Intelligent Detection Method and System for Structural Deformation and Deviation Based on Machine Vision

Abstract

The present invention discloses an intelligent detection method and system for structural deformation and deviation based on machine vision, including obtaining working data and status data of a detection target, performing distortion processing on a camera image to obtain a corrected image, processing the corrected image to obtain target point coordinate information, constructing a three-dimensional model of the detection target according to the target point coordinate information, comparing the model to obtain basic structural deformation and deviation, obtaining first structural deformation and deviation and second structural deformation and deviation according to the status data, constructing an intelligent detection model for structural deformation and deviation, and inputting the working data and status data of the target to be detected into the intelligent detection model for structural deformation and deviation to obtain a detection result of structural deformation and deviation. This method can not only improve the accuracy of structural deformation and deviation detection, but also has good interpretability and can be directly applied to an intelligent detection system for structural deformation and deviation based on machine vision.

Description

Technical Field

[0001] The present invention relates to the field of non-destructive testing, and particularly to an intelligent detection method and system for structural deformation and deviation based on machine vision. Background Art

[0002] With the rapid development of buildings and infrastructure, the requirements for structural safety and stability are increasing day by day. The detection of structural deformation and deviation is of great significance for preventing catastrophic accidents and ensuring the safety of people's lives and property. At the same time, intelligent detection methods based on machine vision have gradually emerged with the rapid development of machine vision and artificial intelligence technologies, providing new solutions for the detection of structural deformation and deviation.

[0003] Most traditional detection methods rely on manual contact measurement and visual inspection, which have problems such as difficult application in large and high-rise buildings, time-consuming and laborious measurement, great influence of human factors, and ignoring the influence of environmental state factors on the detection results, resulting in limited accuracy and reliability of the detection results. In addition, most existing machine vision-based detection methods only stay at the level of two-dimensional image processing and are difficult to accurately reflect the true three-dimensional shape and deformation of the structure. By fully considering factors such as transmission influence, image conversion, and environmental state data, combining machine vision technology, image recognition technology, three-dimensional reconstruction technology, BIM technology, and machine learning technology, an accurate, efficient, and fast intelligent detection method and system for structural deformation and deviation based on machine vision are designed to overcome the deficiencies of existing structural deformation and deviation detection methods and systems, realize efficient and high-precision detection of structural deformation and deviation, and provide strong technical support for structural safety monitoring and evaluation in fields such as bridges, buildings, and machinery. Summary of the Invention

[0004] The object of the present invention is to provide an intelligent detection method and system for structural deformation and deviation based on machine vision.

[0005] To achieve the above object, the present invention is implemented according to the following technical solutions:

[0006] The present invention includes the following steps:

[0007] Obtain the working data and status data of the detection target, and preprocess the working data and the status data; the working data includes monitoring data and camera images; the status data includes environmental data and orientation data;

[0008] Perform distortion processing on the camera image to obtain a corrected image, perform image processing on the corrected image to obtain point cloud data, and obtain target point coordinate information based on the point cloud data and the corrected image;

[0009] Construct a three-dimensional model of the detection target based on the target coordinate information, establish a BIM model of the detection target, and compare the three-dimensional model and the BIM model to obtain the basic structural deformation and deviation;

[0010] Determine the first structural deformation and deviation according to the environmental data and the basic structural deformation and deviation, and determine the second structural deformation and deviation according to the orientation data and the basic structural deformation and deviation;

[0011] Construct an intelligent detection model of structural deformation and deviation based on the structural deformation and deviation and the monitoring data, and input the working data and status data of the target to be detected into the intelligent detection model of structural deformation and deviation to obtain the detection result of structural deformation and deviation.

[0012] Further, a method for performing distortion processing on the camera image to obtain a corrected image includes:

[0013] Use the camera distortion model to correct the camera image coordinates, and the expression is:

[0014] x corr = x(1 + k1r 2 + k2r 4 + k3r 6 ) + 2p1xy + p2(r 2 + 2x 2 ) + δ - μ / x

[0015] y corr = y(1 + k1r 2 + k2r 4 + k3r 6 ) + p1(r 2 + 2x 2 ) + 2p2xy + δ - μ / y

[0016] where (x corr , y corr ) is the corrected position, (x, y) is the position before correction, k1, k2, and k3 are radial distortion correction parameters, p1 and p2 are tangential distortion correction parameters, δ is the orientation correction mean, and μ is the edge distortion value;

[0017] Calibrate the camera parameters based on the OpenCV function to obtain the camera internal parameter matrix, tangential distortion correction parameters, and radial distortion correction parameters;

[0018] Use a checkerboard calibration board for calibration: capture calibration board images at multiple positions and angles, detect corner points to obtain rough positioning, perform sub-pixel processing to obtain corner point coordinates, establish the relationship between chessboard points and image points to obtain the internal parameter calibration result and distortion correction parameters, and correct the image to obtain a corrected image.

[0019] Further, the method for obtaining the target coordinate information includes:

[0020] Load the depth image of the corrected image, extract the depth image pixel values, and convert the depth image into point cloud data corresponding to three-dimensional coordinates according to the depth camera-detection target distance, image pixel values, and the camera internal parameter matrix;

[0021] Construct a CNN machine learning model, input the corrected image and the point cloud data into the CNN machine learning model, use the CNN machine learning model to perform image recognition and feature extraction on the corrected image to obtain the image coordinates of the target target, use the CNN machine learning model to perform data preprocessing and data enhancement on the point cloud data, and determine the coordinate information of the target based on the matching of the point cloud data and the image coordinates by the CNN machine learning model.

[0022] Further, the method for constructing a three-dimensional model of the detection target according to the target coordinate information includes:

[0023] Combine the theodolite turning point calculation algorithm and the dynamic local window algorithm, and determine the target coordinates on the connecting traverse according to the coordinate information of the single observation point target output by the CNN machine learning model;

[0024] Construct a spatial index structure for the point cloud data near each target, including: calculating the normal vector of the point cloud data, calculating the vector space, inversely solving the Poisson equation through the gradient relationship, performing isosurface extraction and reconstructing the point cloud model of the measured object, and the expression is:

[0025]

[0026] U = a1u 2 + a2uv + a3v 2 + a4u + a5v + a6

[0027] c g = (4a1a3 - a2 2 ) / (a4 2 + a5 2 + 1) 2

[0028]

[0029] where is the vector space of the target q, t is the point in the area near the target, p is an arbitrary point, is different surface areas, is the smoothing function, is the inward surface normal vector, U(u, v) is the quadratic surface equation, a1, a2, a3, a4, a5, a6 are the coefficients of the quadratic surface equation, c g is the Gaussian curvature, is the mean curvature;

