Bridge tower base uneven settlement monitoring method based on visual detection

Abstract

The application discloses a bridge tower base uneven settlement monitoring method based on visual detection and relates to the technical field of bridge structure health monitoring, and comprises the following steps: selecting natural feature points and stable reference points on the surface of a bridge tower base, arranging a camera system, collecting a reference image, collecting time sequence images through the camera system and pre-processing, matching the natural feature points through an improved SIFT feature extraction algorithm, estimating a homography matrix through the RANSAC algorithm in combination with the stable reference points, realizing the unification of coordinate systems and obtaining accurate feature point coordinates; comparing the feature point coordinates in the reference image and current image, calculating two-dimensional pixel displacement, converting the two-dimensional pixel displacement into a settlement amount in the vertical direction of the tower base based on camera parameters, performing time sequence analysis on the settlement amount to obtain a differential settlement value, and generating an uneven settlement early warning according to a preset multi-level safety threshold. The application realizes non-contact, high-precision and full-automatic monitoring of the uneven settlement of the bridge tower base.

Description

Technical Field

[0001] This invention relates to the technical field of bridge structural health monitoring, and in particular to a method for monitoring uneven settlement of bridge tower foundations based on visual inspection. Background Technology

[0002] In recent years, with the rapid development of computer vision technology, image processing algorithms, and high-resolution imaging sensors, vision-based detection methods have gradually become an important means of monitoring the health of bridge structures. Traditional methods for monitoring tower foundation settlement suffer from problems such as low efficiency, lack of real-time automation, and severe dependence on environmental conditions, making it difficult to capture the potential structural risk of uneven tower foundation settlement in a timely and effective manner. By deploying an automated image acquisition system, combined with precise target recognition and displacement calculation models, non-contact, high-frequency, and high-precision measurement of tower foundation deformation can be achieved, providing crucial data support for structural safety early warning and preventive maintenance.

[0003] Currently, Chinese invention patent CN120491062A discloses a method for assessing uneven settlement of bridges based on SAR images. This method extracts bridge information by processing SAR image data and uses time-series analysis technology to monitor the deformation of the bridge. It can achieve wide-area, non-contact monitoring of the structural health of bridges. Although it demonstrates the application potential of SAR technology in infrastructure monitoring, the related technology lacks the ability to specifically assess uneven settlement of bridge tower foundations. It lacks a precise physical model conversion from image displacement to vertical settlement. At the same time, due to the side-view geometry of SAR, its monitoring accuracy is easily affected by the complex structure of the bridge and track errors. It is difficult to effectively separate the overall displacement of the bridge from the differential settlement between key tower foundations, thus failing to provide accurate safety warnings for bridge structural risks. Summary of the Invention

[0004] The technical problem solved by this invention is that existing methods for monitoring uneven settlement of bridge tower foundations, such as visual monitoring with a single camera, have the fundamental flaw of being unable to distinguish between actual vertical settlement and horizontal rigid body motion. Methods relying on artificial targets are difficult to implement on existing bridges. Methods based on natural features are susceptible to environmental interference and have poor long-term stability. Traditional manual measurement is inefficient and cannot be automated. This invention aims to comprehensively solve the above problems and provide an automated monitoring method that is highly accurate, highly reliable, and has strong engineering applicability.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0006] A method for monitoring uneven settlement of bridge tower foundations based on visual inspection includes the following steps:

[0007] Step S1: Select natural feature points and deploy the camera system to acquire a reference image, which includes natural feature points and stable reference points;

[0008] Step S2: Acquire time-series images through the camera system, and preprocess the time-series images to obtain high-quality current images;

[0009] Step S3: The feature description vector is obtained by processing the benchmark image and the high-quality current image using the improved SIFT feature extraction algorithm;

[0010] The precise coordinates of the feature points are obtained by using a coordinate system based on the homography matrix and stable reference points.

