Bridge displacement monitoring system and method based on visual guidance

Abstract

The invention relates to the technical field of bridge monitoring, in particular to a bridge displacement monitoring system and method based on visual guidance, and the system comprises an image collection module which is used for collecting a bridge structure image and illumination data around a bridge; the light field analysis module is used for separating noise signals from structure signals; the self-adaptive denoising module is used for calculating noise characteristics of different regions of the bridge structure image, performing region division on the bridge structure image based on the noise characteristics, and applying a non-local mean denoising technology to each region of the divided bridge structure image to obtain the bridge structure image; performing intelligent denoising processing on each region of the divided bridge structure image through the optimal regression model; the edge detection module is used for carrying out bridge edge contour extraction; and the monitoring module outputs a bridge displacement monitoring result. According to the invention, through high-precision and high-adaptability vision-oriented segmented denoising and continuous linear feature extraction, the method is used for bridge full-field displacement monitoring in a complex light field environment.

Description

Technical Field

[0001] The present invention relates to the technical field of bridge monitoring, and in particular to a bridge displacement monitoring system and method based on vision guidance. Background Art

[0002] Bridge displacement monitoring technology is an important means to ensure the safety and stability of bridge structures. Existing bridge displacement monitoring technologies are mainly divided into two categories: contact and non-contact.

[0003] Among them, the contact method usually relies on sparse measurement points where sensors are arranged on the bridge structure to obtain structural responses. This type of method is difficult to accurately reflect the status of the entire bridge structure due to the incompleteness of the test data, especially in large-span bridges. Multi-point layout not only increases costs, but also faces problems such as wiring transmission and installation difficulties.

[0004] The non-contact monitoring method obtains the full-field displacement data of the bridge through technologies such as 3D laser scanning, radar systems and photogrammetry. Although 3D laser scanning technology is suitable for static displacement monitoring, its scanning speed is slow and its ability to capture dynamic deformation (such as vibration and instantaneous deformation) is limited. At the same time, the equipment and data processing costs are high. Although the radar system can achieve millimeter-level precision monitoring, its equipment is expensive and it is difficult to balance economy and applicability.

[0005] Photogrammetry technology has high resolution and real-time monitoring capabilities, and can capture the full-field displacement of bridges in complex environments through video. However, due to the complexity of the environment in which the bridge is located, there is a common problem of significant noise in the image, especially under strong light, shadows and variable lighting conditions. The coupling of noise and structural feature signals significantly affects the accuracy and continuity of feature extraction.

[0006] Existing image noise processing methods are mainly divided into global processing methods and local processing methods. Global processing methods (such as histogram thresholding and frequency domain filtering) are prone to segmentation errors of noise and structural features under uneven lighting conditions. On the other hand, global processing methods usually rely on fixed parameters or rules and lack the ability to adapt to local changes. Although local processing methods (such as local mean filtering and non-local mean filtering) can be adjusted according to regional characteristics, they may cause inconsistent processing results between regions, resulting in discontinuities in edges or textures. On the other hand, local processing methods usually need to adjust parameters according to the characteristics of the local area, such as window size, threshold or filter strength. How to choose appropriate parameters to adapt to the characteristics of different regions is another challenge.

[0007] In summary, non-contact monitoring technology generally has the following disadvantages:

[0008] 1. Existing monitoring equipment is not adaptable enough to complex light field conditions and it is difficult to ensure data accuracy in a strong noise environment.

[0009] 2. Image denoising technology in large-scale bridge health monitoring has significant limitations in balancing noise suppression and detail retention, which can easily lead to blurring and fragmentation of feature extraction.

[0010] 3. There is a lack of full-field displacement monitoring methods that can comprehensively consider different light field characteristics and adaptively adjust denoising parameters. Summary of the invention

[0011] To solve the above problems, the present invention provides a bridge displacement monitoring system and method based on vision guidance, which is used for full-field bridge displacement monitoring in complex light field environments through high-precision and highly adaptable vision-guided segmented denoising and continuous linear feature extraction.