[0030] Perform meshing on the point cloud model of the object to be measured, including: constructing seed triangles and extracting boundary edges, determining adaptive neighborhood points according to the average distance density, calculating projection coordinates, and determining the optimal expansion points based on the projection coordinates. The expressions are as follows:

[0031] Z: n x (X - P x ) + n y (Y - P y ) + n z (Z - P z ) = 0

[0032] Q t3 = [cosθ·I + (1 - cosθ)kk T + sinθ·ROK]·Q t2

[0033] where Z is the point cloud coordinate P (P t1 , P x , P y , P z ) of the midpoint of the inner boundary edge in the neighborhood Q and the corresponding normal vector n P (n x , n y , n z ) of the tangent plane equation, Q t2 (X, Y, Z) is the projection coordinate of the tangent plane, Q t3 is the rotated coordinate after rotation of the projection coordinate, I is the 3D identity matrix, θ is the angle between the normal vector n P and the normal vector of the XOY plane, k is the rotation axis, which is the cross product of the normal vector n P and the normal vector of the XOY plane;

[0034] Perform hole repair on the point cloud model of the object to be measured to obtain the three-dimensional model of the detection target, including: hole boundary recognition, determination of the hole boundary direction, addition of vertices, addition of vertices according to the vector sum and curvature of the normal vector, addition of vertex inspection, and mesh smoothing processing. The expressions are as follows:

[0035]

[0036] 1 / σ + 1 / ξ = ρ, σ > 0, ξ < -ξ

[0037] where N n is the normal vector of the newly added mesh vertex, N l is the normal vector of the adjacent triangle of the newly added mesh vertex, g is the number of adjacent triangles in the vertex one-ring, λ m is the shape factor of the adjacent triangle m, B m is the centroid distance of the adjacent triangle m, P nFor the newly added mesh vertex, P n ” is the newly added mesh vertex after mesh smoothing, σ is the positive influence factor in the contraction process, ξ is the negative influence factor in the stretching process, P u 、P v is a one-ring neighborhood, P u ', P v ' is the newly added mesh vertex P n The one-ring neighborhood after contraction, h is the number of one-ring neighborhoods.

[0038] Furthermore, the method for obtaining the basic structure deformation and deviation includes:

[0039] Establish a BIM model of the detection target, and compare the three-dimensional model and the BIM model to obtain the basic deformation and deviation

[0040] Import the building information of the detection target into the REVIT software to construct a BIM model of the detection target;

[0041] Register the three-dimensional model and the BIM model of the detection target through grid nodes, and determine the relationship between the deviation mode quantity by comparing the grid sizes and model reconstruction accuracies of each target point. The expression is:

[0042] f(m,n,o) = f P (m,n,o) - f BIM (m,n,o)

[0043] where f(m,n,o) is the deviation field of the detection target, m, n, o are the numbers of the target points in the x, y, and z directions, f P (m,n,o) is the measured value of the target point on the three-dimensional model, f BIM (m,n,o) is the theoretical value of the target point on the BIM model;

[0044] Perform a forward DCT transformation on the detection target deviation field to obtain an orthogonal transformation coefficient matrix. The expression is:

[0045]

[0046] where C(μ,v,ω) is the orthogonal transformation coefficient matrix, M, N, O are the total number of sampling points in the x, y, and z directions, and μ, v, ω are frequency values;

[0047] Perform key mode identification of the target deviation field according to the orthogonal transformation coefficient matrix C(μ,v,ω), and select important deviation data according to the energy compaction degree and contribution degree of the deviation field. The expression is:

[0048]

[0049] where E i is the energy compaction degree of the important deviation mode, Ci 2 (μ, v, ω) is the energy of the important deviation mode, C i (μ, v, ω) ∈ Ω, where Ω is the mode set, η i is the contribution degree of the important deviation mode;

[0050] Select the deviation modes with energy compactness and contribution degree greater than the set threshold to form the key orthogonal transformation coefficient matrix For the orthogonal transformation coefficient matrix Perform the inverse DCT transformation to obtain the key deviation field of the detection target. The expression is:

[0051]

[0052] where is the key deviation field of the detection target. Determine the basic structure deformation and deviation according to the key deviation field of the detection target.

[0053] Furthermore, the method for determining the first structure deformation and deviation includes:

[0054] Input the environmental data into the environmental impact function to obtain the environmental impact factor. Determine its first deviation according to the environmental impact factor and the basic deviation of the coordinates at each target point. Determine its first structure deformation according to the first deviation of the coordinates at the overall target points of the detection target. The expression is:

[0055]

[0056] where Effect1 is the environmental impact factor, w1, w2, w3, w4 are the weights of the environmental impact factor, is the average illumination, L Rm 、L Gm 、L Bm are the measured values of red, green, and blue light in the environment, L Rs 、L Gs 、L Bs are the standard values of red, green, and blue light in the environment, T m is the working temperature of the camera, T s is the standard working temperature of the camera, H m is the measured value of atmospheric humidity, H s is the standard value of atmospheric humidity, N m is the measured value of environmental noise, N s is the acceptable threshold of environmental noise, Deviation1 is the first deviation of the coordinates at the overall target points of the detection target, Deviation0 is the basic deviation of the coordinates at the overall target points of the detection target, P n ” is the target point coordinate after mesh fairing, and P0 is the target point coordinate in the BIM model.

[0057] Further, the method for determining the second structural deformation and deviation includes:

[0058] Construct a scale influence factor according to the orientation data, determine its second deviation according to the scale influence factor and the basic deviation of the coordinates at each target point, and determine its second structural deformation according to the second deviation of the coordinates at the overall target points of the detection target. The expression is:

[0059]

[0060] where Effect2 is the scale influence factor, w5, w6, w7 are the weights of the scale influence factor, D is the distance from the camera to the detection target, S BIM is the maximum value of the length, width, and height in the BIM model of the detection target, S0 is the standard size detected by the camera, is the inclination angle between the camera and the detection object, is the rotation angle between the camera and the detection object, Deviation2 is the second deviation of the coordinates at the overall target points of the detection target, and w8 is the weight of the second structural deviation.