[0011] Step S4: Compare the precise feature point coordinates with the natural feature point coordinates and calculate the two-dimensional pixel displacement. Then, convert the two-dimensional pixel displacement to obtain the tower foundation settlement in the direction perpendicular to the ground.

[0012] Step S5: Perform time-series analysis on the settlement amount to obtain differential settlement values, determine the threshold based on a preset safety threshold, and generate an uneven settlement early warning.

[0013] As a preferred embodiment of the visual detection-based bridge tower foundation uneven settlement monitoring method of the present invention, wherein: the reference image in step S1 includes natural feature points and stable reference points;

[0014] The natural feature points include the surface features of the tower base concrete, structural joint features, and features of ancillary facilities;

[0015] The surface features of the tower base concrete include high-contrast textured areas, durable stain patches, and intersections of native cracks;

[0016] The structural joint features include the edge of the construction joint, the anchor point of the expansion joint, and the corner point of the embedded part outline;

[0017] The features of the ancillary facilities include permanent maintenance ladder connecting bolts, safety railing mounting bases, and original monitoring point markings;

[0018] Based on preset camera parameters, this integrated device acquires reference images and time-series images through a camera system and performs image coordinate to spatial coordinate transformation.

[0019] The camera parameters include intrinsic camera parameters and extrinsic camera parameters;

[0020] The camera intrinsic parameters include focal length, principal point coordinates, and distortion coefficients, while the camera extrinsic parameters include camera mounting position and attitude parameters.

[0021] The attitude parameters include the azimuth, pitch, and roll angles of the camera beam relative to the world coordinate system.

[0022] The stable reference points include fixed points on natural exposed rock bases or hard concrete benchmark piers, artificial marker points formed by pre-embedded steel nails, bolts, and measuring piles, and permanently installed reflective sheets and prism positioning marker points.

[0023] As a preferred embodiment of the vision-based detection-based method for monitoring uneven settlement of bridge tower foundations according to the present invention, in step S2, the time-series image is preprocessed to obtain a high-quality current image, and the processing logic includes:

[0024] Distortion correction processing is performed on the acquired time-series images. The focal length, principal point coordinates, and distortion coefficients obtained from camera calibration are used to recover the distortion coordinates of each pixel in the time-series image to the true position through inverse distortion mapping, thus obtaining the distortion-corrected image.

[0025] The distortion-corrected image is converted into a grayscale image and the grayscale distribution probability of the entire image is statistically analyzed. The grayscale levels are then redistributed using the histogram equalization method to obtain a brightness-equalized image.

[0026] Noise filtering is performed on the brightness-equalized image, and the image is smoothed by convolution using a Gaussian filtering algorithm to obtain a high-quality current image;

[0027] The Gaussian filtering formula is as follows:

[0028] ;

[0029] in, Indicates the Gaussian filter weights. This indicates the center offset of the filter window. Indicates the standard deviation of the filter. Represents the natural constant.

[0030] As a preferred embodiment of the vision-based detection-based method for monitoring uneven settlement of bridge tower foundations according to the present invention, in step S3, a feature description vector is obtained by processing the reference image and the high-quality current image using an improved SIFT feature extraction algorithm. The calculation logic includes:

[0031] Centered on natural feature points, an arc window is used as the feature description region. The radius of the arc window is 24 pixels, and it is divided into 4 arc sub-regions from the inside out, with each arc differing by 6 pixels.

[0032] Calculate the magnitude and direction angle of the gradient of each pixel in each arc sub-region. The angle of each arc sub-region is divided into 30°. Calculate the gradient histogram of each arc sub-region in 12 directions and generate a 48-dimensional feature description vector.

[0033] The formula for calculating the magnitude of the pixel gradient is:

[0034] ;

[0035] The formula for calculating the direction angle of the pixel gradient is:

[0036] ;

[0037] in, Represents the value of the x-axis. This represents the value of the ordinate. Indicates coordinates grayscale value at that location Indicates in Gradient magnitude at point, Indicates coordinates The gradient direction angle;

[0038] The pixels within the arc window are weighted using a Gaussian weighting function to generate a reduced-dimensional feature description vector.

