is taken to obtain a new binary image f4 = f1 & f2 & f3 to remove the artifacts caused by the local threshold;

[0113] And for each small area, an initial binary image is obtained to separate the local non-continuous deformation area in the image from the background. Then, a binary process is performed to convert the image after threshold segmentation into a binary image for further analysis and processing.

[0114]

[0115] Among them, T(x,y) is the local threshold.

[0116] S3.3 performs grayscale transformation to convert the color image into a grayscale image, makes important features more prominent by improving the contrast and brightness of the image, and adjusts the pixel value distribution of the image to increase the contrast of the image;

[0117] According to the characteristic that the area of ​​noise is much smaller than the area of ​​shear band, the maximum noise area interval is determined to obtain the optimal area threshold, and then the value in this interval is taken as the area threshold for de-noising operation, and then smoothing processing is performed using Gaussian filtering to reduce the noise in the image and retain the main features.

[0118] The grayscale transformation formula is:

[0119] I g =0.299×R+0.587×G+0.114×B;

[0120] Among them, R, G, and B represent the values ​​of the red, green, and blue channels of the color image respectively.

[0121] cv2.cvtColor(image,cv2.COLOR_BGR2GRAY) converts a color image to a grayscale image, simplifying the processing steps.

[0122] cv2.GaussianBlur(gray_image,(5,5),1.4) applies a Gaussian filter to smooth the image, reducing noise and retaining the main features.

[0123] cv2.threshold(blurred_image,128,255,cv2.THRESH_BINARY) sets the pixel values ​​in the grayscale image above the threshold to white (255) and the pixel values ​​below the threshold to black (0).

[0124] The cv2.equalizeHist(image) function implements the above process.

[0125] S4. After preprocessing, the image is segmented and feature extracted to identify and extract the local discontinuous deformation area. This process mainly includes two processes: segmenting the image into different areas by setting a threshold, extracting the local discontinuous deformation, and using the Canny edge detection algorithm to detect the edge of the local discontinuous deformation.

[0126] S4.1The gradient calculation formula for Canny edge detection is:

[0127] Gradient detection formula:

[0128]

[0129] Among them, G x and G y are the gradients in the horizontal and vertical directions respectively, and G is the gradient strength.

[0130] cv2.threshold(image,threshold,255,cv2.THRESH_BINARY);

[0131] cv2.Canny(binary_image,100,200);

[0132] cv2.morphologyEx(binary_image,cv2.MORPH_CLOSE,kernel) Segmentation extraction is implemented by the above function.

[0133] S4.2 Due to the lighting effect when the shear band is photographed, the image processed by the above operation may still show discontinuity in the original continuous shear band image. In further recognition, the system will identify it as multiple shear bands, so the shear bands must be bridged. The noise in the image is removed through morphological operations, and the clear outline of the shear band is extracted;

[0134] cv2.dilate(binary_image,kernel,iterations=1) expands the white area in the image and fills the small holes in the clipping band;

[0135] cv2.erode(dilated_image,kernel,iterations=1) removes noise points in the image and retains the main shear band features;

[0136] cv2.morphologyEx(binary_image,cv2.MORPH_CLOSE,kernel) performs a closing operation, first dilating and then corroding, to fill the small holes in the shear band.

[0137] S4.3 Since the physical size of the noise point and the shear band are similar, the opening operation causes some shear bands to be excessively corroded and disappear. The scan line algorithm is now used to record the intersection of the current scan line and the polygon edge by scanning the image or graphic line by line, and dynamically update the intersection information when the scan line moves. The specific implementation steps are as follows:

[0138] Enter the initialized edge table (ET) to traverse all the edges of the polygon, calculate the inverse slope of each edge (slope_inverse), and sort them according to the y coordinate of the starting point of the edge. It should be noted that horizontal edges are not counted in the edge table;

[0139] Initialize the Active Edge Table (AET) Before the scan line starts, the Active Edge Table (AET) is empty;

[0140] Perform a line-by-line scan, starting from the minimum y coordinate of the polygon and scanning line by line to the maximum y coordinate of the polygon;

[0141] Update the AET on the current scan line, delete the edges that are no longer active, and add new active edges;

[0142] Calculate the intersection points of the current scan line with all edges in the AET, sorted by x coordinates. Fill the polygons based on the intersection points, taking the area between the intersection points in pairs;

[0143] Update the edges in AET, mainly update the x-coordinates of the intersection points, and delete the edges that are no longer active;

[0144] Repeat the above steps until the scan is complete.

