Bridge bottom disease positioning method
Through image acquisition and deep learning recognition combined with panoramic image matching technology, the problem of low manual inspection dependence and automation in bridge bottom disease positioning is solved, and accurate and rapid positioning of bridge bottom disease is achieved, and the detection is improved. The visual effect has important bridge safety significance.
Patent Information
- Application Number
- CN202310610312.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-05-29
- Publication Date
- 2024-12-03
AI Technical Summary
The existing bridge beam bottom disease positioning methods rely on manual inspection and are low in degree of automation. It is difficult for traditional methods to achieve accurate and intuitive visual display of disease location, especially when the visual environment at the bottom of bridge beam bottom is poor.
Using technologies such as image acquisition, deep learning disease recognition, quantitative positioning, grayscale normalization processing and image feature matching, the global positioning of disease points is achieved by generating panoramic images and mapping transformation matrix algorithms, reducing the dependence on the subjective judgment of detectors. Improve automation.
It realizes more convenient and accurate positioning of the disease conditions at the bottom of the bridge beam, generates visual two-dimensional images, improves the speed and accuracy of disease detection, and is of great significance to bridge safety.
Smart Images

Figure CN119067903A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of bridge safety, and in particular relates to a method for locating bridge beam bottom defects. Background Art
[0002] In the past, the most commonly used method for locating bridge beam bottom defects was regular manual inspections and manual measurement and positioning. However, this method is time-consuming and laborious, and the visual environment at the bottom of the bridge beam is poor, which often leads to misjudgment and omission, and the data presentation is not intuitive. Therefore, how to automatically find the location of bridge beam bottom defects and achieve better visual presentation is an urgent need in the current field of bridge beam bottom defects.
[0003] Existing solutions generally use drones to replace manual inspections, with the focus on using deep learning to detect diseases. However, the location of the diseases still relies on the subjective judgment of the inspectors operating the drones, and the degree of automation is low.
[0004] In addition, the following scheme emphasizes the use of laser ranging to determine the location of the defect and the rangefinder, but ignores the fact that the bottom defects of the bridge beam have different relative positions on the bottom plate in different areas, lacks corresponding change algorithms, and the defect positioning form is too rigid. The degree of damage of the defect can only be determined through two-dimensional distance data, and the data performance is poor. Summary of the invention
[0005] The purpose of the present invention is to provide a method for locating bridge beam bottom defects, which avoids the defects of traditional detection means and can locate the defects of the bridge beam bottom more conveniently and accurately.
[0006] In order to achieve the above-mentioned object of the invention, the technical solution provided by the present invention is as follows:
[0007] 1. A method for locating bridge beam bottom defects, comprising the following steps:
[0008] S1: Image acquisition, sequentially collecting n images covering the bottom of all bridge beams;
[0009] S2: Disease identification, detecting the type of disease in the image through a deep learning disease classification model;
[0010] S3: Quantitative positioning of defective images: quantitative positioning of defective images. Binarization of defective image pixels to 0 or 255 is performed. A collision-first-then-erosion closing operation is used to eliminate hollow points in the binary image. The minimum enclosing rectangle is calculated for the defective area with 0 pixels, that is, the defective points are quantitatively positioned on the image.
[0011] S4: Grayscale normalization processing, grayscale normalization processing is performed on all collected images to make their brightness uniform;
[0012] S5: Image feature matching, generating a panoramic image, using the SIFT algorithm to extract feature points from all images in turn, matching and fusing them two by two, and performing a smooth transition to form a panoramic image of the entire beam bottom. The specific steps are as follows:
[0013] (1) Match the features of image 1 and image 2 and merge them into image 1&2;
[0014] (2) Match the features of images 1&2 and 3 and merge them into images 1&3;
[0015] (3) Repeat the above steps until all images are matched and merged into images 1&n, which is the final stitched image, that is, the panorama;
[0016] S6: Global positioning of the defect point. The local coordinates of the defect point in the defect image in step S3 are converted to the global coordinates in the panoramic image through the mapping transformation M matrix algorithm. The specific steps of calculating the M matrix algorithm are as follows:
[0017] (1) In step S3, the coordinates (x, y) of the defect point in the defect image are present. Image 1 is used as the standard axis system O1, and the defect image is put into the queue from any position i to start pairwise feature fusion;
[0018] (2) If i divided by 2 is greater than or equal to 1, the number of transformations of the defect image i is n=n+1, otherwise the number of transformations is n=n and the process is skipped;
[0019] (3) Record matrices M1, M2, ...Mn during feature fusion;
[0020] (4) The transformation matrix of image i relative to image 1 is M = M1*M2...*Mn;
[0021] (5) Therefore, the equivalent distance of the image relative to the O1 coordinate system is (X, Y) = M(x, y).
