Tunnel wall surface deformation detection method based on computer image recognition

Abstract

The invention relates to the technical field of tunnel detection, in particular to a tunnel wall surface deformation detection method based on computer image recognition, which comprises the following steps: S1, setting an image acquisition system; s2, image preprocessing; s3, extracting wall surface features; s4, deformation detection; and S5, deformation early warning and reporting. In an image acquisition link, image acquisition is not singly carried out, and three-dimensional point cloud data of a tunnel wall surface is acquired by utilizing a laser radar technology to assist image acquisition and is fused with image data. By making the features in the image correspond to the space coordinates in the three-dimensional point cloud data, the deformation condition of the tunnel wall surface can be analyzed more accurately, especially when the wall surface with a complex geometrical shape is processed, a more comprehensive space information preprocessing link can be provided, and the processing efficiency of the tunnel wall surface is improved. And in combination with an image registration technology in image preprocessing, the analysis efficiency among multiple images can be improved, and repeated processing caused by image differences is avoided, so that the deformation detection precision is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of tunnel detection, and in particular to a tunnel wall deformation detection method based on computer image recognition. Background Art

[0002] In recent years, with the development of science and technology, the country's infrastructure construction has continued to increase, and more and more tunnels have been built in the fields of highways, railways, and subways. The safety of tunnels directly affects the smooth flow of roads below and the safety of passing vehicles, and has a significant impact on the safety of people's lives and property and economic construction. Once a collapse occurs, it will cause inestimable direct and indirect losses. Therefore, tunnel deformation monitoring is a top priority in the tunnel construction process and subsequent tunnel maintenance process.

[0003] With the rapid development of computer image recognition technology, its application in tunnel wall deformation detection has great potential. Computer image recognition technology can quickly and accurately obtain the image information of the tunnel wall, and detect the deformation of the wall by analyzing and processing the image. However, the current tunnel wall deformation detection method based on computer image recognition still has some shortcomings, such as the inability to pre-process the image during image acquisition, which affects the accuracy of detection. Summary of the invention

[0004] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a tunnel wall deformation detection method based on computer image recognition.

[0005] In order to achieve the above object, the present invention adopts the following technical solution: a tunnel wall deformation detection method based on computer image recognition, comprising the following steps:

[0006] S1: Image acquisition system setup: multiple image acquisition devices are installed in the tunnel, and image acquisition with multi-source data fusion is performed;

[0007] S2: Image preprocessing: graying, filtering, denoising and image registration are performed on the collected tunnel wall images. The image registration includes feature-based image registration and elastic deformation registration for images of walls with large deformations.

[0008] S3: Wall feature extraction: Use edge detection algorithm and feature point marking algorithm to extract edge features and feature points of tunnel wall, and construct feature template of tunnel wall;

[0009] S4: Deformation detection: Regularly collect tunnel wall images, perform preprocessing and feature extraction, match them with the initial feature template, and calculate the displacement of feature points to determine the deformation;

[0010] S5: Deformation warning and report: Send out a deformation warning signal according to the pre-set deformation threshold and generate a detection report.

[0011] further:

[0012] The image acquisition device in step S1 uses a high-definition camera, and the installation of the high-definition camera should be perpendicular to the tunnel wall.

[0013] further:

[0014] The image acquisition of multi-source data fusion in step S1 uses laser radar technology to collect three-dimensional point cloud data. The laser radar technology obtains three-dimensional point cloud data of the tunnel wall and fuses it with the image data.

[0015] further:

[0016] The edge detection algorithm in step S3 combines the Canny edge detection algorithm with the morphological gradient edge detection method, and optimizes the edge detection result using the convolutional neural network in deep learning.

[0017] further:

[0018] The feature point marking algorithm in step S3 adopts the SIFT algorithm, and combines it with the feature extraction technology in deep learning, uses a pre-trained deep neural network model to extract features from the tunnel wall image, and then fuses the extracted features with the feature points marked by the SIFT algorithm.

[0019] further:

[0020] In step S4, the nearest neighbor matching algorithm based on the descriptor is used to match the initial feature template, and the RANSAC algorithm is used to remove the wrong matches. The formula of the nearest neighbor matching algorithm is:

[0021] Among them, NN(q): refers to the nearest neighbor point of the query point q; q: refers to the query point, that is, the feature point to be matched; x i : refers to the points in the data set, that is, the feature points extracted from the image; : refers to the query point q and the data point x i The square of the Euclidean distance between them.