[0039] The formula for calculating the Gaussian weighting function is as follows:

[0040] ;

[0041] in, Indicates coordinates The weighting coefficients at the point, Indicates the location of natural feature points. The standard deviation of the Gaussian function is represented by... Indicates the reference x-axis, Indicates the reference ordinate. Represents the natural constant;

[0042] The feature description vector matching algorithm is used to match natural feature points in the reference image to obtain matching point pairs. The matching point pairs are then filtered according to the Euclidean distance ratio and cosine similarity to obtain accurate matching point pairs.

[0043] The formula for calculating cosine similarity is:

[0044] ;

[0045] in, Represents cosine similarity. The first feature descriptor vector representing the feature descriptor vector of natural feature points in the reference image. Each component value The first feature descriptor vector representing the feature description vector of natural feature points in a time series image. Each component value;

[0046] Based on a stable reference point, the homography matrix is ​​estimated using the RANSAC algorithm, and coordinate system unification is performed to obtain accurate feature point coordinates.

[0047] As a preferred embodiment of the vision-based detection method for monitoring uneven settlement of bridge tower foundations described in this invention, the Euclidean distance ratio and cosine similarity are used to filter the matching point pairs to obtain accurate matching point pairs. The processing logic includes:

[0048] The matching distance between the feature description vector and the corresponding feature description vector of the reference image is calculated using Euclidean distance, and the first minimum distance is selected for each pair of matching points. Second minimum distance ;

[0049] when When the value is less than the first threshold, the matching point pairs are retained to form the first set of matching point pairs, which is the set of all retained matching points.

[0050] when If the value is greater than or equal to the first threshold, the matching point pair is discarded.

[0051] After filtering by Euclidean distance ratio, the cosine similarity of the feature description vectors is calculated for the matching point pairs in the first set of matching point pairs.

[0052] When the cosine similarity is less than or equal to the second threshold, the matching point pair is deemed invalid, and the matching point pair is removed and deleted.

[0053] When the cosine similarity is greater than the second threshold, the matching point pair is determined to be valid, and an exact matching point pair is obtained.

[0054] As a preferred embodiment of the vision-based bridge tower foundation uneven settlement monitoring method described in this invention, the method involves: applying geometric constraints to precisely matched point pairs based on stable reference points; estimating the homography matrix using the RANSAC algorithm; and unifying the coordinates of precisely matched point pairs in the high-quality current image to the coordinate system of the reference image to obtain precise feature point coordinates.

[0055] As a preferred embodiment of the vision-based detection method for monitoring uneven settlement of bridge tower foundations according to the present invention, the coordinate unification processing logic includes:

[0056] Using stable reference points already contained in the baseline image, geometric constraints are applied to the precisely matched point pairs. The homography matrix is ​​estimated using the RANSAC algorithm. The processing logic includes:

[0057] Iterative calculations are performed using the RANSAC algorithm. At least four pairs of matching points containing stable reference points are arbitrarily selected to calculate the homography matrix. The projection error is calculated for all matching point pairs, and pairs with projection errors greater than a preset error threshold are discarded. Matching point pairs with projection errors not exceeding the preset error threshold are considered interior points. When the proportion of interior points to the total number of matching point pairs exceeds a preset proportion threshold, the estimated homography matrix is ​​deemed valid. The formula for the homography matrix is:

[0058] ;

[0059] in, Represents the homography matrix. Indicates the high-quality current image of the th The original x-coordinates of the natural feature points Indicates the high-quality current image of the th The original ordinates of the natural feature points This indicates that the homography matrix is ​​mapped to the reference image coordinate system. The x-coordinates of natural feature points This indicates that the homography matrix is ​​mapped to the reference image coordinate system. The ordinates of the natural feature points;

[0060] Coordinates of natural feature points in the high-quality current image Mapping to the coordinate system of the reference image yields precise feature point coordinates. .