[0012] In order to achieve the above object, the technical solution of the present invention is as follows:

[0013] On the one hand, a bridge displacement monitoring system based on vision guidance is provided, comprising:

[0014] An image acquisition module, used to acquire images of the bridge structure and illumination data around the bridge;

[0015] Light field analysis module, used to analyze the light intensity distribution and its dynamic changes in the bridge structure image, identify the distribution differences between noise and structural features, and separate noise signals from structural signals;

[0016] Adaptive denoising module, used to calculate the noise characteristics of different areas of the bridge structure image, and divide the bridge structure image into regions based on the noise characteristics, apply non-local mean denoising technology to each area of ​​the divided bridge structure image, and perform intelligent denoising on each area of ​​the divided bridge structure image through the optimal regression model;

[0017] An edge detection module is used to extract the edge contour of the bridge from the bridge structure image after intelligent denoising;

[0018] The monitoring module is used to compare the edge profile of the bridge at different times and output the bridge displacement monitoring results.

[0019] Furthermore, the image acquisition module is used to obtain the bridge structure image collected by the high-resolution camera; and is also used to obtain the lighting data around the bridge collected by the meteorological station.

[0020] Furthermore, the light field analysis module is used to extract the region of interest from the bridge structure image and calculate the total light intensity of each row of the bridge structure image:

[0021]

[0022] Among them, I v(y) is the total light intensity of the yth row, W is the width of the bridge structure image, and I(x, y) represents the light intensity at the coordinate (x, y);

[0023] Compute the second-order derivative of the light field distribution to extract structural features:

[0024]

[0025] Identify edge contours of bridge structures;

[0026] Calculate the extreme points of the vertical light field distribution, calculate the local histogram near the extreme points, and calculate the skewness of the bridge structure image from the local histogram:

[0027]

[0028] Where n is the sample size of the specified window near the extreme point, (n-1)(n-2) is the correction factor used to correct the deviation caused by small sample size; x i is the second-order derivative value of each point in the window, is the average value of all second-order derivative values ​​in the window, and σ is the distribution range or discreteness of the second-order derivative values ​​in the window;

[0029] Split based on median skewness:

[0030]

[0031] Among them, Segmentation Point is the segmentation point set, cp i represents the i-th potential segmentation point, Sk i is the absolute value of the skewness of the i-th potential segmentation point, Represents the median of the absolute values ​​of all skewnesses.

[0032] Furthermore, the adaptive denoising module is used to perform local variance analysis on the vertically segmented image segments and calculate the noise intensity of each image segment:

[0033]

[0034] Among them, σ 2 (i, j) is the local variance of pixel (i, j) in the specified window, I(p, q) is the pixel value in the image window, and μ(i, j) is the average pixel brightness in the window;

[0035] Average each column of the local variance matrix for each vertical segment:

[0036]

[0037] Where M is the total number of rows of the vertical segmentation, and V(i, j) is the local variance value at the i-th row and j-th column;

[0038] Convert the average value result into a one-dimensional data array. Each element of the one-dimensional data array represents the average local variance of the corresponding column, and use it as the variability index for each column.

[0039] Furthermore, use the one-dimensional local variance matrix to perform clustering on the vertical segmentation:

[0040]

[0041] Where x represents the evaluation point in the feature space, x i is the number of data samples, K is the kernel function, a non-negative function satisfying ∫K(x)dx = 1, h is the bandwidth, and d is the dimension of the data space;

[0042] For any x, its Mean Shift vector M(x) is defined as:

[0043]

[0044] Update the position of each sample point iteratively through the Mean Shift vector until convergence. The update rule is:

[0045] x new = x + M(x)

[0046] During the process of iteratively updating the position of each sample point through the Mean Shift vector, move the data points to the local maximum of the density, connect the points with the maximum density, and form the clustering result, thereby obtaining the regional division result of the bridge structure image.

[0047] Furthermore, the adaptive denoising module is used to extract image patches based on the regional division result of the bridge structure image, and use non-local means denoising to process the image patches:

[0048]

[0049] Where I denoised (i) is the denoised value of pixel i, I(j) is the original value of pixel j in the image, w(i, j) is the weight calculated based on the similarity between pixel i and j, C(i) is the normalization constant to ensure that the sum of weights is 1, and N(i) represents the set of surrounding pixels considered for denoising.