[0061] Further, the method for obtaining the detection results of the structural deformation and deviation includes:

[0062] Measure the structural deformation and deviation of the detection target to obtain the actual structural deformation and deviation;

[0063] Form a comprehensive detection data set with the actual structural deformation and deviation, the first structural deformation and deviation, the second structural deformation and deviation, the environmental influence factor, and the scale influence factor, and divide the comprehensive detection data set into a training set and a test set;

[0064] Construct an intelligent detection model for structural deformation and deviation. The intelligent detection model for structural deformation and deviation includes an attention layer, a feature fusion layer, and a BP neural network;

[0065] The attention layer extracts the main features of the training set data through two parallel SE attention mechanism channels, and learns the dependence relationship between the features to obtain the required feature data. The expression is:

[0066]

[0067] where F' Deviation is the required feature data matrix, σ(·) is the sigmoid function used to compress the numerical value between (0,1) to generate the weight coefficient, W9, W 10 is the weight matrix of the fully connected layer used to compress and restore the number of channels C, δ(·) is the non-linear activation function, I, J are the row and column dimensions of the feature data matrix, f c (i,j) is the value of the c-th channel of the feature data matrix at the position (i,j), F DeviationIt is the feature data matrix formed after the extraction of input data features;

[0068] The feature fusion layer performs probability calculation and feature fusion on the data processed by the attention layer through the first classification layer and the second classification layer, and outputs the fusion result of the structural deformation and deviation of the detection target. The cross-entropy loss function is used to optimize the classification result of the first classification layer, and the MSE function is used to optimize the fusion result of the second classification layer;

[0069] The BP neural network is used to learn the relationship between the fusion result of the structural deformation and deviation of the detection target and the actual structural deformation and deviation, environmental impact factors, and scale impact factors, and performs regression prediction to output the detection result of the structural deformation and deviation of the detection target. The MAE loss function is used to uniformly penalize the prediction error of the model, and the RMSProp optimizer is used to automatically adjust the learning rate of the model;

[0070] The test set is used to evaluate the intelligent detection model of structural deformation and deviation;

[0071] The working data and status data of the target to be detected are input into the intelligent detection model of structural deformation and deviation to obtain the detection result of structural deformation and deviation.

[0072] In the second aspect, an intelligent detection system for structural deformation and deviation based on machine vision includes:

[0073] Data acquisition module: including a high-resolution binocular camera, a target, and a measurement track; used to obtain the working data and status data of the detection target, and preprocess the working data and the status data;

[0074] Image module: used to perform distortion processing on the camera image to obtain a corrected image, perform image processing on the corrected image to obtain point cloud data, and obtain target point coordinate information based on the point cloud data and the corrected image;

[0075] BIM model module: used to establish a BIM model of the detection target according to building information;

[0076] Data processing module: used to construct a three-dimensional model of the detection target according to the target point coordinate information, used to compare the three-dimensional model and the BIM model to obtain the basic structural deformation and deviation, used to construct environmental impact factors and scale impact factors according to environmental data and azimuth data, and used to determine the first structural deformation and deviation and the second structural deformation and deviation according to the environmental impact factors, the scale impact factors, and the basic structural deformation and deviation;

[0077] Detection model module: It is used to construct an intelligent detection model for structural deformation and deviation based on the first structural deformation and deviation, the second structural deformation and deviation, the environmental impact factor, the scale impact factor, and the actual structural deformation and deviation, perform feature fusion and regression prediction, and input the working data and status data of the target to be detected into the intelligent detection model for structural deformation and deviation to obtain the detection result of structural deformation and deviation;

[0078] Intelligent supervision module: It is used to store, view, and manage the working data, the status data, and the detection result of structural deformation and deviation.

[0079] The beneficial effects of the present invention are:

[0080] The present invention is an intelligent detection method and system for structural deformation and deviation based on machine vision. Compared with the prior art, the present invention has the following technical effects:

[0081] Through image distortion processing, coordinate recognition, three-dimensional model reconstruction, construction of environmental impact factors and scale impact factors, correction of structural deformation and deviation, and model construction steps, the present invention can improve the accuracy of structural deformation and deviation detection, thereby improving the efficiency and precision of structural deformation and deviation detection based on machine vision. The intelligent detection of structural deformation and deviation can greatly save resources and improve the detection efficiency of structural deformation and deviation. It can realize the measurement of structural deformation and deviation of the target to be detected, quickly perform high-precision detection of structural deformation and deviation on the target to be detected, provide strong technical support for the construction field, and is of great significance for realizing ultra-high-precision non-contact measurement of civil engineering structures. It can adapt to different intelligent detection systems for structural deformation and deviation based on machine vision and the intelligent detection requirements of different users for structural deformation and deviation based on machine vision, and has a certain universality. Description of the Drawings

[0082] Figure 1 It is a flowchart of the steps of the intelligent detection method for structural deformation and deviation based on machine vision of the present invention. Detailed Embodiments

[0083] The present invention will be further described below through specific embodiments. The illustrative embodiments and descriptions of the present invention are used to explain the present invention, but do not limit the present invention.

[0084] The intelligent detection method and system for structural deformation and deviation based on machine vision of the present invention include the following steps:

[0085] As Figure 1 shown, in this embodiment, it includes the following steps:

[0086] Obtain the working data and status data of the detection target, and preprocess the working data and the status data; the working data includes monitoring data and camera images; the status data includes environmental data and azimuth data;

[0087] Perform distortion processing on the camera image to obtain a corrected image, perform image processing on the corrected image to obtain point cloud data, and obtain target point coordinate information based on the point cloud data and the corrected image;

[0088] Construct a three-dimensional model of the detection target based on the target point coordinate information, establish a BIM model of the detection target, and compare the three-dimensional model and the BIM model to obtain basic structural deformation and deviation;

[0089] Determine the first structural deformation and deviation based on the environmental data and the basic structural deformation and deviation, and determine the second structural deformation and deviation based on the azimuth data and the basic structural deformation and deviation;

[0090] Construct an intelligent detection model for structural deformation and deviation based on the structural deformation and deviation and the monitoring data, and input the working data and status data of the target to be detected into the intelligent detection model for structural deformation and deviation to obtain the detection result of structural deformation and deviation.

