Concrete member defect detection method and system

Abstract

The invention provides a concrete member defect detection method and system, and relates to the technical field of automatic detection.The method comprises the following steps that a surface image of a to-be-detected concrete member is obtained, and the surface image is preprocessed; performing defect identification and calculation on the preprocessed surface image to obtain a corresponding defect measurement result; completing three-dimensional reconstruction of the defect according to the preprocessed surface image and the defect measurement result; according to the method, through a series of image processing and calculation steps, an automatic process from surface image acquisition to defect three-dimensional reconstruction and grading is realized, manual intervention is greatly reduced, the detection efficiency and consistency are improved, and a large number of concrete members can be rapidly and accurately detected and graded in a relatively short time.

Description

Technical Field

[0001] This invention relates to the field of automated inspection technology, and more specifically to a method and system for detecting defects in concrete components. Background Technology

[0002] Currently, during long-term service, concrete components can develop various degrees of damage, such as holes, cracks, grooves, and corrosion pits, due to changes in the external environment and human errors such as improper material selection or design, thus reducing the overall structural safety. Therefore, accurately identifying damage to concrete components has always been a research hotspot for professionals in related industries.

[0003] However, traditional methods for detecting surface defects in components rely on manual visual inspection, which is inefficient, susceptible to subjective factors, and difficult to accurately measure the three-dimensional characteristics of surface defects.

[0004] Therefore, how to provide a method for detecting defects in concrete components that can solve the above problems is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] In view of this, the present invention provides a method and system for detecting defects in concrete components. Through a series of image processing and calculation steps, it realizes an automated process from obtaining surface images to three-dimensional reconstruction and classification of defects, which greatly reduces manual intervention, improves detection efficiency and consistency, and can quickly and accurately detect and classify a large number of concrete components in a short period of time.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A method for detecting defects in concrete components includes the following steps:

[0008] S1: Acquire a surface image of the concrete component to be tested, and preprocess the surface image;

[0009] S2: Perform defect identification and calculation on the preprocessed surface image to obtain the corresponding defect measurement results;

[0010] S3: Complete the three-dimensional reconstruction of the defect based on the preprocessed surface image and the defect measurement results.

[0011] Preferred options also include:

[0012] S4: Classify defects based on the results of three-dimensional reconstruction.

[0013] Preferably, S1 specifically includes:

[0014] S11: Pre-calibrate the camera lens parameters, correct distortion, and perform epipolar calibration;

[0015] S12: Obtain a surface image of the concrete component to be tested using a camera lens processed by S11;

[0016] S13: Preprocess the surface image, wherein the preprocessing process includes noise reduction, histogram equalization and Laplacian sharpening.

[0017] Preferably, S2 specifically includes:

[0018] S21: Perform defect identification on the preprocessed surface image to obtain the corresponding target point and the corresponding point that matches the target point. At the same time, select multiple points on the surface of the concrete component to be tested that do not have defects and are not on the same straight line.

[0019] S22: Determine the three-dimensional coordinates of the target point and the corresponding point, and simultaneously determine the coordinates of multiple points;

[0020] S23: Determine the length of the defect based on the three-dimensional coordinates of the target point;

[0021] S24: Determine the depth and area of ​​the defect based on the coordinates of multiple points.

[0022] Preferably, S3 specifically includes:

[0023] S31: The preprocessed surface image is scaled using a Gaussian kernel function to obtain the corresponding scale space, and the Harris second-order matrix is ​​calculated in the scale space.

[0024] S32: Use a quadratic polynomial to fit the surface in the scale space to find the location of the maximum points at each scale in the scale space, and use Taylor expansion at the extreme points.

[0025] S33: Combine the calculation results obtained in S32 to determine the magnitude and direction of the pixel gradient within the neighborhood of the target point;

[0026] S34: Generate feature descriptors based on the pixel's magnitude and orientation, and process the feature descriptors to obtain the corresponding disparity map;

[0027] S35: Based on the disparity map and the coordinates obtained in S22, calculate the three-dimensional point cloud information of the defect to achieve three-dimensional reconstruction.