[0050] Furthermore, the adaptive denoising module is used to set the range of the denoising intensity h for the image patches. The value of the denoising intensity h increases from 0 to 50 with a step size of 1. Traverse each image patch, apply a series of denoising intensities h, and record the peak signal-to-noise ratio after denoising;

[0051] Select the optimal regression model by calculating the mean square error and coefficient of determination of logarithmic regression, exponential regression and polynomial regression models;

[0052] De-noise each image block separately, evaluate the impact of different denoising strengths h, and establish a mapping relationship between the image block cluster center and the optimal denoising strength h;

[0053] The optimal regression model is trained based on the recorded cluster centers of the image blocks, the corresponding denoising intensity h, and the corresponding peak signal-to-noise ratio; the selected optimal regression model is used to estimate the logarithmic relationship between the cluster centers and the denoising intensity h, thereby completing the adaptive denoising process for the image blocks.

[0054] Furthermore, the edge detection module is used to calculate the gradient amplitude and direction of the bridge structure image after intelligent denoising:

[0055]

[0056] Among them, G x and G y are the gradients in the x-direction and y-direction respectively, and I is the bridge structure image after intelligent denoising;

[0057]

[0058] Where G is the gradient amplitude and θ is the direction;

[0059] Non-maximum suppression and double threshold method are applied successively to identify potential edges, and bridge edge tracking is completed by suppressing isolated weak edges.

[0060] On the other hand, a bridge displacement monitoring method based on vision guidance is provided, comprising the following steps:

[0061] Collect images of bridge structures and lighting data around the bridge;

[0062] Analyze the light intensity distribution and its dynamic changes in the bridge structure image, identify the distribution differences between noise and structural features, and separate noise signals from structural signals;

[0063] Calculate the noise characteristics of different areas of the bridge structure image, and divide the bridge structure image into regions based on the noise characteristics. Apply the non-local mean denoising technology to each area of ​​the divided bridge structure image, and perform intelligent denoising on each area of ​​the divided bridge structure image through the optimal regression model.

[0064] Extract the edge contour of the bridge from the bridge structure image after intelligent denoising;

[0065] Compare the edge profiles of the bridge at different times and output the bridge displacement monitoring results.

[0066] Furthermore, the bridge structure image is divided into regions, including vertical segments:

[0067] Extract the region of interest from the bridge structure image and calculate the total light intensity of each row of the bridge structure image:

[0068]

[0069] Among them, I v (y) is the total light intensity of the yth row, W is the width of the bridge structure image, and I(x, y) represents the light intensity at the coordinate (x, y);

[0070] Compute the second-order derivative of the light field distribution to extract structural features:

[0071]

[0072] Identify edge contours of bridge structures;

[0073] Calculate the extreme points of the vertical light field distribution, calculate the local histogram near the extreme points, and calculate the skewness of the bridge structure image from the local histogram:

[0074]

[0075] Where n is the sample size of the specified window near the extreme point, (n-1)(n-2) is the correction factor used to correct the deviation caused by small sample size; x i is the second-order derivative value of each point in the window, is the average value of all second-order derivative values ​​in the window, and σ is the distribution range or discreteness of the second-order derivative values ​​in the window;

[0076] Split based on median skewness:

[0077]

[0078] Among them, Segmentation Point is the segmentation point set, cp i represents the i-th potential segmentation point, Sk i is the absolute value of the skewness of the ith potential segmentation point, Represents the median of the absolute values ​​of all skewnesses.

[0079] The solution of the present invention has the following beneficial effects:

[0080] 1. The present invention uses a method combining light field skewness histogram analysis with adaptive denoising technology to effectively separate noise and structural feature signals under complex lighting conditions. Experiments show that the proposed segmented denoising algorithm increases the amount of image feature points extracted by 35% compared with traditional methods and effectively suppresses the influence of background noise.