[0091] In this embodiment, the method for performing distortion processing on the camera image to obtain a corrected image includes:

[0092] Use the camera distortion model to correct the camera image coordinates, and the expression is:

[0093] x corr = x(1 + k1r 2 + k2r 4 + k3r 6 ) + 2p1xy + p2(r 2 + 2x 2 ) + δ - μ / x

[0094] y corr = y(1 + k1r 2 + k2r 4 + k3r 6 ) + p1(r 2 + 2x 2 ) + 2p2xy + δ - μ / y

[0095] where (x corr , y corr ) is the corrected position, (x, y) is the position before correction, k1, k2, and k3 are radial distortion correction parameters, p1 and p2 are tangential distortion correction parameters, δ is the directional correction mean, and μ is the edge distortion value;

[0096] Calibrate the camera parameters based on OpenCV functions to obtain the camera internal parameter matrix, tangential distortion correction parameters, and radial distortion correction parameters;

[0097] Use a checkerboard calibration board for calibration: capture calibration board images at multiple positions and angles, detect corner points to obtain rough positioning, perform sub-pixel processing to obtain corner point coordinates, establish the relationship between checkerboard points and image points to obtain the internal parameter calibration results and distortion correction parameters, and correct the image to obtain the corrected image;

[0098] In the actual evaluation, conduct structural deformation and deviation detection on a stadium applying a steel structure-concrete beam composite structure, and obtain the image coordinates of key target points in the concrete beam image through a binocular camera: 1. (800, 600); 2. (1200, 400); 3. (500, 800); 4. (1500, 700); 5. (1000, 1000); 6. (1300, 500); 7. (700, 1100); 8. (1600, 300); 9. (900, 1200); 10. (1100, 600);

[0099] The internal parameter matrix of the camera is [718.856, 0, 960.5; 0, 718.856, 540; 0, 0, 1], the distortion coefficients are [-0.2635, 0.05159, -0.0001, 0, 0], and the image coordinates of the key target points after distortion processing are: 1. (802, 598); 2. (1203, 398); 3. (498, 802); 4. (1500, 700); 5. (1000, 1000); 6. (1300, 500); 7. (700, 1100); 8. (1600, 300); 9. (900, 1200); 10. (1100, 600).

[0100] In this embodiment, the method for obtaining the target point coordinate information includes:

[0101] Load the depth image of the corrected image, extract the depth image pixel values, and convert the depth image into point cloud data corresponding to three-dimensional coordinates according to the depth camera-detection target distance, image pixel values, and the camera internal parameter matrix;

[0102] Construct a CNN machine learning model, input the corrected image and point cloud data into the CNN machine learning model, use the CNN machine learning model to perform image recognition and feature extraction on the corrected image to obtain the image coordinates of the target point, use the CNN machine learning model to perform data preprocessing and data enhancement on the point cloud data, and determine the coordinate information of the target point based on the matching of the point cloud data and the image coordinates by the CNN machine learning model;

[0103] In the actual evaluation, image recognition is performed on the corrected image, and in combination with the distance between the camera and the detection target, the coordinate information of the target points is obtained: Top steel structure: 1. (205, -2, 10009); 2. (1503, 1, 10006); 3. (3498, 2, 10003); 4. (503, -1, 10002); 5. (4497, 1, 10007); Lower concrete beam: 1. (2499, -1501, 7001); 2. (1501, -1499, 6999); 3. (3499, -1502, 7003); 4. (502, -1500, 6998); 5. (4499, -1501, 7002).

[0104] In this embodiment, the method for constructing a three-dimensional model of the detection target according to the target point coordinate information includes:

[0105] Combining the measuring instrument turning point calculation algorithm and the dynamic local window algorithm, determining the target point coordinates on the connecting traverse according to the coordinate information of the single observation point target output by the CNN machine learning model;

[0106] Using a kd-tree to construct a spatial index structure for the point cloud data near each target point, including: using the nearest neighbor search to calculate the normal vector of the point cloud data, calculating the vector space, inversely solving the Poisson equation through the gradient relationship, using the stereocube algorithm for isosurface extraction and reconstructing the point cloud model of the object under test, and the expression is:

[0107]

[0108] U = a1u 2 + a2uv + a3v 2 + a4u + a5v + a6

[0109] c g =(4a1a3 - a2 2 ) / (a4 2 + a5 2 + 1) 2

[0110]

[0111] where is the vector space of the target point q, t is the point in the area near the target point, p is an arbitrary point, are different surface areas, is the smoothing function, is the inward surface normal vector, U(u, v) is the quadratic surface equation, a1, a2, a3, a4, a5, a6 are the coefficients of the quadratic surface equation, c g is the Gaussian curvature, is the mean curvature;

[0112] Perform meshing on the point cloud model of the object under test, including: constructing seed triangles and extracting boundary edges, determining adaptive neighborhood points according to the average distance density, calculating projection coordinates, and determining the optimal expansion points based on the projection coordinates. The expressions are as follows:

[0113] Z: n x (X - P x ) + n y (Y - P y ) + n z (Z - P z ) = 0

[0114] Q t3 = [cosθ·I + (1 - cosθ)kk T + sinθ·ROK]·Q t2

[0115] Where Z is the point cloud coordinate P(P t1 , P x , P y , P z ) of the midpoint of the inner boundary edge in the neighborhood Q and the corresponding normal vector n P (n x , n y , n z ) of the tangent plane equation, Q t2 (X, Y, Z) is the projection coordinate of the tangent plane, Q t3 is the rotated coordinate after the projection coordinate is rotated, I is the 3D identity matrix, θ is the angle between the normal vector n P and the normal vector of the XOY plane, k is the rotation axis, which is the cross product of the normal vector n P and the normal vector of the XOY plane;

[0116] Perform hole repair on the point cloud model of the object under test to obtain the three-dimensional model of the detection target, including: hole boundary recognition, determination of the hole boundary direction, addition of vertices, addition of vertices according to the vector sum and curvature of the normal vector, addition of vertex inspection, and mesh smoothing processing. The expressions are as follows:

[0117]

[0118] 1 / σ + 1 / ξ = ρ, σ > 0, ξ < -ξ

[0119] Where N n is the normal vector of the newly added mesh vertex, N l is the normal vector of the adjacent triangle of the newly added mesh vertex, g is the number of adjacent triangles in the vertex one-ring, λ m is the shape factor of the adjacent triangle m, B m is the centroid distance of the adjacent triangle m, P n is the newly added mesh vertex, Pn ” is the newly added grid vertex after grid fairing, σ is the positive influence factor in the contraction process, ξ is the negative influence factor in the stretching process, P u 、P v is a ring neighborhood, P u ', P v ' is the newly added grid vertex P n 's ring neighborhood after contraction, and h is the number of ring neighborhoods.