[0022] The formula of the RANSAC algorithm is:

[0023] Where model refers to the model fitted by the RANSAC algorithm; M refers to the model parameters; I M : refers to the internal point set, that is, the point set that fits the model well; yi : refers to the observed value; f(x i ;M): refers to the model parameters M and input x i Calculated predicted values; : refers to the observed value y i and the predicted value f(x i ;M) is the square of the error between them.

[0024] further:

[0025] The detection report in step S5 includes the tunnel wall deformation position, deformation amount, deformation mode, deformation trend and treatment suggestions.

[0026] The present invention has the following beneficial effects:

[0027] 1. Compared with the prior art, the present invention not only performs image acquisition in the image acquisition link, but also uses laser radar technology to assist image acquisition. Laser radar can obtain three-dimensional point cloud data of the tunnel wall and fuse it with the image data. By matching the features in the image with the spatial coordinates in the three-dimensional point cloud data, the deformation of the tunnel wall can be analyzed more accurately, especially when dealing with walls with complex geometric shapes. It can provide a more comprehensive spatial information preprocessing link, and combined with the use of image registration technology in image preprocessing, it can improve the analysis efficiency between multiple images, avoid repeated processing due to image differences, and thus improve the accuracy of deformation detection.

[0028] 2. Compared with the prior art, the present invention adopts deep optimization and innovation of multiple algorithms in the wall feature extraction link, and uses convolutional neural network to optimize the edge detection results in the edge detection link, which can remove pseudo edges and improve the accuracy of edge detection, thereby providing a more reliable basis for subsequent tunnel deformation detection; SIFT algorithm is adopted in the feature point marking link, and combined with the feature extraction technology in deep learning, a pre-trained deep neural network model is used to extract features of the tunnel wall image, and then the extracted features are fused with the feature points marked by the SIFT algorithm, so that the powerful feature representation ability of the deep learning model can be utilized to improve the accuracy and stability of feature point marking. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 The present invention is a flow chart of the method. DETAILED DESCRIPTION

[0030] Reference Figure 1 The present invention provides a method for detecting tunnel wall deformation based on computer image recognition, comprising the following steps:

[0031] S1: Image acquisition system setting: multiple image acquisition devices are installed in the tunnel, and image acquisition with multi-source data fusion is performed; the image acquisition device in step S1 uses a high-definition camera, and the installation of the high-definition camera should be perpendicular to the tunnel wall; the image acquisition with multi-source data fusion in step S1 uses laser radar technology to collect three-dimensional point cloud data, and the laser radar technology obtains the three-dimensional point cloud data of the tunnel wall and fuses it with the image data.

[0032] Specifically, multiple image acquisition devices are installed in the tunnel, and the layout of the image acquisition devices is reasonably planned according to factors such as the length, shape and curvature of the tunnel to ensure that images of all parts of the tunnel wall can be effectively captured; after the image acquisition device is installed, the tunnel wall is imaged. In addition to single image acquisition, multi-source data fusion image acquisition is also performed simultaneously. When performing multi-source data fusion image acquisition, laser radar technology is used to assist image acquisition. Laser radar can obtain three-dimensional point cloud data of the tunnel wall and fuse it with the image data. By matching the features in the image with the spatial coordinates in the three-dimensional point cloud data, the deformation of the wall can be analyzed more accurately, especially when dealing with walls with complex geometric shapes, more comprehensive spatial information can be provided.

[0033] S2: Image preprocessing: grayscale, filter denoising and image registration are performed on the collected tunnel wall images. Image registration includes feature-based image registration and elastic deformation registration for images with large deformation walls.

[0034] Specifically, when performing image preprocessing, grayscale processing is first performed. When using the weighted average method for grayscale processing, the weighting coefficient can be fine-tuned for different types of tunnel wall materials. For example, for concrete walls, since their reflectivity to red light is relatively high, the weighting coefficient of the red component can be appropriately increased, and the formula can be adjusted to Gray = 0.32*R+0.55*G+0.13*B. This can better preserve the characteristic information of the concrete wall in the grayscale image.

[0035] Next, filtering and denoising are performed. Gaussian filtering is used to smooth the image. The size of the filter kernel can be selected according to the estimated intensity of the noise, so that the noise in the image can be removed and the signal-to-noise ratio of the image can be improved.