[0081] 2. The present invention combines local variance analysis with non-local mean denoising, and the system can intelligently adjust the denoising intensity and effectively retain the edge details of the structure. This method improves the continuity of the structure contour by 60% under different lighting conditions, ensuring that the structural features extracted in complex environments are more complete.

[0082] 3. The system of the present invention can adaptively suppress noise according to the illumination changes in the local area of ​​the image by intelligently adjusting the denoising parameters. It is not only applicable to static conditions, but also to dynamically changing illumination conditions. Compared with traditional methods, the inventive scheme shows higher stability and robustness under various illumination intensities.

[0083] 4. The present invention realizes the accurate monitoring of the full-field displacement of the bridge in a complex light field environment, and the measurement error can be controlled within 3%, which is significantly lower than the error range of traditional sensors. This method can provide more accurate data support for the health monitoring of bridge structures and provide a reliable basis for structural damage detection and safety assessment.

[0084] 5. The present invention adopts a high-resolution camera and an automated image processing module. The system can capture bridge displacement information in real time and automatically perform data analysis and processing, which reduces the need for manual intervention and greatly improves monitoring efficiency.

[0085] In summary, the technical solution of the present invention not only effectively solves the problems of noise interference and discontinuous structural feature extraction in the prior art, but also significantly improves the accuracy, continuity and stability of bridge displacement monitoring. It has broad application prospects, and is particularly suitable for high-precision bridge monitoring and structural health management.

[0086] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] Figure 1 A schematic diagram of the system structure of an embodiment of the present invention;

[0088] Figure 2 is a general flow chart of the method of an embodiment of the present invention;

[0089] Figure 3 A flow chart of a method for distinguishing noise from structural characteristic signals according to an embodiment of the present invention;

[0090] Figure 4 A method flow chart of an adaptive denoising process according to an embodiment of the present invention;

[0091] Figure 5 A flow chart of a method for extracting continuous contour lines according to an embodiment of the present invention;

[0092] Figure 6 Schematic diagram of the total number of pixels at the lower edge of the beam and the number of pixel breakages at different loads and 157 Lux illumination according to an embodiment of the present invention;

[0093] Figure 7 This is a schematic diagram of the total number of pixels at the lower edge of the beam and the number of pixel breakages at different loads and 4000 Lux illumination according to an embodiment of the present invention;

[0094] Figure 8 This is a schematic diagram of the total number of pixels at the lower edge of the beam and the number of pixel breakages under different loads and 8000 Lux illumination according to an embodiment of the present invention;

[0095] Fig. 9 This is a schematic diagram of the total number of pixels at the lower edge of the beam and the number of pixel breakages under different loads and 12000 Lux illumination according to an embodiment of the present invention. DETAILED DESCRIPTION

[0096] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. 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.

[0097] The following is further described in detail through specific implementation methods:

[0098] Embodiment Combination Figure 1 and Figure 2 As shown: A bridge displacement monitoring system based on vision guidance is mainly composed of an image acquisition module, a light field analysis module, an adaptive denoising module, an edge detection module and a monitoring module:

[0099] The image acquisition module is used to collect bridge structure images and illumination data around the bridge. The bridge structure image is collected by a high-resolution industrial camera, preferably a Canon 5DSR, to ensure the quality of the collected bridge structure image. The illumination data around the bridge is collected by a small weather station, which can record real-time illumination data. Preferably, the small weather station is set around the bridge.

[0100] Combination Figure 3 As shown in the figure, the light field analysis module is used to analyze the light intensity distribution and its dynamic changes in the bridge structure image, identify the distribution differences between noise and structural features, and separate the noise signal from the structural signal. Specifically, firstly, the ROI (region of interest) automatic cropping algorithm is used to extract the region of interest of the acquired bridge structure image; then, it is divided into the following three steps:

[0101] Step 1: Calculate the total light intensity of the image in each row:

[0102]

[0103] Among them, I v (y) is the total light intensity of the yth row, W is the width of the bridge structure image, and i(x, y) represents the light intensity at the coordinate (x, y).