[0120] In this embodiment, the method for obtaining the basic structure deformation and deviation includes:

[0121] Establish a BIM model of the detection target, and compare the three-dimensional model and the BIM model to obtain the basic deformation and deviation

[0122] Import the building information of the detection target into the REVIT software to construct a BIM model of the detection target;

[0123] Register the three-dimensional model and the BIM model of the detection target through grid nodes, and determine the deviation mode quantity relationship by comparing the grid sizes and model reconstruction accuracies of each target point. The expression is:

[0124] f(m,n,o) = f P (m,n,o) - f BIM (m,n,o)

[0125] where f(m,n,o) is the deviation field of the detection target, m, n, o are the numbers of the target points in the x, y, z directions, f P (m,n,o) is the measured value of the target point on the three-dimensional model, and f BIM (m,n,o) is the theoretical value of the target point on the BIM model;

[0126] Perform a forward DCT transformation on the deviation field of the detection target to obtain an orthogonal transformation coefficient matrix. The expression is:

[0127]

[0128] where C(μ,v,ω) is the orthogonal transformation coefficient matrix, M, N, O are the total number of sampling points in the x, y, z directions, and μ, v, ω are frequency values;

[0129] Perform key mode identification of the target deviation field according to the orthogonal transformation coefficient matrix C(μ,v,ω), and select important deviation data according to the energy compaction degree and contribution degree of the deviation field. The expression is:

[0130]

[0131] where E i is the energy compaction degree of the important deviation mode, C i2 (μ, v, ω) is the energy of the important deviation mode, C i (μ, v, ω) ∈ Ω, where Ω is the mode set, η i is the contribution degree of the important deviation mode;

[0132] Select the deviation modes with energy compactness and contribution degree greater than the set threshold to form the key orthogonal transformation coefficient matrix For the orthogonal transformation coefficient matrix Perform the inverse DCT transformation to obtain the key deviation field of the detection target, and the expression is:

[0133]

[0134] where is the key deviation field of the detection target, and the basic structure deformation and deviation are determined according to the key deviation field of the detection target;

[0135] In the actual evaluation, combine the multi-measurement binocular camera data to determine the target point coordinate information on the attached traverse. According to the target point coordinate information, perform three-dimensional information reconstruction to obtain the three-dimensional model of the detection target. At the same time, construct a BIM model. The original coordinates of the model corresponding to the target points are: For the top steel structure: 1. (2500, 0, 10000); 2. (1500, 0, 10000); 3. (3500, 0, 10000); 4. (500, 0, 10000); 5. (4500, 0, 10000); For the lower concrete beam: 1. (2500, -1500, 7000); 2. (1500, -1500, 7000); 3. (3500, -1500, 7000); 4. (500, -1500, 7000); 5. (4500, -1500, 7000);

[0136] Compare the three-dimensional model of the detection target with the BIM model to determine the coordinate deviation at the target points: For the top steel structure: 1. (5, -2, 9), 2. (3, 1, 6), 3. (-2, 2, 3), 4. (3, -1, 2), 5. (-3, 1, 7); For the lower concrete beam: 1. (-1, -1, 1), 2. (1, 1, -1), 3. (-1, -2, 2), 4. (2, 0, -2), 5. (-2, -1, 3). The target point coordinate deviation reflects that the concrete beam only has minor deformations, and the deformations in all directions are within 2 mm. The steel structure support on the upper part of the concrete beam has relatively small deformations, with minor deformations in the horizontal direction and a maximum vertical deformation of 9 mm. At the same time, the deformation at the mid-span of the steel structure is greater than that at both ends.

[0137] In this embodiment, the method for determining the first structure deformation and deviation includes:

[0138] Input environmental data into the environmental impact function to obtain the environmental impact factor. Determine its first deviation based on the environmental impact factor and the basic deviation of the coordinates at each target point. Determine its first structural deformation based on the first deviation of the coordinates at the overall target points of the detection target. The expression is as follows:

[0139]

[0140] Where Effect1 is the environmental impact factor, and w1, w2, w3, w4 are the weights of the environmental impact factor. is the average illumination, L Rm 、L Gm 、L Bm are the measured values of red, green, and blue light in the environment, L Rs 、L Gs 、L Bs are the standard values of red, green, and blue light in the environment, T m is the working temperature of the camera, T s is the standard working temperature of the camera, H m is the measured value of atmospheric humidity, H s is the standard value of atmospheric humidity, N m is the measured value of environmental noise, N s is the acceptable threshold of environmental noise. Deviation1 is the first deviation of the coordinates at the overall target points of the detection target, Deviation0 is the basic deviation of the coordinates at the overall target points of the detection target, P n ” is the target point coordinate after grid fairing, and P0 is the target point coordinate in the BIM model;

[0141] In the actual evaluation, the environmental data is obtained: the ambient RGB (245, 250, 253) when the camera is working, the ambient standard value RGB (255, 255, 255), the working temperature of the camera is 31 °C, the standard working temperature of the camera is 25 °C, the humidity in the stadium is 60%, the standard atmospheric humidity is 45%, the measured value of the environmental noise is 65 dB, the acceptable noise for the camera to work is 45 dB, w1 = 0.5, w2 = 0.6, w3 = 0.4, w4 = 0.5. According to the environmental data, Effect1 = 0.177 is calculated, and the first structural deviation of the corresponding target points is: for the upper steel structure: 1. (7.02, -2.808, 12.636), 2. (4.212, 1.404, 8.424), 3. (-2.808, 2.808, 4.212), 4. (4.212, -1.404, 2.808), 5. (-4.212, 1.404, 9.828); for the lower concrete beam: 1. (-1.404, -1.404, 1.404), 2. (1.404, 1.404, -1.404), 3. (-1.404, -2.808, 2.808), 4. (2.808, 0, -2.808), 5. (-2.808, -1.404, 4.212).

[0142] In this embodiment, the method for determining the second structural deformation and deviation includes:

[0143] Construct a scale influence factor according to the azimuth data, determine its second deviation according to the scale influence factor and the basic deviation of the coordinates at each target point, and determine its second structural deformation according to the second deviation of the coordinates at the overall target points of the detection target. The expression is:

[0144]

[0145] where Effect2 is the scale influence factor, w5, w6, w7 are the weights of the scale influence factor, D is the distance from the camera to the detection target, S BIM is the maximum value of the length, width, and height in the BIM model of the detection target, S0 is the standard size detected by the camera, is the inclination angle between the camera and the detection object, is the rotation angle between the camera and the detection object, Deviation2 is the second deviation of the coordinates at the overall target points of the detection target, and w8 is the weight of the second structural deviation;

[0146] In the actual evaluation, the orientation data is obtained: the shooting distance between the camera and the lower concrete beam is 5000mm, the shooting rotation angle is 30°, the shooting inclination angle is 5°, the shooting distance between the camera and the upper steel structure is 15000mm, the shooting rotation angle is 45°, the shooting inclination angle is -25°, w5 = 0.5, w6 = 0.5, w7 = 0.1, w8 = 0.1. According to the environmental data, the Effect2 of the upper steel structure and the lower concrete are calculated to be 1.142 and 0.819 respectively. The second structural deviations of the corresponding target points are: for the upper steel structure: 1. (5.688, -1.763, 10.395), 2. (3.365, 1.087, 6.858), 3. (-1.763, 2.218, 3.365), 4. (3.365, -0.924, 2.218), 5. (-2.555, 1.087, 8.034); for the lower concrete beam: 1. (-0.940, -0.940, 1.077), 2. (1.077, 1.077, -0.940), 3. (-0.9404, -1.806, 2.197), 4. (2.197, 0, -1.806), 5. (-1.806, -0.940, 3.333).