[0145] The above four operations are selected and used according to the different feature forms of different parts of the image.

[0146] S5. Identify the nodes of the shear band network in the image, obtain the trunk composed of the shear bands and the pixels contained in each shear band, and obtain the nodes of the shear band network on the basis of skeletonization, and trace the shear band lines on the trunk starting from the nodes.

[0147] Detect shear bands to extract the area, perimeter and shape parameters of local discontinuous deformations;

[0148] Area: Area = Σ i,j I(i,j);

[0149] Perimeter: Use the count of edge pixels or calculate the contour length;

[0150] The texture features and deformation features of local discontinuous deformation are analyzed, the segmented local discontinuous deformation image is processed, the location, shape and evolution process of local discontinuous deformation are determined, and the instability and failure form of soil is estimated.

[0151] S5.1 First, the skeletonization algorithm can be used to extract the main structure of the shear band. The skeletonization operation refines the shear band to only one pixel wide, retaining the basic framework of the shear band. Through this process, the skeleton line of the shear band is obtained;

[0152] S5.2 Node and shear band identification is based on the shear band trunk. For each trunk point, count the number of transitions from black to white in the eight adjacent pixels in a counterclockwise direction. When N = 1, the point is an endpoint; when N = 2, it is a general trunk point; when N ≥ 3, it is an intersection point. In some special cases, such as the node where four shear bands intersect, multiple nodes may be generated after skeletonization. Therefore, after the node search, it is necessary to calculate the centroid of these close nodes and select the trunk point closest to the centroid as the formal node.

[0153] S5.3 Any shear band trunk connects two nodes. In the skeletonized image, starting from a node, extend and search along the trunk, record the trunk pixels found, until another node is found, so as to determine the trunk of a shear band. For the remaining non-trunk shear band pixels, merge them to the nearest shear band trunk connected to them through the merging algorithm, and finally form a complete shear band.

[0154] The number of pixels inside the shear zone according to the soil morphology in the image

[0155] Use cv2.findContours function to detect the contour of the shear zone;

[0156] Use cv2.contourArea function to calculate the area of ​​the contour;

[0157] Use cv2.arcLength function to calculate the perimeter of the contour;

[0158] Finally, the binary image is read through cv2.imread('binary_image.jpg',0) and the geometric features are extracted.

[0159] S6. Use the subset matching algorithm to find the correspondence between the target sub-areas before and after deformation, and then calculate the displacement vector. The displacement of the center point of the target subset can be expressed as:

[0160] u=(u x ,u y );

[0161] where u x and u y are the displacements in the x and y directions respectively.

[0162] Based on the calculated displacement field, the displacement gradient (deformation gradient) of each target subset is calculated using the numerical differentiation method:

[0163]

[0164] in, are the displacement gradients of the subset in the x and y directions.

[0165] For small deformations, the strains can be calculated using the linear strain tensor:

[0166]

[0167] For large deformations, the Green-Lagrange strain tensor can be used:

[0168]

[0169] where E is the Green-Lagrange strain tensor and I is the identity matrix.

[0170] Strain calculation method is as follows Figure 7 shown.

[0171] S7, carrying out a three-dimensional reconstruction process of the shearing process based on geometric features;

[0172] Use SIFT algorithm to detect and match feature points of multi-view images to obtain corresponding feature point pairs;

[0173] S7.1 Based on the matched feature point pairs, use the cv2.findFundamentalMat(pts1,pts2,cv2.FM_RANSAC) function to calculate the basic matrix F, which is used to describe the geometric relationship between the two viewpoints. The intrinsic matrix E is calculated through the intrinsic matrix K:

[0174]

[0175] Use the essential matrix E and the camera projection matrices P1 and P2 to triangulate the feature points, use the cv2.triangulatePoints function to triangulate the feature points, reconstruct the three-dimensional points, and reconstruct the three-dimensional coordinates X;

[0176] S7.2 converts homogeneous coordinates into three-dimensional coordinates, creates a point cloud object using Open3D, and calculates the normal of the point cloud for surface reconstruction;

[0177] S7.3 uses the Poisson Surface Reconstruction algorithm to reconstruct the surface of point cloud data.