[0104] Step 2: Calculate the second-order derivative of the light field distribution to extract structural features:

[0105]

[0106] This step helps to accurately identify the edge contours of the bridge structure.

[0107] Step 3: Decoupling noise from structural features

[0108] The extreme points of the vertical light field distribution are calculated, and the local histogram near the extreme points is calculated; in the process of signal differentiation and denoising, the light field skewness histogram technology is used to further analyze the relationship between noise and structural features, and the skewness of the bridge structure image is calculated from the local histogram to reveal the asymmetry of the structural features in the image:

[0109]

[0110] Where n is the sample size of the specified window near the extreme point, (n-1)(n-2) is the correction factor used to correct the deviation caused by small sample size; x i is the second-order derivative value of each point in the window, is the average value of all second-order derivative values ​​in the window, and σ is the distribution range or discreteness of the second-order derivative values ​​in the window. A high skewness value indicates the edge characteristics of the structure, which can help further distinguish noise from structural information.

[0111] Split based on median skewness:

[0112]

[0113] Among them, Segmentation Point is the segmentation point set, cp i represents the i-th potential segmentation point, Sk i is the absolute value of the skewness of the ith potential segmentation point, Represents the median of the absolute values ​​of all skewnesses.

[0114] Combination Figure 4As shown, the adaptive denoising module is used to calculate the noise characteristics of different areas of the bridge structure image, and divide the bridge structure image into regions based on the noise characteristics, apply the non-local mean denoising technology to each area of ​​the divided bridge structure image, and perform intelligent denoising on each area of ​​the divided bridge structure image through the optimal regression model. Specifically:

[0115] First, local variance analysis is performed on the vertically segmented image segments to calculate the noise intensity of each image segment. The calculation formula is as follows:

[0116]

[0117] Among them, σ 2 (i, j) is the local variance of pixel (i, j) in the specified window, I(p, q) is the pixel value in the image window, and μ(i, j) is the average pixel brightness in the window.

[0118] After calculating the local variance, the average of each column of the local variance matrix of each vertical segment is calculated as follows:

[0119]

[0120] Where M is the total number of rows in the vertical segment, and V(i, j) is the local variance value in row i and column j. The average value of each column of the local variance matrix of each vertical segment is calculated and the result is converted into a one-dimensional data set. By calculating the average value, the overall variation characteristics in the vertical segment can be extracted.

[0121] Therefore, the average result is converted into a one-dimensional data array. By performing this operation on all columns j, a one-dimensional array is obtained, each element of which represents the average local variance of the corresponding column. This one-dimensional array summarizes the local variance characteristics of the entire vertical segment and provides a representative variability indicator for each column.

[0122] In the step of clustering the vertical segments, a one-dimensional local variance matrix is ​​used to cluster the vertical segments. This process is an adaptive clustering process, which is used to find clusters with similar characteristics based on local variance values ​​on the vertical segments of the image, thereby revealing the corresponding noise characteristics:

[0123]

[0124] Among them, x represents the evaluation point in the feature space, x i is the number of data samples, K is the kernel function, which is a non-negative function that satisfies ∫K(x)dx=1, h is the bandwidth, and d is the dimension of the data space. The bandwidth h determines the scale of the kernel function and affects the smoothness of the estimate. The kernel function K is used to calculate the distance from point x to sample point x. iThe contribution of the distance to the density estimation. Commonly used kernel functions include Gaussian kernels, which are expressed as:

[0125]

[0126] is the normalization coefficient, ensuring that the integral of the kernel function in the entire space is 1; This is the exponential part of the Gaussian function; where ||x|| 2 is the square of the Euclidean norm of vector x; it represents the square of the distance of x from the origin. The negative exponent indicates that the value of the kernel function decreases exponentially as x moves away from the center of the kernel.