[0028] Preferably, S34 specifically includes:

[0029] S341: Normalize the feature descriptors;

[0030] S342: Use a KD tree based on the BBF algorithm to perform nearest neighbor search on feature points;

[0031] S343: Detection is performed using the RANSAC algorithm.

[0032] Preferably, S4 specifically includes:

[0033] S41: Determine whether the defect depth meets the threshold requirement;

[0034] S42: When the threshold requirement is met, calculate the correlation between defect length, defect area and defect depth, select the parameter with the highest correlation, determine the corresponding weight based on the parameter with the highest correlation and the defect depth, and calculate the corresponding defect feature parameters based on the defect depth, defect area and corresponding weights, and complete the classification based on the defect feature parameters.

[0035] S43: When the threshold requirement is not met, construct a grading model and input the defect length, defect area and defect depth into the grading model for processing to obtain the corresponding grading result.

[0036] The present invention also provides a concrete component defect detection system, comprising:

[0037] The acquisition module is used to acquire a surface image of the concrete component to be tested and to preprocess the surface image.

[0038] The calculation module is used to identify and calculate defects in the preprocessed surface image to obtain the corresponding defect measurement results;

[0039] The reconstruction module is used to complete the three-dimensional reconstruction of the defect based on the preprocessed surface image and the defect measurement results;

[0040] The grading module is used to grade defects based on the results of 3D reconstruction.

[0041] As can be seen from the above technical solution, compared with the prior art, the present invention discloses a method and system for detecting defects in concrete components, which has the following beneficial effects:

[0042] 1. The defect identification and calculation process provided by this invention, by determining and analyzing multiple non-collinear points of the target point, corresponding point, and defect-free part of the component surface, can more accurately calculate key measurement results such as the length, depth, and area of ​​the defect. This is an important basis for subsequent three-dimensional reconstruction and grading, and helps to improve the accuracy and reliability of the overall detection.

[0043] 2. This invention quantifies defects from multiple dimensions such as length, area, and depth, comprehensively describing the geometric characteristics and severity of defects, which helps to understand the potential impact of defects on the performance of concrete components more deeply.

[0044] 3. This invention achieves an automated process from surface image acquisition to three-dimensional reconstruction and grading of defects through a series of image processing and calculation steps, which greatly reduces manual intervention, improves detection efficiency and consistency, and enables rapid and accurate detection and grading of a large number of concrete components in a short period of time.

[0045] 4. This invention utilizes Gaussian kernel functions for scaling, Harris second-order matrix calculation, surface fitting, and extreme point solving to accurately determine the scale-space extreme points of image feature points, providing crucial information for subsequent feature descriptor generation and disparity map calculation. Based on pixel gradient magnitude and direction, feature descriptors are generated, and through normalization, KD-tree nearest neighbor search, and RANSAC algorithms, the accuracy and robustness of feature matching are effectively improved, resulting in a more accurate disparity map. Combining the disparity map with the determined coordinate information, the three-dimensional point cloud information of the defect can be calculated, achieving accurate three-dimensional reconstruction of concrete defects and providing intuitive and detailed spatial information for subsequent defect analysis and classification.

[0046] 5. This invention flexibly employs different grading strategies based on the correlation between parameters such as defect depth, length, and area, as well as pre-set threshold requirements. For defects that meet the threshold requirements, the weights of each parameter can be accurately determined, defect characteristic parameters can be reasonably calculated, and grading can be completed. For defects that do not meet the threshold requirements, a grading model is constructed for comprehensive processing to obtain reliable grading results. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0048] Figure 1 This invention provides an overall flowchart of a method for detecting defects in concrete components.