[0061] As a preferred embodiment of the vision-based detection-based method for monitoring uneven settlement of bridge tower foundations according to the present invention, in step S4, the precise feature point coordinates are compared with the natural feature point coordinates, and the two-dimensional pixel displacement is calculated. The processing logic includes:

[0062] The coordinate system obtains the precise feature point coordinates of natural feature points in the current high-quality image. Coordinates of natural feature points in the reference image Point-by-point comparison is performed to calculate the 2D pixel displacement. The specific formula is as follows:

[0063] ;

[0064] ;

[0065] in, Indicates the high-quality current image of the th The pixel coordinates of each natural feature point Indicating the first in the reference image The pixel coordinates of each natural feature point Indicates the first The two-dimensional pixel displacement of a natural feature point in the horizontal direction of the image. Indicates the first The two-dimensional pixel displacement of a natural feature point in the vertical direction of the image;

[0066] Based on the camera intrinsic and extrinsic parameters in the camera parameters, the two-dimensional pixel displacement is projected and converted into the tower foundation settlement in the direction perpendicular to the ground. The formula for the tower foundation settlement is:

[0067] ;

[0068] in, Indicates the camera's focal length in the vertical direction. Indicates the optical center of the camera to the first The straight-line distance between each natural feature point on the monitoring surface This indicates the angle between the camera's optical axis and the normal to the monitoring surface. This indicates the amount of tower foundation settlement in the direction perpendicular to the ground.

[0069] As a preferred embodiment of the vision-based detection-based method for monitoring uneven settlement of bridge tower foundations according to the present invention, in step S5, the differential settlement value is obtained by performing time-series analysis on the settlement of the tower foundation, and the processing logic includes:

[0070] The settlement of the tower base at natural feature points during the monitoring period is stored in chronological order to form a settlement time series.

[0071] Based on the settlement time series, the difference between each natural feature point on the same tower base is calculated. The settlement difference between any two natural feature points at the same monitoring time is calculated. The absolute value of the settlement difference is taken to obtain the local differential settlement value of the corresponding tower base.

[0072] For each tower foundation of the bridge, the absolute value of the difference in settlement between any two natural feature points with the largest settlement at the same monitoring time is calculated to obtain the overall differential settlement value between the tower foundations.

[0073] As a preferred embodiment of the vision-based detection-based method for monitoring uneven settlement of bridge tower foundations according to the present invention, the processing logic for generating an early warning of uneven settlement based on a preset safety threshold includes:

[0074] A first-level safety threshold and a second-level safety threshold are set for the local differential settlement value of the tower base, wherein the second-level safety threshold is greater than the first-level safety threshold;

[0075] A third-level safety threshold and a fourth-level safety threshold are set for the overall differential settlement value between the tower bases, wherein the fourth-level safety threshold is greater than the third-level safety threshold;

[0076] When any local differential settlement value of the tower base exceeds the first-level safety threshold but does not exceed the second-level safety threshold, a local first-level early warning is generated;

[0077] When any local differential settlement value of the tower base exceeds the second-level safety threshold, a local second-level early warning is generated;

[0078] When any overall differential settlement value between tower bases exceeds the third-level safety threshold but does not exceed the fourth-level safety threshold, an overall first-level early warning is generated;

[0079] When the differential settlement value of any individual tower base exceeds the fourth-level safety threshold, an overall second-level early warning is generated.