[0022] Preferably, the method for disease identification in step S2 is to determine whether the image has diseases based on a deep learning convolutional neural network disease detection model.
[0023] Preferably, the grayscale normalization processing method in step S4 is to map the pixel values of the captured image to a range of 0-1, that is, the pixel value of any point in the image is divided by the difference between the maximum pixel value and the minimum pixel value in the image.
[0024] Preferably, the technical solution of the present invention is applicable to small and medium-sized bridges, especially prefabricated T-beam bridges and hollow slab beam bridges.
[0025] Compared with the prior art, the present invention has the following beneficial effects:
[0026] The present invention avoids the influence of the personal level of the inspectors on the results in traditional detection methods, converts the on-site image of the bottom of the bridge beam into a visualized two-dimensional image, and can more conveniently and accurately detect the defects at the bottom of the bridge beam through the defect detection model, and realizes the rapid positioning of the defects, which is of great significance to bridge safety. BRIEF DESCRIPTION OF THE DRAWINGS
[0027] Figure 1 It is a flow chart of the bridge beam bottom disease locating method of the present invention. DETAILED DESCRIPTION
[0028] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0029] The specific embodiment is described as follows: a method for locating bridge beam bottom defects, combined with Figure 1 Content, including the following steps:
[0030] S1: Image acquisition, sequentially collecting n images covering the bottom of all bridge beams;
[0031] Generally, multiple camera devices are used to collect images of the bottom of the bridge beam. The number of cameras depends on the type of bridge being tested. After relevant tests, this method generally measures a hollow slab bridge with 4 cameras. When the bridge is a prefabricated T-beam bridge, 6 cameras are more appropriate. The number and arrangement of cameras are customized according to the cross-sectional form of the bridge. For prefabricated T-beam bridges, the number of cameras used can cover the bottom of a beam, and full coverage can be achieved through multiple rounds of conversion. For hollow slab beam bridges, since the bottom surface is relatively simple, the number of cameras can be simply increased to increase the coverage area and improve the collection efficiency. Generally, 6 cameras are used for simultaneous collection.
[0032] S2: Disease identification, using a deep learning disease classification model to detect the type of disease in the image. Specifically, a deep learning-based convolutional neural network disease detection model is used to determine whether the image has a disease.
[0033] S3: Quantitative positioning of defective images: quantitative positioning of defective images. Binarization of defective image pixels to 0 or 255 is performed. A collision-first-then-erosion closing operation is used to eliminate hollow points in the binary image. The minimum enclosing rectangle is calculated for the defective area with 0 pixels, that is, the defective points are quantitatively positioned on the image.
[0034] S4: Grayscale normalization processing, grayscale normalization processing is performed on all the collected images to make their brightness uniform, specifically, the pixel value of the captured image is mapped to the range of 0-1, that is, the pixel value of any point in the image is divided by the difference between the maximum pixel value and the minimum pixel value in the image;
[0035] S5: Image feature matching, generating a panoramic image, using the SIFT algorithm to extract feature points from all images in turn, matching and fusing them two by two, and performing a smooth transition to form a panoramic image of the entire beam bottom. The specific steps are as follows:
[0036] (1) Match the features of image 1 and image 2 and merge them into image 1&2;
[0037] (2) Match the features of images 1&2 and 3 and merge them into images 1&3;
[0038] (3) Repeat the above steps until all images are matched and merged into images 1&n, which is the final stitched image, that is, the panorama;
[0039] S6: Global positioning of the defect point. The local coordinates of the defect point in the defect image in step S3 are converted to the global coordinates in the panoramic image through the mapping transformation M matrix algorithm. The specific steps of calculating the M matrix algorithm are as follows:
[0040] (1) In step S3, the coordinates (x, y) of the defect point in the defect image are present. Image 1 is used as the standard axis system O1, and the defect image is put into the queue from any position i to start pairwise feature fusion;
[0041] (2) If i divided by 2 is greater than or equal to 1, the number of transformations of the defect image i is n=n+1, otherwise the number of transformations is n=n and the process is skipped;
[0042] (3) Record matrices M1, M2, ...Mn during feature fusion;
[0043] (4) The transformation matrix of image i relative to image 1 is M = M1*M2...*Mn;
[0044] (5) Therefore, the equivalent distance of the image relative to the O1 coordinate system is (X, Y) = M(x, y).