[0178]

[0179] Among them, φ is the implicit function of reconstruction, and v is the vector field generated by the point cloud normal vector.

[0180] S8. After the test, the system automatically generates a detailed experimental report, including the development curve of local discontinuous deformation and the analysis results of the failure morphology. The report should include the key steps of image processing, the analysis results of local discontinuous deformation characteristics, and the images and data of 3D reconstruction. The test data and processing results are stored in the data processing system for subsequent analysis and research. The data storage should include the original image, the processed image, the local discontinuous deformation feature data, and the 3D reconstruction results.

[0181] The specific implementation steps are as follows:

[0182] S1. Install two high-resolution cameras in front and on the side of the soil test equipment. These cameras need to be fixed on a stable bracket to ensure that they do not change position during the experiment. Adjust the position and angle of the camera to ensure that the shear process of the soil sample can be fully captured. Arrange LED light sources evenly around the experimental equipment and adjust the brightness and position of the lights to avoid shadows and reflections. Use a diffuser or reflector to further even out the light. Visual perception devices such as Figure 5 shown.

[0183] Connect the camera system to the image acquisition module to ensure that the image signal can be stably transmitted to the data processing system. The image acquisition module needs to be configured with an appropriate sampling rate and resolution to ensure image clarity and real-time performance. Connect the image acquisition module to the data processing system to ensure that the data processing system can receive and process real-time image data. The configuration of the data processing system needs to be optimized based on the test requirements and processing capabilities. Configure the control interface to establish a communication connection with the geotechnical test equipment to ensure coordination and synchronization during data acquisition and processing. The control interface needs to have a reliable communication protocol and interface standard.

[0184] S2. Before the test begins, start the data processing system and perform a system self-check to ensure that the connection between the camera, image acquisition module and data processing system is normal. Check whether the image acquisition module can receive and transmit image data in real time to confirm that the system is in working condition. During the unit shear test, two cameras will simultaneously capture the shear process of the soil sample in real time. Set the camera shooting frequency to 45 frames per second to ensure that the detailed changes in the shear process can be captured. The image is transmitted to the data processing system through the image acquisition module.

[0185] cv2.VideoCapture(camera_1);

[0186] cv2.VideoCapture(camera_2);

[0187] cv2.getTickCount() This process is implemented by two functions: capture and output;

[0188] S3. After installation, calibrate the camera accordingly. Install the calibration board as shown in the figure. Figure 2 As shown in the figure, fix the calibration plate in an appropriate position, ensure that the calibration plate is within the field of view of the camera, and take multiple images of the calibration plate from different angles and positions. Each image should cover different viewing angles and distances as much as possible to ensure the accuracy of calibration. It is recommended to take more than 10 images of the calibration plate. The sample measurement image is as follows: Figure 4 As shown. Then load the calibration board image and convert it to a grayscale image. Use the cv2.findChessboardCorners function to detect the corners on the calibration board. If a corner is detected, use the cv2.cornerSubPix function to optimize the sub-pixel accuracy. Store the detected corner coordinates in an array as the corner coordinates in the image coordinate system. The specific operation process is as follows Figure 3 shown.

[0189] The input image is screened to remove those images that do not meet the processing standards and ensure the quality of the image for subsequent processing. The image gradient value G is calculated to evaluate the image clarity. The gradient value reflects the degree of change in the pixel value in the image. The higher the gradient value, the clearer the image. If the gradient value G is greater than 40%, the image is considered clear, otherwise the image is deleted. At the same time, the image contrast c is calculated. The contrast reflects the degree of difference in light and dark in the image. The higher the contrast, the easier it is to distinguish different areas in the image. The image contrast c needs to be greater than 0.25. The screening process is as follows: Figure 6 shown.

[0190] S4, the real-time received image is first preprocessed by grayscale conversion of the input image, and then Gaussian blur is applied to reduce image noise. The kernel size of Gaussian blur is set to 5x5 and the standard deviation is set to 1.4, which helps to smooth the image while retaining important structural information. Then, adaptive thresholding technology is used to dynamically adjust the threshold by analyzing the brightness of the local area of ​​the image, thereby improving the sensitivity and accuracy of edge detection.