[0049] Figure 2 The present invention provides a structural principle block diagram of a concrete component defect detection system. Detailed Implementation

[0050] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] See Figure 1 As shown in the figure, an embodiment of the present invention discloses a method for detecting defects in concrete components, including the following steps:

[0052] S1: Acquire the surface image of the concrete component to be tested and preprocess the surface image. The above process can be realized by remotely transmitting the surface image to the relevant host computer through the wireless network module. At the same time, the two cameras on the crack detector can be used to acquire images of the concrete component to be tested. The wireless network module is implemented using Zigbee2007.

[0053] S2: Perform defect identification and calculation on the preprocessed surface image to obtain the corresponding defect measurement results;

[0054] S3: Complete the three-dimensional reconstruction of the defect based on the preprocessed surface image and defect measurement results.

[0055] In one specific embodiment, it also includes:

[0056] S4: Classify defects based on the results of three-dimensional reconstruction.

[0057] In one specific embodiment, S1 specifically includes:

[0058] S11: Pre-calibrate the camera lens parameters, correct distortion, and perform epipolar calibration;

[0059] S12: Obtain the surface image of the concrete component to be tested using the camera lens processed by S11;

[0060] S13: Preprocess the surface image, including noise reduction, histogram equalization, and Laplacian sharpening.

[0061] Specifically, S11 can be implemented using Zhang Zhengyou's calibration method based on 2D planar targets. The specific implementation process includes:

[0062] (1) Adjust the field of view of the left and right cameras and fix them in place;

[0063] (2) Take a total of 9 pairs of images of the chessboard calibration board, including translation and rotation;

[0064] (3) The internal and external parameters of the left and right cameras were calculated respectively using Zhang Zhengyou's calibration method based on 2D planar targets;

[0065] (4) Finally, the structural parameters of the binocular vision sensor are calculated.

[0066] Camera lenses exhibit nonlinear distortion. The purpose of distortion correction is to eliminate the impact of lens distortion on the image, making it approximate an ideal pinhole camera model, thus preparing for subsequent epipolar calibration and image matching. The steps of distortion correction are:

[0067] (1) The radial distortion coefficients k1, k2 and the tangential distortion coefficients p1, p2 of the camera lens were calibrated using the Zhang Zhengyou method;

[0068] (2) Substitute the obtained distortion coefficients into the formula to obtain the corrected image coordinates, i.e. the ideal image coordinates (X,Y);

[0069] (3) Perform the above processing on each pixel in the original image to obtain the image after distortion correction.

[0070] Epipolar calibration transforms the perspective transformation matrices of the original left and right cameras into new perspective transformation matrices, ensuring that the epipolar lines of the transformed stereo image pairs are horizontal and parallel, with corresponding epipolar lines collinear. This essentially converts a non-parallel binocular stereo vision model into a parallel binocular stereo vision model, with the epipolar point at infinity. By using epipolar geometry to calibrate image pairs, the collinearity of corresponding epipolar lines in the calibrated image pairs reduces the two-dimensional search during region matching to a one-dimensional search, improving matching accuracy and reducing computational load.

[0071] The specific process of S13 may include: the filtering process may include mean filtering and Gaussian filtering. Mean filtering is used to remove random noise. It means replacing the pixel value of the target point in the template with the mean value of the pixels in the template, using a 5×5 template. Gaussian filtering is used to eliminate Gaussian noise, which is the main noise in the image, that is, random noise that conforms to a normal distribution.

[0072] Histogram equalization is used to eliminate brightness differences between left and right image pairs. By performing the above processing on both images in the image pair, the brightness of the image becomes uniform and the brightness difference between the image pairs is balanced. In order to make the details and edges of the image clear, Laplacian sharpening is performed on the image pair, and the sharpened result is the grayscale of the final image.

[0073] In one specific embodiment, S2 specifically includes:

[0074] S21: Defect identification is performed on the preprocessed surface image to obtain the corresponding target points and their matching points. Simultaneously, multiple points not on the same straight line are selected from the defect-free portion of the concrete component's surface. Specifically, arbitrary target points on the defective surface are manually selected in the left image, and their corresponding matching points are automatically calculated in the right image using a region matching algorithm. After obtaining the matching point pairs, the 3D coordinates of these points can be calculated based on the principle of parallel binocular stereo vision, thus enabling the measurement of 3D information for any point of interest.