[0036] Finally, the images are registered. When multiple tunnel wall images are collected, these images need to be registered because there may be slight differences in the position and angle of the image acquisition device. The feature-based image registration method is used to extract feature points in the image, such as corner points, edge points, etc., and then use the correspondence between these feature points to map different images to the same coordinate system, providing an accurate image basis for subsequent deformation detection. For example, suppose images I1 and I2, extract feature points such as corner points and edge points p 1i and p 2j (i, j are the sequence numbers of the feature points), and the transformation matrix T is found through the correspondence between the feature points, so that p 2j =T×p 1i , I2 is mapped to the coordinate system of I1. For wall images with large deformation, the elastic deformation registration method is used. This method can handle the nonlinear deformation relationship between images. By establishing an elastic deformation model, the deformed image is registered to the coordinate system of the initial image, so as to more accurately analyze the deformation area and deformation degree of the wall. For example, suppose the deformed image I d , initial image I0, establish elastic deformation model E (I d ,I0), by minimizing the energy function E d Align to the I0 coordinate system.

[0037] S3: Wall feature extraction: The edge features and feature points of the tunnel wall are extracted using the edge detection algorithm and the feature point marking algorithm, and a feature template of the tunnel wall is constructed; the edge detection algorithm in step S3 combines the Canny edge detection algorithm with the morphological gradient edge detection method, and uses the convolutional neural network in deep learning to optimize the edge detection result; the feature point marking algorithm in step S3 uses the SIFT algorithm, combined with the feature extraction technology in deep learning, and uses the pre-trained deep neural network model to extract features from the tunnel wall image, and then fuses the extracted features with the feature points marked by the SIFT algorithm.

[0038] Specifically, in the application of Canny edge detection algorithm in tunnel wall detection, the parameter selection of Gaussian filter has a great influence on the edge detection results. The standard deviation of Gaussian filter can be optimized by analyzing the noise characteristics and edge blur of tunnel wall image. For example, when the noise of tunnel wall image is large and the edge is fuzzy, a larger standard deviation is selected, such as 1.5 - 2.0; when the noise is small and the edge is relatively clear, a smaller standard deviation is selected, such as 0.8 - 1.2. After calculating the gradient amplitude and direction, the threshold of non-maximum suppression can also be dynamically adjusted according to the specific situation of the image. For tunnel wall images with rich textures, the threshold of non-maximum suppression can be appropriately increased to reduce the detection of pseudo edges.

[0039] When extracting feature points using the SIFT algorithm, in order to improve the stability and accuracy of the feature points, the way of constructing the scale space can be improved. For example, when constructing the scale space in the tunnel wall image, more intermediate scales are added to make the scale change more continuous, so that feature points at different scales can be detected more accurately, which can ensure the accuracy of the feature points while improving the detection speed.

[0040] When constructing the feature template of the tunnel wall, in addition to the marked feature points of the tunnel wall image and their related feature information, the geometric relationship information between the feature points can also be added. For example, the distance and angle relationship between the feature points. These geometric relationship information can provide more references in the subsequent tunnel wall deformation detection and improve the accuracy of deformation detection.

[0041] S4: Deformation detection: Regularly collect tunnel wall images, perform preprocessing and feature extraction, match them with the initial feature template, and calculate the displacement of the feature points to determine the deformation; in step S4, the nearest neighbor matching algorithm based on the descriptor is used to match the initial feature template, and the RANSAC algorithm is used to remove wrong matches. The formula of the nearest neighbor matching algorithm is:

[0042] Among them, NN(q): refers to the nearest neighbor point of the query point q; q: refers to the query point, that is, the feature point to be matched; x i : refers to the points in the data set, that is, the feature points extracted from the image; : refers to the query point q and the data point x i The square of the Euclidean distance between them.

[0043] The formula of the RANSAC algorithm is:

[0044] Where model refers to the model fitted by the RANSAC algorithm; M refers to the model parameters; I M : refers to the internal point set, that is, the point set that fits the model well; y i : refers to the observed value; f(x i ;M): refers to the model parameters M and input x i Calculated predicted values; : refers to the observed value y i and the predicted value f(x i ;M) is the square of the error between them.

[0045] Specifically, regular deformation detection is carried out. According to the real-time monitoring data of the tunnel, such as changes in traffic flow, slight changes in geological conditions (obtained through surrounding geological monitoring equipment), etc., the deformation detection cycle is dynamically adjusted; the tunnel wall image is regularly collected, preprocessed and feature extracted, and then matched with the initial feature template, using the nearest neighbor matching algorithm based on the descriptor, and the RANSAC algorithm is used to remove the wrong match. The deformation of the tunnel wall is determined by the position change of the matching feature points. For the successfully matched feature points, the displacement in the horizontal and vertical directions is calculated, and the displacement is calculated using the Euclidean distance formula. According to the displacement of the feature points, the deformation mode of the tunnel wall is analyzed to determine whether it is local deformation or overall deformation, uniform deformation or non-uniform deformation, etc.