[0147] In this embodiment, the method for obtaining the structural deformation and deviation detection result includes:

[0148] Performing structural deformation and deviation measurement on the detection target to obtain the actual structural deformation and deviation;

[0149] Combining the actual structural deformation and deviation, the first structural deformation and deviation, the second structural deformation and deviation, the environmental impact factor, and the scale impact factor to form a comprehensive detection data set, and dividing the comprehensive detection data set into a training set and a test set;

[0150] Constructing a structural deformation and deviation intelligent detection model, where the structural deformation and deviation intelligent detection model includes an attention layer, a feature fusion layer, and a BP neural network;

[0151] The attention layer includes two parallel SE attention mechanism channels. The first structural deformation and deviation, the second structural deformation and deviation, and the corresponding environmental impact factor and scale impact factor in the training set are respectively input into the two parallel SE attention mechanism channels to extract the main features of the input data and learn the dependence relationship between the features to obtain the required feature data. The expression is:

[0152]

[0153] where F' Deviation is the required feature data matrix, σ(·) is the sigmoid function used to compress the numerical value between (0, 1) to generate the weight coefficient, W9, W 10is the weight matrix of the fully connected layer, which is used to compress and recover the number of channels C. δ(·) is the non-linear activation function. I and J are the row and column dimensions of the feature data matrix, and f c (i, j) is the value of the c-th channel of the feature data matrix at the position (i, j). F Deviation is the feature data matrix formed after the input data feature extraction;

[0154] The feature fusion layer includes a first classification layer and a second classification layer. The first classification layer consists of a concatenate fusion, a batch normalization layer, and a softmax classifier. The concatenate fuses the data features along the axis to obtain the representative features. The batch normalization layer normalizes the representative features. The softmax classifier converts the normalized representative features into the probability results of each category. The second classification layer consists of a support vector machine (SVM). The input probability results are meta-classified to determine the fusion result of the feature data. The cross-entropy loss function is used to optimize the classification result of the first classification layer, and the MSE function is used to optimize the fusion result of the second classification layer;

[0155] The BP neural network is used to learn the relationship between the structural deformation and deviation fusion result of the detection target and the actual structural deformation and deviation, environmental impact factors, and scale impact factors, and perform regression prediction to output the structural deformation and deviation detection result of the detection target. The MAE loss function is used to uniformly penalize the prediction error of the model, and the RMSProp optimizer is used to automatically adjust the learning rate of the model;

[0156] The test set is used to evaluate the structural deformation and deviation intelligent detection model;

[0157] The working data and status data of the target to be detected are input into the structural deformation and deviation intelligent detection model to obtain the structural deformation and deviation detection result;

[0158] In the actual evaluation, the working data and status data of the target to be detected are input into the intelligent detection model for structural deformation and deviation to obtain the detection results of structural deformation and deviation. Taking the detection results of the monitoring point cameras as an example, the deviation of the steel structure targets in the upper part of this span is as follows: 1. (5.903, -2.190, 10.677), 2. (3.526, 1.164, 7.091), 3. (-2.190, 2.342, 3.529), 4. (3.526, -1.109, 2.342), 5. (-3.256, 1.164, 8.287); the deviation of the concrete beam targets in the lower part of this span is as follows: 1. (-1.115, -1.115, 1.160), 2. (1.160, 1.160, -1.115), 3. (-1.115, -2.205, 2.335), 4. (2.335, 0, -2.205), 5. (-2.205, -1.115, 3.515). According to the analysis of the target coordinate deviation, there is only slight deformation in the concrete beam in the lower part of this span, and there is small deformation in the steel structure above the concrete beam. The deformation at the mid-span is greater than that at both ends, and the maximum deformation at the mid-span reaches 10.677 mm.

[0159] In the second aspect, an intelligent detection system for structural deformation and deviation based on machine vision includes:

[0160] Data acquisition module: including a high-resolution binocular camera, a target, and a measurement track; used to obtain the working data and status data of the detection target and preprocess the working data and the status data;

[0161] Image module: used to perform distortion processing on the camera image to obtain a corrected image, perform image processing on the corrected image to obtain point cloud data, and obtain target coordinate information based on the point cloud data and the corrected image;

[0162] BIM model module: used to establish a BIM model of the detection target according to building information;

[0163] Data processing module: used to construct a three-dimensional model of the detection target according to the target coordinate information, used to compare the three-dimensional model and the BIM model to obtain the basic structural deformation and deviation, used to construct an environmental impact factor and a scale impact factor according to environmental data and azimuth data, and used to determine the first structural deformation and deviation and the second structural deformation and deviation according to the environmental impact factor, the scale impact factor, and the basic structural deformation and deviation;

[0164] Detection model module: It is used to construct an intelligent detection model for structural deformation and deviation based on the first structural deformation and deviation, the second structural deformation and deviation, the environmental impact factor, the scale impact factor and the actual structural deformation and deviation, perform feature fusion and regression prediction, and input the working data and status data of the target to be detected into the intelligent detection model for structural deformation and deviation to obtain the detection result of structural deformation and deviation;

[0165] Intelligent supervision module: It is used to store, view and manage the working data, the status data and the detection result of structural deformation and deviation.

[0166] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. An intelligent detection method for structural deformation and deviation based on machine vision, characterized in that Including the following steps: S1. Obtain the working data and status data of the detection target, and preprocess the working data and the status data; the working data includes monitoring data and camera images; the status data includes environmental data and orientation data; S2. Perform distortion processing on the camera image to obtain a corrected image, perform image processing on the corrected image to obtain point cloud data, and obtain target point coordinate information based on the point cloud data and the corrected image; S3. Construct a three-dimensional model of the detection target according to the target point coordinate information, establish a BIM model of the detection target, and compare the three-dimensional model and the BIM model to obtain basic structural deformation and deviation; S4. Determine the first structural deformation and deviation according to the environmental data and the basic structural deformation and deviation, and determine the second structural deformation and deviation according to the orientation data and the basic structural deformation and deviation; S5. Construct an intelligent detection model for structural deformation and deviation according to the structural deformation and deviation and the monitoring data, and input the working data and status data of the target to be detected into the intelligent detection model for structural deformation and deviation to obtain the detection result of structural deformation and deviation.