[0045] The present invention converts the on-site image of the bottom of the bridge beam into a visualized two-dimensional image. Through the defect detection model, the defect condition of the bottom of the bridge beam can be detected more conveniently and accurately, and the defect can be quickly located, which is of great significance to bridge safety.
Claims
1. A method for locating bridge beam bottom defects, characterized in that: The following steps are involved: S1: Image acquisition, sequentially collecting n images covering the bottom of all bridge beams; S2: Disease identification, detecting the type of disease in the image through a deep learning disease classification model; S3: Quantitative positioning of defective images: quantitative positioning of defective images. Binarization of defective image pixels to 0 or 255 is performed. A collision-first-then-erosion closing operation is used to eliminate hollow points in the binary image. The minimum enclosing rectangle is calculated for the defective area with 0 pixels, that is, the defective points are quantitatively positioned on the image. S4: Grayscale normalization processing, grayscale normalization processing is performed on all collected images to make their brightness uniform; S5: Image feature matching, generating a panoramic image, using the SIFT algorithm to extract feature points from all images in turn, matching and fusing them two by two, and performing a smooth transition to form a panoramic image of the entire beam bottom. The specific steps are as follows: (1) Match the features of image 1 and image 2 and merge them into image 1&2; (2) Match the features of images 1&2 and 3 and merge them into images 1&3; (3) Repeat the above steps until all images are matched and merged into images 1&n, which is the final stitched image, that is, the panorama; S6: Global positioning of the defect point. The local coordinates of the defect point in the defect image in step S3 are converted to the global coordinates in the panoramic image through the mapping transformation M matrix algorithm. The specific steps of calculating the M matrix algorithm are as follows: (1) In step S3, the coordinates (x, y) of the defect point in the defect image are present. Image 1 is used as the standard axis system O1, and the defect image is put into the queue from any position i to start pairwise feature fusion; (2) If i divided by 2 is greater than or equal to 1, the number of transformations of the defect image i is n=n+1, otherwise the number of transformations is n=n and the process is skipped; (3) Record matrices M1, M2, ...Mn during feature fusion; (4) The transformation matrix of image i relative to image 1 is M = M1*M2...*Mn; (5) Therefore, the equivalent distance of the image relative to the O1 coordinate system is (X, Y) = M(x, y).
2. The bridge beam bottom defect locating method according to claim 1, characterized in that: The method for disease identification in step S2 is to determine whether the image has diseases based on a convolutional neural network disease detection model based on deep learning.
3. The bridge beam bottom defect locating method according to claim 1, characterized in that: The grayscale normalization processing method in step S4 is to map the pixel value of the captured image to a range of 0-1, that is, the pixel value of any point in the image is divided by the difference between the maximum pixel value and the minimum pixel value in the image.
4. The bridge beam bottom defect locating method according to claim 1, characterized in that: The bridge is a small or medium-sized bridge.
5. The bridge beam bottom defect locating method according to claim 4, characterized in that: The small and medium-sized bridges are prefabricated T-beam bridges and hollow slab beam bridges.
Citation Information
Patent Citations
Method for detecting surface diseases of bridges
CN109064449A
Bridge disease identification method adopting unmanned aerial vehicle image
CN110070008A
Crack detection method based on remote sensing image
CN110084302A
Beam bottom inspection and disease visualization method and system based on unmanned aerial vehicle
CN112669278A
Bridge bottom disease all-morphology detection method and system
CN114627107A
Cited By
Intelligent analysis and diagnosis method for visual data of railway steel arch bridge inspection robot
CN121305161A