[0080] The beneficial effects of this invention are as follows: This invention employs an improved SIFT feature extraction and a dual matching strategy combining Euclidean distance ratio and cosine similarity. It also introduces a stable reference point and homography matrix to establish a coordinate system. By establishing a precise physical transformation from two-dimensional pixel displacement to vertical settlement of the tower base, it solves the fundamental defect that single-camera visual monitoring cannot distinguish between real settlement and horizontal rigid body motion. Combined with time series analysis and multi-level safety thresholds, it realizes automated, real-time, accurate monitoring and intelligent early warning of local and overall uneven settlement of bridge tower bases, providing a full-process, highly reliable monitoring method for bridge structural health. Attached Figure Description

[0081] Figure 1 This is a schematic diagram of the basic process of a vision-based method for monitoring uneven settlement of bridge tower foundations, provided in one embodiment of the present invention. Detailed Implementation

[0082] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0083] Example 1, referring to Figure 1 As an embodiment of the present invention, a method for monitoring uneven settlement of bridge tower foundations based on visual inspection is provided, comprising the following steps:

[0084] Step S1: Select natural feature points and deploy the camera system to acquire a reference image, which includes natural feature points and stable reference points;

[0085] Step S2: Acquire time-series images through the camera system, and preprocess the time-series images to obtain high-quality current images;

[0086] Step S3: The feature description vector is obtained by processing the benchmark image and the high-quality current image using the improved SIFT feature extraction algorithm;

[0087] The precise coordinates of the feature points are obtained by using a coordinate system based on the homography matrix and stable reference points.

[0088] Step S4: Compare the precise feature point coordinates with the natural feature point coordinates and calculate the two-dimensional pixel displacement. Then, convert the two-dimensional pixel displacement to obtain the tower foundation settlement in the direction perpendicular to the ground.

[0089] Step S5: Perform time-series analysis on the settlement amount to obtain differential settlement values, determine the threshold based on the preset safety threshold, and generate an uneven settlement early warning.

[0090] This invention employs an improved SIFT feature extraction method combined with a dual matching strategy of Euclidean distance ratio and cosine similarity, and introduces a stable reference point and homography matrix for coordinate system one. By establishing a precise physical transformation from two-dimensional pixel displacement to vertical settlement of the tower base, it solves the fundamental defect of single-camera visual monitoring in distinguishing between real settlement and horizontal rigid body motion. Combined with time series analysis and multi-level safety thresholds, it realizes automated, real-time, accurate monitoring and intelligent early warning of local and overall uneven settlement of bridge tower bases, providing a full-process, highly reliable monitoring method for bridge structural health.

[0091] The reference image in step S1 includes natural feature points and stable reference points;

[0092] Natural features include the surface features of the tower base concrete, structural joint features, and features of ancillary facilities;

[0093] The concrete surface features of the tower base include areas of high-contrast texture, durable stain patches, and intersections of native cracks;

[0094] Structural joint features include construction joint edges, expansion joint anchor points, and embedded part contour corners;

[0095] Features of the ancillary facilities include permanent maintenance ladder connection bolts, safety railing mounting bases, and existing monitoring point markings;

[0096] Based on preset camera parameters, this integrated device acquires reference images and time-series images through a camera system and performs image coordinate to spatial coordinate transformation.

[0097] Camera parameters include intrinsic camera parameters and extrinsic camera parameters;

[0098] Camera intrinsic parameters include focal length, principal point coordinates, and distortion coefficients; camera extrinsic parameters include camera mounting position and attitude parameters.

[0099] Attitude parameters include the azimuth, pitch, and roll angles of the camera beam relative to the world coordinate system;

[0100] Stable reference points include fixed points on natural exposed rock bases or hard concrete benchmark piers, artificial markers formed by pre-embedded steel nails, bolts, and surveying stakes, as well as permanently installed reflectors and prism positioning markers.