[0075] S22: Determine the three-dimensional coordinates of the target point and its corresponding points, and simultaneously determine the coordinates of multiple points;

[0076] S23: Determine the length of the defect based on the three-dimensional coordinates of the target point. The specific expression is as follows:

[0077]

[0078] In the formula, the world coordinates of the two points in space are P1(X1,Y1,Z1) and P2(X2,Y2,Z2), and the formula for calculating the width is the same as that in formula (1).

[0079] S24: Determine the depth and area of ​​the defect based on the coordinates of multiple points. The specific process may include:

[0080] S241: The specific process of depth calculation

[0081] Three points on the undamaged surface of the component, not collinear, are selected as reference points to establish a reference surface. Let the world coordinates of these three points be P1(X1,Y1,Z1), P2(X2,Y2,Z2), and P3(X3,Y3,Z3), and the coordinates of the point whose depth is to be determined be P(X,Y,Z). Then the normal vector of the reference surface is:

[0082]

[0083] The depth h of point P is:

[0084]

[0085] First, select three non-collinear points in the non-defect area of ​​the left figure, and automatically match their corresponding points in the right figure; then load the parameters of the binocular stereo vision system, calculate the three-dimensional world coordinates of the three points according to the parallel binocular stereo vision model, and calculate the normal vector of the reference surface according to equation (2); use the same method to obtain the world coordinates of the depth target point, and substitute them into equation (3) to obtain the defect depth.

[0086] S242: Detailed process of area calculation

[0087] The semi-automatic defect area detection method first manually selects the target points of the defect edge in the left image, and then automatically calculates the corresponding matching points in the right image based on a region matching algorithm. This semi-automatic defect edge point selection method is used sequentially to obtain the curve of the entire defect edge in both the left and right images. The specific steps are as follows:

[0088] Three points are selected in the non-defective part of the object under test to establish a reference surface. The normal vector of the reference surface is calculated according to equation (2). The defect edge points are selected sequentially in the left figure, and the corresponding points are automatically matched in the right figure. The world coordinates of these points are calculated according to the parallel binocular stereo vision model as P1(X1,Y1,Z1), P2(X2,Y2,Z2), P3(X3,Y3,Z3), P4(X4,Y4,Z4)..., P n (X n ,Y n Z n );

[0089] According to the formula (4) for the foot of the perpendicular from a point to a plane, the selected edge point is projected onto the reference plane. The specific expression is as follows:

[0090]

[0091] In the formula, (X′ i ,Y′ i Z′ i P1(X1,Y1,Z1), P2(X2,Y2,Z2), and P3(X3,Y3,Z3) are the three points selected in the first step to establish the reference plane; (X′) are the perpendicular coordinates of point i. i ,Y′ i Z′ i () represents the point selected in the second step.

[0092] Connect the orthographic projection points in sequence to form an irregular polygon in the plane. Divide the polygon into n-2 triangles with a certain vertex as the reference. Calculate the area of ​​each triangle and sum them up to get the total area of ​​the defect.

[0093] In a specific embodiment, S3 specifically includes:

[0094] S31: The preprocessed surface image I(x,y,σ) is scaled using a Gaussian kernel function G(x,y,σ) to obtain the corresponding scale space. The Harris second-order matrix is ​​then calculated in this scale space, with the specific expression as follows:

[0095] L(x,y,σ)=G(x,y,σ)*I(x,y,σ) (5)

[0096]

[0097] In the formula, μ(x,y,σ) i ,σ D ) is a Harris second-order matrix, σ i For the integral scale, σ D For the differential scale, σ i =sσ D L x (x,y,σ D ) and L y (x,y,σ D ) represent the first-order derivatives of the image in the x and y directions after smoothing, respectively;

[0098] Meanwhile, embodiments of the present invention can use the Harris algorithm to process the transformed surface image to eliminate some irrelevant isolated points and improve the image's noise resistance.