[0127] The core of the Mean Shift algorithm is to calculate the Mean Shift vector and use this vector to update the position of the sample points. The Mean Shift vector points to the direction where the density of sample points increases. For any x, its Mean Shift vector M(x) is defined as:

[0128]

[0129] The position of each sample point can be iteratively updated through the Mean Shift vector until convergence. The update rule is:

[0130] x new =x+M(x)

[0131] This process will be repeated until the position change of all points is less than a certain threshold, or the predetermined number of iterations is reached. Through the Mean Shift iterative process, the data points can be moved to the local maximum value of the density, and the points belonging to the same density peak are considered to be the same class. Finally, the clustering result is formed by connecting the points with the maximum density; thus, the regional division result of the bridge structure image is obtained.

[0132] Then, image blocks are extracted based on the regional division results of the bridge structure image, and the image blocks are processed using non-local mean denoising, which removes noise based on the self-similarity within the image, rather than relying solely on the neighborhood of local pixels. The main advantage of this method is that it can better preserve image details and structures while denoising. Its expression is:

[0133]

[0134] Among them, I denoised where (i) is the denoised value of pixel i, I(j) is the original value of pixel j in the image, w(i, j) is the weight calculated based on the similarity between pixels i and j, C(i) is a normalization constant to ensure that the sum of the weights is 1, and N(i) represents the set of surrounding pixels considered for denoising.

[0135] In order to enable image blocks to be adaptively denoised, it is necessary to count the values ​​of the cluster centers of all image blocks and the PSNR (peak signal-to-noise ratio) values ​​corresponding to different denoising intensities h. By setting the range of denoising intensity h for the image blocks, the denoising intensity h value increases from 0 to 50 with a step size of 1, traverse each image block, apply a series of denoising intensities h, and record the peak signal-to-noise ratio after denoising. The optimal regression model is selected by calculating the mean square error and determination coefficient of the logarithmic regression, exponential regression and polynomial regression models; denoising is performed on each image block separately, and the influence of different denoising intensities h is evaluated to establish a mapping relationship between the image block cluster center and the optimal denoising intensity h; the optimal regression model is trained based on the recorded image block cluster center, the corresponding denoising intensity h and the corresponding peak signal-to-noise ratio; the selected optimal regression model is used to estimate the logarithmic relationship between the cluster center and the denoising intensity h, thereby completing the adaptive denoising process of the image block.

[0136] Combination Figure 5 As shown, the edge detection module is used to extract the edge contour of the bridge from the bridge structure image after intelligent denoising. Specifically:

[0137] By calculating the gradient amplitude and direction of the bridge structure image after intelligent denoising:

[0138]

[0139] Among them, G x and G y are the gradients in the x-direction and y-direction respectively, and I is the bridge structure image after intelligent denoising;

[0140]

[0141] Where G is the gradient amplitude and θ is the direction. Non-maximum suppression and double threshold methods are applied in sequence to identify potential edges, and bridge edge tracking is completed by suppressing isolated weak edges. After adaptive denoising of different clustered image blocks of vertical segments, the denoised image can locate the edge more accurately. This edge detection method that combines segmented denoising and adaptive threshold technology can effectively reduce the occurrence of false edges and broken edges while maintaining accurate edge positioning. It is suitable for edge detection tasks in complex environments and separates the signal coupling between structural contour features and strong image noise.

[0142] The monitoring module is used to compare the edge profiles of the bridge at different times and output the bridge displacement monitoring results. Specifically, by processing the data after denoising and edge detection, the displacement response information of the bridge is output with an accuracy of less than 3%. The data output module can automatically convert the processed results into displacement data and further optimize the monitoring results in combination with sensor data.

[0143] Combination Figure 6-Figure 9 As shown in the figure, the number of feature points and the number of fractures of the lower edge contour of the beam were measured under different loads (0kg, 30kg, 40kg, 48kg, 55kg) and different illuminations (lighting conditions are controlled by LED stepless dimming lamps) (157Lux, 4000Lux, 8000Lux, 12000Lux). It can be seen that the proposed measurement method is basically consistent with the number of pixels in the field of view of the visual camera, and there is almost no feature point loss and fracture phenomenon. Compared with the traditional method, the algorithm increases the extraction of image feature points by 35% and the continuity of the contour line by 60% under different illumination.