[0101] In practice, images are acquired using an industrial camera (45 megapixels resolution, 3.45μm pixel size) equipped with a 35mm fixed-focus lens (focal length tolerance ±0.008mm). The camera's intrinsic parameters are calibrated to the focal length using the Zhang Zhengyou calibration method. =3200px =3200px, principal point coordinates ( =2250px, =1600px), radial distortion coefficients k1=-0.12, k2=0.08, tangential distortion coefficients p1=0.0003, p2=0.0002; external parameters were calibrated using a total station (angle measurement accuracy 0.5), and the camera was positioned at east coordinates 286734.852m, north coordinates 3345671.293m, and elevation 45.362m (coordinate system CGCS2000). The attitude parameters were set as follows: azimuth 287.36°±0.01°, pitch -2.17°±0.005°, and roll 0.34°±0.003°. The selection criteria for natural feature points include: using the SIFT algorithm to extract high-contrast texture areas (grayscale variance ≥ 8000), durable stain patches (LAB color space b component difference > 15), and original crack intersections (angle tolerance ± 5°) on the concrete surface of the tower base; structural joint features are located using the Canny operator (high-low threshold ratio 1:3) to locate the edges of construction joints, expansion joint anchor points (roundness > 0.85), and corner points of embedded parts (curvature > 0.1mm); and features of ancillary facilities focus on permanent maintenance ladder connection bolts (thread contour recognition accuracy 99.2%), safety railing installation bases (flatness detection error ± 0.8mm), and existing monitoring point markers (decoding success rate ≥ 99.5%). Stable reference points are established on natural exposed rock foundations (point load strength ≥ 85 MPa) or C40 concrete reference piers (strength standard deviation ≤ 3 MPa). Manual markers (8mm diameter stainless steel nails, centering error of measuring piles ± 0.3mm) are fixed using a forced centering device. A 360° reflecting prism (distance measurement accuracy ± 1mm ​​+ 1ppm) and coded targets (ID recognition rate 100%) are also deployed. In a suspension bridge application, this scheme achieved an image coordinate to spatial coordinate transformation error ≤ 0.08 pixels and a repeatability accuracy of 0.2mm, effectively ensuring the reliability of the long-term monitoring reference.

[0102] In step S2, the time-series image is preprocessed to obtain a high-quality current image. The processing logic includes:

[0103] Distortion correction processing is performed on the acquired time-series images. The focal length, principal point coordinates, and distortion coefficients obtained from camera calibration are used to recover the distortion coordinates of each pixel in the time-series image to the true position through inverse distortion mapping, thus obtaining the distortion-corrected image.

[0104] The distortion-corrected image is converted into a grayscale image and the grayscale distribution probability of the entire image is statistically analyzed. The grayscale levels are then redistributed using the histogram equalization method to obtain a brightness-equalized image.

[0105] Noise filtering is performed on the brightness-equalized image, and the image is smoothed by convolution using a Gaussian filtering algorithm to obtain a high-quality current image;

[0106] The Gaussian filtering formula is:

[0107] ;

[0108] in, Indicates the Gaussian filter weights. This indicates the center offset of the filter window. Indicates the standard deviation of the filter. Represents the natural constant.

[0109] In specific implementation, based on the intrinsic parameters obtained from camera calibration (focal length f=3200px, principal point coordinates (2250, 1600), distortion coefficients k1=-0.12, k2=0.08, p1=0.0003, p2=0.0002), the pixel coordinates (u, v) of the temporal image are iteratively calculated to the true position (u', v') using the Brown-Conrady model through an inverse distortion mapping algorithm. The reprojection error of the corrected image is ≤0.15 pixels. Subsequently, the corrected image is converted into a grayscale image (weighting coefficients: 0.299R+0.587G+0.114B), and a limited... Contrast Adaptive Histogram Equalization (CLAHE) redistributes gray levels in 8×8 pixel grids (cropping limit 2.0), increasing image information entropy from 6.3 bit / pixel to 7.5 bit / pixel. Finally, Gaussian filtering (standard deviation σ = 1.2, convolution kernel 5×5) is used for noise suppression, with the weight calculation formula G(x,y) = (1 / 2πσ²)exp(-(x²+y²) / 2σ²). After processing, the peak signal-to-noise ratio (PSNR) reaches 41.6 dB, and the root mean square error (RMSE) is reduced to 3.8. This preprocessing workflow is accelerated in parallel on the NVIDIA Jetson Xavier edge computing platform, with a single frame processing time of 0.8 seconds, effectively ensuring the stability of subsequent feature extraction. In actual testing, the feature point matching success rate after preprocessing increased from 73% to 96.5% of the original image, while the false matching rate decreased to 1.2%.