2. The intelligent detection method for structural deformation and deviation based on machine vision according to claim 1, characterized in that The method for performing distortion processing on the camera image to obtain a corrected image includes: Using a camera distortion model to correct the camera image coordinates, and the expression is: x corr = x(1 + k1r 2 + k2r 4 + k3r 6 ) + 2p1xy + p2(r 2 + 2x 2 ) + δ - μ / x y corr = y(1 + k1r 2 + k2r 4 + k3r 6 ) + p1(r 2 + 2x 2 ) + 2p2xy + δ - μ / y where (x corr , y corr ) is the corrected position, (x, y) is the position before correction, k1, k2, k3 are the radial distortion correction parameters, p1, p2 are the tangential distortion correction parameters, δ is the orientation correction mean, and μ is the edge distortion value; Calibrating the camera parameters based on OpenCV functions to obtain the camera internal parameter matrix, tangential distortion correction parameters, and radial distortion correction parameters; Using a checkerboard calibration board for calibration: taking calibration board images at multiple positions and angles, detecting corner points to obtain rough positioning, performing sub-pixel processing to obtain corner point coordinates, establishing the relationship between checkerboard points and image points to obtain the internal parameter calibration result and distortion correction parameters, and correcting the image to obtain a corrected image.

3. The intelligent detection method for structural deformation and deviation based on machine vision according to claim 1, characterized in that The method for obtaining the target point coordinate information includes: Loading the depth image of the corrected image, extracting the depth image pixel values, and converting the depth image into point cloud data corresponding to three-dimensional coordinates according to the depth camera-detection target distance, image pixel values, and camera internal parameter matrix; Constructing a CNN machine learning model, inputting the corrected image and the point cloud data into the CNN machine learning model, using the CNN machine learning model to perform image recognition and feature extraction on the corrected image to obtain the image coordinates of the target point, using the CNN machine learning model to perform data preprocessing and data enhancement on the point cloud data, and determining the coordinate information of the target point based on the matching of the point cloud data and the image coordinates by the CNN machine learning model.

4. The intelligent detection method for structural deformation and deviation based on machine vision according to claim 1, wherein, The method for constructing a three-dimensional model of the detection target according to the target point coordinate information includes: Combining the measuring instrument turning point calculation algorithm and the dynamic local window algorithm, and determining the target point coordinates on the connecting traverse according to the coordinate information of the single observation point target output by the CNN machine learning model; Constructing a spatial index structure of the point cloud data near each target point, including: calculating the normal vector of the point cloud data, calculating the vector space, inversely solving the Poisson equation through the gradient relationship, performing isosurface extraction and reconstructing the point cloud model of the object to be measured, and the expression is: U = a1u 2 + a2uv + a3v 2 + a4u + a5v + a6 c g = (4a1a3 - a2 2 ) / (a4 2 + a5 2 + 1) 2 Among them is the vector space of target q, t is a point in the area near the target, and p is an arbitrary point are different surface areas is a smoothing function is the inward surface normal vector, U(u, v) is the quadratic surface equation, and a1, a2, a3, a4, a5, a6 are the coefficients of the quadratic surface equation, c g is the Gaussian curvature is the mean curvature Mesh the point cloud model of the object to be measured, including: constructing seed triangles and extracting boundary edges, determining adaptive neighborhood points according to the average distance density, calculating projection coordinates, and determining the optimal expansion points according to the projection coordinates. The expression is as follows: Z: n x (X - P x ) + n y (Y - P y ) + n z (Z - P z ) = 0 Q t3 = [cosθ·I + (1 - cosθ)kk T + sinθ·ROK]·Q t2 where Z is the neighborhood Q t1 The point cloud coordinates P (P x , P y , P z ) of the midpoint of the inner boundary edge and the corresponding normal vector n P (n x , n y , n z ) of the tangent plane equation, Q t2 (X, Y, Z) are the projection coordinates of the tangent plane, Q t3 are the rotated coordinates after the projection coordinates are rotated, I is the 3D identity matrix, θ is the angle between the normal vector n P and the normal vector of the XOY plane, k is the rotation axis, which is the cross product of the normal vector n P and the normal vector of the XOY plane; Repair the holes in the point cloud model of the object to be measured to obtain the three-dimensional model of the detection target, including: hole boundary recognition, determining the hole boundary direction, adding new vertices, adding new vertices according to the normal vector sum and curvature, new vertex inspection, and mesh smoothing. The expression is as follows: 1 / σ + 1 / ξ = ρ, σ > 0, ξ < -ξ where N n is the normal vector of the newly added mesh vertex, N l is the normal vector of the adjacent triangles of the newly added mesh vertex, g is the number of adjacent triangles in the one-ring neighborhood of the vertex, λ m is the shape factor of the adjacent triangle m, B m is the centroid distance of the adjacent triangle m, P n is the newly added mesh vertex, P n ” is the newly added mesh vertex after mesh fairing, σ is the positive influence factor in the contraction process, ξ is the negative influence factor in the stretching process, P u 、P v is the one-ring neighborhood, P u '、P v ' are the newly added mesh vertices P n after contraction of the one-ring neighborhood, h is the number of one-ring neighborhoods.

5. The intelligent detection method for structural deformation and deviation based on machine vision according to claim 1, characterized in that The method for obtaining the deformation and deviation of the basic structure includes: Establish a BIM model of the detection target, and compare the three-dimensional model and the BIM model to obtain the basic deformation and deviation Import the building information of the detection target into the REVIT software to construct the BIM model of the detection target; Register the three-dimensional model and the BIM model of the detection target through grid nodes, and determine the deviation mode quantity relationship by comparing the grid sizes and model reconstruction accuracies of each target point. The expression is as follows: f(m,n,o) = f P (m,n,o) - f BIM (m,n,o) where f(m,n,o) is the detection target deviation field, m, n, and o are the numbers of the target points in the x, y, and z directions, and f P (m,n,o) is the measured value of the target point on the three-dimensional model, and f BIM (m,n,o) is the theoretical value of the target point on the BIM model; Perform a forward DCT transformation on the deviation field of the detection target to obtain an orthogonal transformation coefficient matrix. The expression is as follows: Where C(μ, v, ω) is the orthogonal transformation coefficient matrix, M, N, and O are the total number of sampling points in the x, y, and z directions, and μ, v, and ω are frequency values; Perform key mode identification of the target deviation field according to the orthogonal transformation coefficient matrix C(μ, v, ω), and select important deviation data according to the energy compaction degree and contribution degree of the deviation field. The expression is as follows: Among which E i is the energy compaction degree of the important deviation mode, C i 2 (μ, v, ω) is the energy of the important deviation mode, C i (μ, v, ω) ∈ Ω, where Ω is the mode set, η i is the contribution degree of the important deviation mode; Select the deviation modes with energy compactness and contribution degree greater than the set threshold to form the key orthogonal transformation coefficient matrix For the orthogonal transformation coefficient matrix Perform the inverse DCT transformation to obtain the key deviation field of the detection target, and the expression is: Among them For detecting the key deviation field of the target, the basic structure deformation and deviation are determined according to the key deviation field of the detection target.