[0099] S32: Perform surface fitting using a quadratic polynomial in scale space to solve for the locations of the maxima m(x,y) at each scale in scale space. Simultaneously, apply Taylor expansion at the maxima. The specific expression is as follows:

[0100] m(x,y)=a0+a1x+a2y+a3x 2 +a4y 2 +a5xy (8)

[0101]

[0102] Since some candidate feature points are easily affected by scaling transformations, the LoG function is used to filter them. The LoG formula is shown below:

[0103]

[0104] In the formula, σ n Let L be the scale of the nth layer. xx With L yy L(x,y,σ) n The second derivative of ).

[0105] In the three dimensions of x, y, and σ, if the offset in any one dimension is greater than 0.5, the position of that point should be changed until all feature points converge.

[0106] The Hessian matrix is ​​used to remove edge feature points with poor reliability, and the principal curvature can be obtained by equation (11).

[0107]

[0108] In the formula, D xx D xy Dyy All can be obtained by the difference between the current pixel and the surrounding pixels;

[0109] Assuming α and β are the eigenvalues ​​of matrix H, and α > β, then we can conclude that:

[0110] Tr(H)=D xx +D yy =α+β (12)

[0112] Det(H) = D xx D yy -(D yy ) 2 =αβ (13)

[0113] In the formula, Tr(H) represents the trace of matrix H, and Det(H) represents the determinant of matrix H;

[0114] Let α = rβ, then we have:

[0115]

[0116] Centered on the candidate feature point, calculate the LoG value of a 3×3 window region within the neighboring scales of the same scale. When the LoG value of the candidate feature point is maximized, select that point as the final feature point.

[0117] S33: Based on the calculation results obtained in S32, determine the pixel gradient m(x,y) and direction θ(x,y) within the neighborhood of the target point. The specific expression is as follows:

[0118]

[0119] In the formula, L represents the scale space value of the feature point. The gradient and direction of each feature point are statistically analyzed using a gradient histogram, which is then divided into 36 bars (each bar representing 10°). The direction of the peak value in the gradient histogram is used as the primary direction of the feature point. To enhance the rotation invariance of the feature points, the SIFT algorithm uses the gradient directions that are greater than 80% of the peak value in the gradient histogram as the secondary directions of the feature points.

[0120] Then generate the description vector.

[0121] Step 1: Select a circular region with a radius of 8 pixels centered on the feature point as the feature point's neighborhood. Divide this circular region into two circles with radii of 4 and 8 pixels respectively. Within the circle with a radius of 4 pixels, divide it into 4 sub-regions by the coordinate axes of the feature point; within the 4 sub-regions of the circle with a radius of 8 pixels, each sub-region is divided into two sub-regions by a 45° line, generating 8 sub-regions. In total, the space within the two circles is divided into 12 parts.

[0122] Step 2: Calculate the magnitude and direction of the gradient direction of each pixel in each sub-region using equations (15) and (16). Since the distance between pixels within the neighborhood of a feature point and the feature point varies, their impact on the feature point also differs. Gaussian weights are applied to the pixel gradients within each sub-region, with higher weights for pixels closer to the feature point.

[0123] Step 3: Calculate the gradient histograms in 8 directions for each sub-region, sort them from largest to smallest gradient values, and generate 8-dimensional vectors. Each feature point generates a total of 8 × (4 + 8) = 96-dimensional vectors.

[0124] S34: Generate feature descriptors based on the pixel's magnitude and orientation, and process the feature descriptors to obtain the corresponding disparity map;

[0125] S35: Based on the disparity map and the coordinates obtained in S22, calculate the three-dimensional point cloud information of the defect to achieve three-dimensional reconstruction.