[0110] In step S3, the improved SIFT feature extraction algorithm is used to process the baseline image and the high-quality current image to obtain feature description vectors. The calculation logic includes:

[0111] Centered on natural feature points, an arc window is used as the feature description region. The radius of the arc window is 24 pixels, and it is divided into 4 arc sub-regions from the inside out, with each arc differing by 6 pixels.

[0112] Calculate the magnitude and direction angle of the gradient of each pixel in each arc sub-region. The angle of each arc sub-region is divided into 30°. Calculate the gradient histogram of each arc sub-region in 12 directions and generate a 48-dimensional feature description vector.

[0113] The formula for calculating the magnitude of the pixel gradient is:

[0114] ;

[0115] The formula for calculating the direction angle of the pixel gradient is:

[0116] ;

[0117] in, Represents the value of the x-axis. This represents the value of the ordinate. Indicates coordinates grayscale value at that location Indicates in Gradient magnitude at point, Indicates coordinates The gradient direction angle;

[0118] The pixels within the arc window are weighted using a Gaussian weighting function to generate a reduced-dimensional feature description vector.

[0119] Gaussian weighted function calculation formula:

[0120] ;

[0121] in, Indicates coordinates The weighting coefficients at the point, Indicates the location of natural feature points. The standard deviation of the Gaussian function is represented by... Indicates the reference x-axis, Indicates the reference ordinate. Represents the natural constant;

[0122] The feature description vector matching algorithm is used to match natural feature points in the reference image to obtain matching point pairs. The matching point pairs are then filtered according to the Euclidean distance ratio and cosine similarity to obtain accurate matching point pairs.

[0123] Cosine similarity calculation formula:

[0124] ;

[0125] in, Represents cosine similarity. The first feature descriptor vector representing the feature descriptor vector of natural feature points in the reference image. Each component value The first feature descriptor vector representing the feature description vector of natural feature points in a time series image. Each component value;

[0126] Based on a stable reference point, the homography matrix is ​​estimated using the RANSAC algorithm, and coordinate system unification is performed to obtain accurate feature point coordinates.

[0127] In practice, a circular description window with a radius of 24 pixels is constructed centered on the natural feature point. This window is then divided into four concentric circular sub-regions (with a radius gradient of 4 pixels) from the inside out. Each sub-region is further divided into 12 directional intervals at 30° intervals. The Sobel operator is used to calculate the gradient magnitude of each pixel. With direction angle θ ,in for The grayscale value is calculated; a 12-directional gradient histogram (30° width) is generated for each sub-region, producing a 48-dimensional feature description vector; a Gaussian weighting function is used for weighting to enhance the central feature response. In the feature matching stage, the Euclidean distance ratio (threshold 0.6) and cosine similarity of the feature vectors are calculated using KD-tree nearest neighbor search.

[0128] (Threshold 0.85) After filtering, the matching accuracy improved to 98.2%. Finally, based on stable reference point coordinates, the RANSAC algorithm (5000 iterations, reprojection error threshold 1.2 pixels) was used to estimate the homography matrix H, unifying the image coordinates to the engineering coordinate system, achieving a feature point localization accuracy of 0.15 pixels (approximately 0.08 mm). The SIFT feature extraction algorithm is generally a 128-dimensional feature description vector, while the improved SIFT feature extraction algorithm reduces the dimensionality to a 48-dimensional feature description vector, effectively improving the computation speed.

[0129] The Euclidean distance ratio and cosine similarity are used to filter matching point pairs to obtain exact matching point pairs. The processing logic includes:

[0130] The matching distance between the feature description vector and the corresponding feature description vector of the reference image is calculated using Euclidean distance, and the first minimum distance is selected for each pair of matching points. Second minimum distance ;

[0131] when When the value is less than the first threshold, the matching point pairs are retained to form the first set of matching point pairs, which is the set of all retained matching points.