[0191] After threshold processing, the Canny algorithm is used for edge detection.

[0192] Gaussian Blur:

[0193] Where σ = 1.4, the kernel size is 5×5;

[0194] Perform histogram equalization on the image, which helps to enhance the contrast of the image and make low-contrast edges easier to detect. Histogram equalization improves edge detection by redistributing the brightness of the image so that the brightness distribution of the entire image is more uniform. The opening operation uses the same 3×3 square kernel as the previous closing operation, but dilates first and then erodes.

[0195] The low threshold of the Canny algorithm is set to 100 and the high threshold is set to 200 to further enhance the continuity and integrity of the edges. We also add a morphological closing operation and use a kernel with a structural element of 3×3 for processing.

[0196] S5. Erosion and dilation operations are performed on the processed image, which helps to eliminate small holes and small breaks in the image and ensure the integrity and continuity of the edge. The erosion operation uses the same 3x3 structure element and iterates once. The subsequent dilation operation also uses the same settings to restore the original width and shape of the edge, including grayscale transformation, image enhancement and smoothing. The preprocessed image will be used for subsequent segmentation and analysis. It is implemented by the following function:

[0197] cv2.cvtColor(image,cv2.COLOR_BGR2GRAY);

[0198] cv2.equalizeHist(gray);

[0199] cv2.GaussianBlur(enhanced, (5, 5), 0);

[0200] Perform threshold segmentation and edge detection on the preprocessed image to extract the local discontinuous deformation area. The extraction process is as follows: Figure 8 and Fig. 9 As shown. Then, morphological processing is performed to remove noise and extract effective local discontinuous deformation features. The segmented local discontinuous deformation image is analyzed to determine the location, morphology and evolution process of the local discontinuous deformation, estimate the soil failure morphology, and automatically extract the geometric and statistical features of the local discontinuous deformation.

[0201] cv2.imshow('Segmented Image', binary_image);

[0202] cv2.imshow('Morphological Processing', processed_image);

[0203] cv2.findContours(binary_image, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE);

[0204] S6. Perform three-dimensional reconstruction of the extracted features, use the SIFT algorithm to detect feature points in multi-view images, and match the same feature points in different view images by calculating the Euclidean distance between SIFT feature descriptors. Use the cv2.findFundamentalMat function to calculate the basic matrix F to describe the geometric relationship between the two viewpoints. Use the essential matrix E and the camera projection matrices P1 and P2 to triangulate the feature points, and use the cv2.triangulatePoints function to reconstruct the three-dimensional coordinates X of the feature points. Then use Open3D to create a point cloud object o3d.geometry.PointCloud(), add point data o3d.utility.Vector3dVector(points) and calculate the normal vector of the point cloud. Finally, use the PoissonSurface Reconstruction algorithm to reconstruct the surface of the point cloud data as follows Fig.10 shown.

[0205] The data processing system automatically generates a test report based on the image processing and feature analysis results. The report includes a test overview, image processing process, and local discontinuous deformation analysis results. The test report can be presented in text and chart form to provide detailed test data and analysis results. The test data is stored in the storage unit of the data processing system, including the original image, processed image, and local discontinuous deformation feature data. Data storage needs to ensure data integrity and security to facilitate subsequent analysis and research.

[0206] This method uses external visual recognition equipment to achieve real-time capture and analysis of local discontinuous deformation development and failure morphology in geotechnical tests. It uses advanced digital image processing technology, including grayscale transformation, image enhancement, smoothing, threshold segmentation, edge detection, morphological processing, geometric feature extraction and statistical feature analysis. At the same time, it combines multi-viewpoint fusion technology. This method provides an accurate and real-time means for geotechnical test research, greatly improving the efficiency of test data acquisition and analysis.

[0207] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for reconstructing the shear deformation process of a triaxial test specimen based on visual perception, characterized in that: The following steps are involved: Set up an external visual device and use a dot calibration plate to calibrate the camera in the external visual device; Using a calibrated camera to capture an experimental image during the shearing experiment, and preprocessing the experimental image to obtain a preprocessed image; Segmenting and extracting features from the preprocessed image to obtain a local discontinuous deformation region image; Identify the local discontinuous deformation area image, obtain each node of the shear band network, obtain the shear band trunk and the pixels contained in each shear band based on each node of the shear band network; and trace each shear band line along the shear band trunk based on skeletonization to obtain a complete shear band; Calculate the displacement gradient and strain tensor of rock and soil specimens during shearing based on the complete shear band; The shear process is reconstructed based on the displacement vector and the strain tensor to obtain a visualized rock and soil shear gradient process.