6. The intelligent detection method for structural deformation and deviation based on machine vision according to claim 1, characterized in that The method for determining the deformation and deviation of the first structure includes: Input the environmental data into the environmental impact function to obtain the environmental impact factor, determine its first deviation according to the environmental impact factor and the basic deviation of the coordinates at each target point, and determine its first structure deformation according to the first deviation of the coordinates at the overall target points of the detection target. The expression is as follows: Among them, Effect1 is the environmental impact factor, and w1, w2, w3, and w4 are the weights of the environmental impact factors. is the average illumination, L Rm , L Gm , L Bm are the measured values of red, green, and blue light in the environment, L Rs , L Gs , L Bs are the standard values of red, green, and blue light in the environment, T m is the operating temperature of the camera, T s is the standard operating temperature of the camera, H m is the measured value of atmospheric humidity, H s is the standard value of atmospheric humidity, N m is the measured value of environmental noise, N s is the acceptable threshold of environmental noise. Deviation1 is the first deviation of the coordinates at the overall target point of the detection target, and Deviation0 is the basic deviation of the coordinates at the overall target point of the detection target. P n ” is the target point coordinate after grid fairing, and P0 is the target point coordinate in the BIM model.

7. The intelligent detection method for structural deformation and deviation based on machine vision according to claim 1, characterized in that The method for determining the deformation and deviation of the second structure includes: Construct a scale impact factor according to the azimuth data, determine its second deviation according to the scale impact factor and the basic deviation of the coordinates at each target point, and determine its second structure deformation according to the second deviation of the coordinates at the overall target points of the detection target. The expression is as follows: Among them, Effect2 is the scale influence factor, w5, w6, and w7 are the weights of the scale influence factor, D is the distance from the camera to the detection target, and S BIM is the maximum value among the length, width, and height in the BIM model of the detection target, S0 is the standard size detected by the camera, is the inclination angle between the camera and the detected object, is the rotation angle between the camera and the detected object, Deviation2 is the second deviation of the coordinates at the overall target point of the detection target, and w8 is the weight of the second structural deviation.

8. The intelligent detection method for structural deformation and deviation based on machine vision according to claim 1, characterized in that The method for obtaining the detection result of the structure deformation and deviation includes: Measure the structure deformation and deviation of the detection target to obtain the actual structure deformation and deviation; Form a comprehensive detection data set with the actual structure deformation and deviation, the deformation and deviation of the first structure, the deformation and deviation of the second structure, the environmental impact factor, and the scale impact factor, and divide the comprehensive detection data set into a training set and a test set; Construct a structure deformation and deviation intelligent detection model, which includes an attention layer, a feature fusion layer, and a BP neural network; The attention layer extracts the main features of the training set data through two parallel SE attention mechanism channels, and learns the dependence relationship between the features to obtain the required feature data. The expression is as follows: Among them, F' Deviation is the required feature data matrix, σ(·) is the sigmoid function used to compress the numerical value between (0,1) to generate the weight coefficient, W9, W 10 is the weight matrix of the fully connected layer, used to compress and restore the number of channels C, δ(·) is the non-linear activation function, I, J are the row and column dimensions of the feature data matrix, f c (i,j) is the value of the c-th channel of the feature data matrix at the position (i,j), F Deviation is the feature data matrix composed after the input data feature extraction; The feature fusion layer performs probability calculation and feature fusion on the data processed by the attention layer through the first classification layer and the second classification layer, and outputs the fusion result of the structural deformation and deviation of the detection target. The cross-entropy loss function is used to optimize the classification result of the first classification layer, and the MSE function is used to optimize the fusion result of the second classification layer; The BP neural network is used to learn the relationship between the fusion result of the structural deformation and deviation of the detection target and the actual structural deformation and deviation, environmental impact factors, and scale impact factors, and performs regression prediction to output the detection result of the structural deformation and deviation of the detection target. The MAE loss function is used to uniformly penalize the prediction error of the model, and the RMSProp optimizer is used to automatically adjust the learning rate of the model; The test set is used to evaluate the intelligent detection model of structural deformation and deviation; The working data and status data of the target to be detected are input into the intelligent detection model of structural deformation and deviation to obtain the detection result of structural deformation and deviation.

9. An intelligent detection system for structural deformation and deviation based on machine vision, which is used to execute the method according to any one of claims 1-8, characterized in that, It includes: Data acquisition module: including a high-resolution binocular camera, a target, and a measurement track; It is used to obtain the working data and status data of the detection target and preprocess the working data and the status data; Image module: used to perform distortion processing on the camera image to obtain a corrected image, perform image processing on the corrected image to obtain point cloud data, and obtain target point coordinate information according to the point cloud data and the corrected image; BIM model module: used to establish a BIM model of the detection target according to building information; Data processing module: used to construct a three-dimensional model of the detection target according to the target point coordinate information, used to compare the three-dimensional model and the BIM model to obtain the basic structural deformation and deviation, used to construct environmental impact factors and scale impact factors according to environmental data and azimuth data, and used to determine the first structural deformation and deviation and the second structural deformation and deviation according to the environmental impact factors, the scale impact factors, and the basic structural deformation and deviation; Detection model module: used to construct an intelligent detection model of structural deformation and deviation according to the first structural deformation and deviation, the second structural deformation and deviation, the environmental impact factors, the scale impact factors, and the actual structural deformation and deviation, perform feature fusion and regression prediction, and input the working data and status data of the target to be detected into the intelligent detection model of structural deformation and deviation to obtain the detection result of structural deformation and deviation; Intelligent supervision module: used to store, view, and manage the working data, the status data, and the detection result of structural deformation and deviation.

Images