[0144] Tables 1 to 4 are comparisons of the displacements of the mid-span of the beam measured by different algorithms under different illuminations (157Lux, 4000Lux, 8000Lux, 12000Lux) and different loads (0kg, 30kg, 48kg, 55kg). The proposed measurement algorithm has no feature point loss under different illumination and load conditions, and the error is less than 3% compared with the standard displacement meter, which meets engineering requirements and has broad application prospects.

[0145] Table 1

[0146]

[0147] Table 2

[0148]

[0149] Table 3

[0150]

[0151] Table 4

[0152]

[0153] Obviously, the above embodiments are merely examples for the purpose of clear explanation, and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived therefrom are still within the scope of protection of the invention.

Claims

1. A bridge displacement monitoring system based on vision guidance, characterized in that: include: An image acquisition module, used to acquire images of the bridge structure and illumination data around the bridge; Light field analysis module, used to analyze the light intensity distribution and its dynamic changes in the bridge structure image, identify the distribution differences between noise and structural features, and separate noise signals from structural signals; Adaptive denoising module, used to calculate the noise characteristics of different areas of the bridge structure image, and divide the bridge structure image into regions based on the noise characteristics, apply non-local mean denoising technology to each area of ​​the divided bridge structure image, and perform intelligent denoising on each area of ​​the divided bridge structure image through the optimal regression model; An edge detection module is used to extract the edge contour of the bridge from the bridge structure image after intelligent denoising; The monitoring module is used to compare the edge profile of the bridge at different times and output the bridge displacement monitoring results.

2. The bridge displacement monitoring system based on vision guidance according to claim 1 is characterized in that: The image acquisition module is used to obtain the bridge structure image collected by the high-resolution camera; it is also used to obtain the lighting data around the bridge collected by the meteorological station.

3. The visual-guided bridge displacement monitoring system according to claim 1 is characterized in that: The light field analysis module is used to extract the region of interest from the bridge structure image and calculate the total light intensity of each row of the bridge structure image: Among them, I v (y) is the total light intensity of the yth row, W is the width of the bridge structure image, and i(x, y) represents the light intensity at the coordinate (x, y); Compute the second-order derivative of the light field distribution to extract structural features: Identify edge contours of bridge structures; Calculate the extreme points of the vertical light field distribution, calculate the local histogram near the extreme points, and calculate the skewness of the bridge structure image from the local histogram: Where n is the sample size of the specified window near the extreme point, (n-1)(n-2) is the correction factor used to correct the deviation caused by small sample size; x i is the second-order derivative value of each point in the window, is the average value of all second-order derivative values ​​in the window, and σ is the distribution range or discreteness of the second-order derivative values ​​in the window; Split based on median skewness: Among them, Segmentaiton Point is the segmentation point set, cp i represents the i-th potential segmentation point, Sk i is the absolute value of the skewness of the i-th potential segmentation point, Represents the median of the absolute values ​​of all skewnesses.

4. The visual-guided bridge displacement monitoring system according to claim 3 is characterized in that: The adaptive denoising module is used to perform local variance analysis on the vertically segmented image segments and calculate the noise intensity of each image segment: Among them, σ 2 (i, j) is the local variance of pixel (i, j) in the specified window, I(p, q) is the pixel value in the image window, and μ(i, j) is the average pixel brightness in the window; Average each column of the local variance matrix for each vertical segment: Where M is the total number of rows in the vertical segment, and V(i, j) is the local variance value in row i and column j; The average result is converted into a one-dimensional data array, each element of which represents the average local variance of the corresponding column, which is used as the variability indicator of each column.

5. The visual-guided bridge displacement monitoring system according to claim 4 is characterized in that: Cluster the vertical segments using a one-dimensional local variance matrix: Among them, x represents the evaluation point in the feature space, x i is the number of data samples, K is the kernel function, which is a non-negative function satisfying ∫K(x)dx=1, h is the bandwidth, and d is the dimension of the data space; For any x, the Mean Shift vector M(x) is defined as: The position of each sample point is iteratively updated through the Mean Shift vector until convergence. The update rule is: x new =x+M(x) In the process of iteratively updating the position of each sample point through the Mean Shift vector, the data point is moved to the local maximum value of the density, and the points with the maximum density are connected to form a clustering result, thereby obtaining the regional division result of the bridge structure image.