[0046] S5: deformation warning and reporting: a deformation warning signal is issued according to a preset deformation threshold, and a detection report is generated; the detection report in step S5 includes the deformation position, deformation amount, deformation mode, deformation trend and treatment suggestions of the tunnel wall.

[0047] Specifically, the deformation threshold is set in a personalized way. The level of the warning signal can be divided according to the degree to which the deformation exceeds the threshold. For example, when the deformation exceeds the threshold but is less than [X]%, a first-level warning is issued to prompt relevant personnel to pay attention; when the deformation exceeds the threshold by more than [X]%, a second-level warning is issued, requiring relevant personnel to conduct an immediate on-site inspection; and the setting of the deformation threshold is no longer based solely on conventional factors such as the tunnel's structural type, geological conditions, and service life, but should also consider factors such as the frequency of seismic activity in the area where the tunnel is located and the surrounding engineering construction. For example, in areas with frequent seismic activity, the deformation threshold should be appropriately lowered to increase the sensitivity of the warning; if there is large-scale construction or underground engineering construction in the surrounding area, the deformation threshold also needs to be adjusted accordingly. In addition to the basic deformation location, deformation amount, deformation mode, deformation trend, and treatment suggestions, the deformation detection report should also include information such as the time of image acquisition, the traffic flow at the time, and the ambient temperature and humidity in the tunnel. This additional information helps to more comprehensively analyze the causes of tunnel wall deformation and provide a more detailed decision-making basis for tunnel maintenance and management.

[0048] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for detecting tunnel wall deformation based on computer image recognition, characterized in that: The following steps are involved: S1: Image acquisition system setup: multiple image acquisition devices are installed in the tunnel, and image acquisition with multi-source data fusion is performed; S2: Image preprocessing: graying, filtering, denoising and image registration are performed on the collected tunnel wall images. The image registration includes feature-based image registration and elastic deformation registration for images of walls with large deformations. S3: Wall feature extraction: Use edge detection algorithm and feature point marking algorithm to extract edge features and feature points of tunnel wall, and construct feature template of tunnel wall; S4: Deformation detection: Regularly collect tunnel wall images, perform preprocessing and feature extraction, match them with the initial feature template, and calculate the displacement of feature points to determine the deformation; S5: Deformation warning and report: Send out a deformation warning signal according to the pre-set deformation threshold and generate a detection report.

2. The method for detecting tunnel wall deformation based on computer image recognition according to claim 1, characterized in that: The image acquisition device in step S1 uses a high-definition camera, and the installation of the high-definition camera should be perpendicular to the tunnel wall.

3. The method for detecting tunnel wall deformation based on computer image recognition according to claim 1, characterized in that: The image acquisition of multi-source data fusion in step S1 uses laser radar technology to collect three-dimensional point cloud data. The laser radar technology obtains three-dimensional point cloud data of the tunnel wall and fuses it with the image data.

4. The method for detecting tunnel wall deformation based on computer image recognition according to claim 1, characterized in that: The edge detection algorithm in step S3 combines the Canny edge detection algorithm with the morphological gradient edge detection method, and optimizes the edge detection result using the convolutional neural network in deep learning.

5. The method for detecting tunnel wall deformation based on computer image recognition according to claim 1, characterized in that: The feature point marking algorithm in step S3 adopts the SIFT algorithm, and combines it with the feature extraction technology in deep learning, uses a pre-trained deep neural network model to extract features from the tunnel wall image, and then fuses the extracted features with the feature points marked by the SIFT algorithm.

6. The method for detecting tunnel wall deformation based on computer image recognition according to claim 1, characterized in that: In step S4, the nearest neighbor matching algorithm based on the descriptor is used to match the initial feature template, and the RANSAC algorithm is used to remove the wrong matches. The formula of the nearest neighbor matching algorithm is: ， Among them, NN(q): refers to the nearest neighbor point of the query point q; q: refers to the query point, that is, the feature point to be matched; x i : refers to the points in the data set, that is, the feature points extracted from the image; : refers to the query point q and the data point x i The square of the Euclidean distance between them. The formula of the RANSAC algorithm is: ， Where model refers to the model fitted by the RANSAC algorithm; M refers to the model parameters; I M : refers to the internal point set, that is, the point set that fits the model well; y i : refers to the observed value; f(x i ;M): refers to the model parameters M and input x i Calculated predicted values; : refers to the observed value y i and the predicted value f(x i ;M) is the square of the error between them.

7. The method for detecting tunnel wall deformation based on computer image recognition according to claim 1, characterized in that: The detection report in step S5 includes the tunnel wall deformation position, deformation amount, deformation mode, deformation trend and treatment suggestions.

Images