2. The method for reconstructing the shear deformation process of a triaxial test specimen based on visual perception according to claim 1 is characterized in that: The external visual equipment includes a high-resolution camera and LED lighting equipment; The high-resolution camera is used to capture images of the entire test area during the shearing process in real time; The LED lighting device is used to provide a stable light source.

3. The method for reconstructing the shear deformation process of a triaxial test specimen based on visual perception according to claim 1, characterized in that: The method for calibrating a camera in an external visual device comprises: Use the cv2.findChessboardCorners function in OpenCV to detect the corners on the calibration board. When the corners are detected, use the cv2.cornerSubPix function to optimize the sub-pixel accuracy. Based on the optimized corner point coordinates, use the cv2.calibrateCamera function to calculate the camera's intrinsic and extrinsic matrix to calibrate the camera. Use feature extraction algorithm to extract feature points in the experimental image and calculate its descriptor; perform feature matching on images with different monocular viewpoints to obtain feature point pairs; Estimating the feature point pair through relative posture to obtain the relative posture of the camera; Based on the relative postures of the cameras, the matched feature point pairs are triangulated to obtain the three-dimensional coordinates of the feature points in the three-dimensional space.

4. The method for reconstructing the shear deformation process of a triaxial test specimen based on visual perception according to claim 3 is characterized in that: The three-dimensional coordinates are solved by minimizing the reprojection error, and the solution expression is: minimize∑∥x i -π(P i ,X)∥ 2 ; Among them, X i Represents the observed three-dimensional coordinate point, P i Represents the three-dimensional point that needs to be projected onto the two-dimensional plane, π(P i ,X) represents the projection function, which projects the three-dimensional point X onto the image plane of the i-th view.

5. The method for reconstructing the shear deformation process of a triaxial test specimen based on visual perception according to claim 1, characterized in that: The method for preprocessing the experimental image includes: Determine the blur degree of the image through blur detection and remove blurry images; Adaptively set the threshold value according to the local features of the local discontinuous deformation in the remaining image; The residual image is converted into a grayscale image and smoothed.

6. The method for reconstructing the shear deformation process of a triaxial test specimen based on visual perception according to claim 5 is characterized in that: The method for smoothing includes: According to the characteristic that the area of ​​the noise points is much smaller than the area of ​​the shear band, the maximum noise point area interval is determined to obtain the optimal area threshold, the noise removal operation is performed using the optimal area threshold, and the image after the noise removal operation is processed using Gaussian filtering.

7. The method for reconstructing the shear deformation process of a triaxial test specimen based on visual perception according to claim 1, characterized in that: The method for segmenting and extracting features from the preprocessed image includes: Segment the local non-continuous deformation area image based on the adaptively set threshold; The edge of the image in the local non-continuous deformation area is obtained through Canny edge detection; The shear band is bridged, and the noise in the bridged image is removed by morphological operations, and the clear outline of the shear band is extracted; The scan line algorithm is used to update the intersection information.

8. The method for reconstructing the shear deformation process of a triaxial test specimen based on visual perception according to claim 1, characterized in that: The method for obtaining a complete shear band comprises: In the skeletonized image, starting from a node, the search is extended along the trunk, and the trunk pixels found are recorded until another node is found. The trunk of a shear band is determined based on the two nodes, and the remaining non-trunk shear band pixels are merged into the nearest connected shear band trunk through a merging algorithm to form a complete shear band.

9. The method for reconstructing the shear deformation process of a triaxial test specimen based on visual perception according to claim 1, characterized in that: The calculation expression of the displacement gradient is: Among them, u x and u y are the displacements in the x and y directions, respectively, are the displacement gradients of the subset in the x and y directions.

10. The method for reconstructing the shear deformation process of a triaxial test specimen based on visual perception according to claim 1, characterized in that: The calculation method of the strain tensor includes: When the deformation is small, the calculation is performed by the linear strain tensor; When large deformations occur the Green-Lagrange strain tensor is used for calculations.

Images