6. The visual-guided bridge displacement monitoring system according to claim 5 is characterized in that: The adaptive denoising module is used to extract image blocks based on the regional division results of the bridge structure image, and process the image blocks using non-local mean denoising: Among them, I denoised where (i) is the denoised value of pixel i, I(j) is the original value of pixel j in the image, w(i, j) is the weight calculated based on the similarity between pixels i and j, C(i) is a normalization constant to ensure that the sum of the weights is 1, and N(i) represents the set of surrounding pixels considered for denoising.

7. The visual-guided bridge displacement monitoring system according to claim 6 is characterized in that: The adaptive denoising module is used to set the range of denoising intensity h for the image block. The denoising intensity h value increases from 0 to 50 with a step size of 1. Each image block is traversed, a series of denoising intensity h is applied, and the peak signal-to-noise ratio after denoising is recorded; Select the optimal regression model by calculating the mean square error and coefficient of determination of logarithmic regression, exponential regression and polynomial regression models; De-noise each image block separately, evaluate the impact of different denoising strengths h, and establish a mapping relationship between the image block cluster center and the optimal denoising strength h; The optimal regression model is trained based on the recorded cluster centers of the image blocks, the corresponding denoising intensity h, and the corresponding peak signal-to-noise ratio; the selected optimal regression model is used to estimate the logarithmic relationship between the cluster centers and the denoising intensity h, thereby completing the adaptive denoising process for the image blocks.

8. The visual-guided bridge displacement monitoring system according to claim 1, characterized in that: The edge detection module is used to calculate the gradient amplitude and direction of the bridge structure image after intelligent denoising: Among them, G x and G y are the gradients in the x-direction and y-direction respectively, and I is the bridge structure image after intelligent denoising; Where G is the gradient amplitude and θ is the direction; Non-maximum suppression and double threshold method are applied successively to identify potential edges, and bridge edge tracking is completed by suppressing isolated weak edges.

9. A bridge displacement monitoring method based on vision guidance, characterized in that: The steps include: Collect images of bridge structures and lighting data around the bridge; Analyze the light intensity distribution and its dynamic changes in the bridge structure image, identify the distribution differences between noise and structural features, and separate noise signals from structural signals; Calculate the noise characteristics of different areas of the bridge structure image, and divide the bridge structure image into regions based on the noise characteristics. Apply the non-local mean denoising technology to each area of ​​the divided bridge structure image, and perform intelligent denoising on each area of ​​the divided bridge structure image through the optimal regression model. Extract the edge contour of the bridge from the bridge structure image after intelligent denoising; Compare the edge profiles of the bridge at different times and output the bridge displacement monitoring results.

10. The method for monitoring bridge displacement based on vision guidance according to claim 9, characterized in that: The bridge structure image is divided into regions, including: Extract the region of interest from the bridge structure image and calculate the total light intensity of each row of the bridge structure image: Among them, I v (y) is the total light intensity of the yth row, W is the width of the bridge structure image, and I(x, y) represents the light intensity at the coordinate (x, y); Compute the second-order derivative of the light field distribution to extract structural features: Identify edge contours of bridge structures; Calculate the extreme points of the vertical light field distribution, calculate the local histogram near the extreme points, and calculate the skewness of the bridge structure image from the local histogram: Where n is the sample size of the specified window near the extreme point, (n-1)(n-2) is the correction factor used to correct the deviation caused by small sample size; x i is the second-order derivative value of each point in the window, is the average value of all second-order derivative values ​​in the window, and σ is the distribution range or discreteness of the second-order derivative values ​​in the window; Split based on median skewness: Among them, Segmentation Point is the segmentation point set, cp i represents the i-th potential segmentation point, Sk i is the absolute value of the skewness of the i-th potential segmentation point, Represents the median of the absolute values ​​of all skewnesses.

Images