[0126] In a specific embodiment, S34 specifically includes:

[0127] S341: Normalize the feature descriptors, the specific expression is as follows:

[0128]

[0129] In the formula, L is the normalized feature descriptor;

[0130] S342: A KD-tree based on the BBF algorithm is used to perform nearest neighbor search on feature points. The KD-tree is used to query the target point, recording the distance between each node in the query path and the target point. The nodes in the query path are sorted according to their distance to the target point, ensuring that each traversal always starts from the node with the highest priority. During the traversal, the two closest nodes are recorded as the nearest and second nearest neighbors. When the queue is empty or the number of traversals reaches a certain threshold, the search ends. The current nearest and second nearest neighbors are then used as the nearest and second nearest neighbors of the target point. The ratio of the nearest and second nearest neighbors is used as the matching point. From the above operations, the nearest neighbor distance n1 and the second nearest neighbor distance n2 can be obtained. The ratio (n1 / n2) is compared with a threshold. If the ratio is less than the threshold (0.8), the matching point pair is retained; otherwise, if the ratio is greater than the threshold, the matching point pair is discarded to reduce the false matching rate.

[0131] A bidirectional matching strategy is used to constrain matching point pairs. That is, when image A is used as the reference image, feature points in image A correspond to feature points in image B; and when image B is used as the reference image, the corresponding feature points in image B should also correspond to feature points in image A. If the feature points obtained from the two matches do not correspond, the matching pair should be deleted to reduce the false matching rate.

[0132] S343: Use the RANSAC algorithm for detection. Randomly select 4 pairs of matching points. If the 4 pairs are collinear, reselect; otherwise, calculate the transformation matrix and use it to detect the remaining points, determining if each pair is an interior point. Count the number of interior points and check if it exceeds the optimal number S in the initial image. If the number is less than S, reselect; otherwise, use the transformation matrix as the best estimate and update S. Continue this process until the maximum number of iterations is reached.

[0133] In this embodiment of the invention, high-quality matching point pairs generated based on the improved SIFT algorithm are used as the initial seed queue in the region growing algorithm through the above process. The Normalized Cross-Correlation (NCC) algorithm is selected for matching similarity measurement, simplifying the computational load of the NCC algorithm. First, a point in the left image (to be matched) is selected, and the candidate point is locked onto the epipolar line using epipolar constraints. Second, most areas in the binary image of concrete cracks are black background areas, with only the crack edges being white areas; therefore, grayscale constraints after epipolar constraints can quickly narrow the search range. Finally, the Normalized Cross-Correlation algorithm is used to find the point among the remaining candidate points with the largest normalization coefficient to the reference point, thus determining the final matching point. Median filtering is applied to the obtained disparity map to obtain the final disparity map.

[0134] By combining the disparity map obtained through stereo matching with the three-dimensional world coordinates of the two points calculated by the parallel binocular stereo vision model, the three-dimensional point cloud information of the measured object can be calculated, thereby recovering the three-dimensional structural information of the object.

[0135] In one specific embodiment, S4 specifically includes:

[0136] S41: Determine whether the defect depth meets the threshold requirement;

[0137] S42: When the threshold requirement is met, calculate the correlation between defect length, defect area, and defect depth. Select the parameter with the highest correlation, determine the corresponding weight based on the parameter with the highest correlation and the defect depth, and calculate the corresponding defect feature parameters based on the defect depth, defect area, and corresponding weights. Complete the classification based on the defect feature parameters. The specific expression is as follows:

[0138] n = s1h + s2b (18)

[0139] In the formula, n represents the defect feature parameter, h represents the defect depth, b represents the parameter with the highest correlation, and s1 and s2 represent the parameter weights.

[0140] S43: When the threshold requirement is not met, construct a hierarchical model and input the defect length, defect area and defect depth into the hierarchical model for processing to obtain the corresponding hierarchical result. The hierarchical model can be any machine vision model.

[0141] See Figure 2 As shown, this embodiment of the invention also provides a system for detecting defects in concrete components using any of the above embodiments, comprising:

[0142] The acquisition module is used to acquire surface images of the concrete component to be tested and to preprocess the surface images.

[0143] The calculation module is used to identify and calculate defects in the preprocessed surface image to obtain the corresponding defect measurement results.

[0144] The reconstruction module is used to complete the three-dimensional reconstruction of defects based on the preprocessed surface image and defect measurement results;

[0145] The grading module is used to grade defects based on the results of 3D reconstruction.

[0146] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0147] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for detecting defects in concrete components, characterized in that, Includes the following steps: S1: Acquire a surface image of the concrete component to be tested, and preprocess the surface image; S2: Perform defect identification and calculation on the preprocessed surface image to obtain the corresponding defect measurement results; S3: Complete the three-dimensional reconstruction of the defect based on the preprocessed surface image and the defect measurement results.

2. The method for detecting defects in concrete components according to claim 1, characterized in that, Also includes: S4: Classify defects based on the results of three-dimensional reconstruction.

3. The method for detecting defects in concrete components according to claim 1, characterized in that, S1 specifically includes: S11: Pre-calibrate the camera lens parameters, correct distortion, and perform epipolar calibration; S12: Obtain a surface image of the concrete component to be tested using a camera lens processed by S11; S13: Preprocess the surface image, wherein the preprocessing process includes noise reduction, histogram equalization and Laplacian sharpening.

4. The method for detecting defects in concrete components according to claim 2, characterized in that, S2 specifically includes: S21: Perform defect identification on the preprocessed surface image to obtain the corresponding target point and the corresponding point that matches the target point. At the same time, select multiple points on the surface of the concrete component to be tested that do not have defects and are not on the same straight line. S22: Determine the three-dimensional coordinates of the target point and the corresponding point, and simultaneously determine the coordinates of multiple points; S23: Determine the length of the defect based on the three-dimensional coordinates of the target point; S24: Determine the depth and area of ​​the defect based on the coordinates of multiple points.

5. The method for detecting defects in concrete components according to claim 4, characterized in that, S3 specifically includes: S31: The preprocessed surface image is scaled using a Gaussian kernel function to obtain the corresponding scale space, and the Harris second-order matrix is ​​calculated in the scale space. S32: Use a quadratic polynomial to fit the surface in the scale space to find the location of the maximum points at each scale in the scale space, and use Taylor expansion at the extreme points. S33: Combine the calculation results obtained in S32 to determine the magnitude and direction of the pixel gradient within the neighborhood of the target point; S34: Generate feature descriptors based on the pixel's magnitude and orientation, and process the feature descriptors to obtain the corresponding disparity map; S35: Based on the disparity map and the coordinates obtained in S22, calculate the three-dimensional point cloud information of the defect to achieve three-dimensional reconstruction.

6. The method for detecting defects in concrete components according to claim 5, characterized in that, S34 specifically includes: S341: Normalize the feature descriptors; S342: Use a KD tree based on the BBF algorithm to perform nearest neighbor search on feature points; S343: Detection is performed using the RANSAC algorithm.

7. The method for detecting defects in concrete components according to claim 4, characterized in that, S4 specifically includes: S41: Determine whether the defect depth meets the threshold requirement; S42: When the threshold requirement is met, calculate the correlation between defect length, defect area and defect depth, select the parameter with the highest correlation, determine the corresponding weight based on the parameter with the highest correlation and the defect depth, and calculate the corresponding defect feature parameters based on the defect depth, defect area and corresponding weights, and complete the classification based on the defect feature parameters. S43: When the threshold requirement is not met, construct a grading model and input the defect length, defect area and defect depth into the grading model for processing to obtain the corresponding grading result.

8. A system utilizing the method for detecting defects in concrete components according to any one of claims 1-7, characterized in that, include: The acquisition module is used to acquire a surface image of the concrete component to be tested and to preprocess the surface image. The calculation module is used to identify and calculate defects in the preprocessed surface image to obtain the corresponding defect measurement results; The reconstruction module is used to complete the three-dimensional reconstruction of the defect based on the preprocessed surface image and the defect measurement results; The grading module is used to grade defects based on the results of 3D reconstruction.

Images