[0132] when If the value is greater than or equal to the first threshold, the matching point pair is discarded.

[0133] After filtering by Euclidean distance ratio, the cosine similarity of the feature description vectors is calculated for the matching point pairs in the first set of matching point pairs.

[0134] When the cosine similarity is less than or equal to the second threshold, the matching point pair is deemed invalid, and the matching point pair is removed and deleted.

[0135] When the cosine similarity is greater than the second threshold, the matching point pair is determined to be valid, and an exact matching point pair is obtained.

[0136] In practice, the Euclidean distance between the feature description vector to be matched and the reference feature vector is calculated, and the first minimum distance is selected. With the second minimum distance When the distance ratio ρ= If the similarity is less than the first threshold of 0.6, the matching point pair is retained in the first set of matching point pairs; otherwise, it is discarded. Then, the cosine similarity of the feature vectors of each matching point pair in the set is calculated. When the cosine similarity is less than or equal to the second threshold of 0.85, it is determined to be an invalid match and discarded. The final retained matching point pairs are the exact matching point pairs. This joint screening strategy, in the monitoring of a bridge tower foundation, reduced the false match rate of feature matching from 7.3% using the Euclidean distance ratio method alone to 1.1%, and improved the matching accuracy to 98.9%. The algorithm is implemented based on OpenCV's FlannBasedMatcher, and the average processing time on the Jetson AGX Orin edge computing platform is 0.2 seconds per thousand feature points.

[0137] Geometric constraints are applied to precisely matched point pairs based on stable reference points. The homography matrix is ​​estimated using the RANSAC algorithm. The coordinates of precisely matched point pairs in the high-quality current image are then unified to the coordinate system of the reference image to obtain the precise feature point coordinates.

[0138] In practice, at least four stable reference points (such as pre-embedded steel nails or reflecting prisms) are selected to form a control network, whose spatial coordinates are measured using a total station (accuracy ±0.6mm). The coordinates of the precisely matched point pairs and the stable reference points are input into the RANSAC algorithm, with 5000 iterations and a reprojection error threshold of 1.2 pixels. The homography matrix H is calculated iteratively using the least squares method. The matrix solution employs the Direct Linear Transform (DLT) algorithm, ultimately obtaining an optimized 3×3 homography matrix with a condition number <103, ensuring numerical stability.

[0139] Through coordinate transformation formula ;

[0140] This method unifies the coordinates of feature points in the current high-quality image to the coordinate system of the reference image, achieving pixel-level coordinate alignment. In actual testing, at a shooting distance of 80 meters, the coordinate transformation residual is less than 0.1 pixels (approximately 0.05 mm), representing a 40% improvement in accuracy compared to the traditional least-squares fitting method. In the monitoring of a cable-stayed bridge, the system maintained a transformation stability of 0.15 pixels even under a level 12 wind vibration environment, and 98.7% of the matched point pairs passed geometric constraint verification, effectively eliminating systematic errors caused by camera micro-motion.

[0141] The processing logic for coordinate unification includes:

[0142] Using stable reference points already contained in the baseline image, geometric constraints are applied to precisely matched point pairs. The homography matrix is ​​estimated using the RANSAC algorithm. The processing logic includes:

[0143] Iterative calculations are performed using the RANSAC algorithm. At least four pairs of matching points containing stable reference points are arbitrarily selected to calculate the homography matrix. The projection error is calculated for all matching point pairs, and pairs with projection errors greater than a preset error threshold are discarded. Matching point pairs with projection errors not exceeding the preset error threshold are considered interior points. When the proportion of interior points to the total number of matching point pairs exceeds a preset proportion threshold, the estimated homography matrix is ​​considered valid. The formula for the homography matrix is:

[0144] ;

[0145] in, Represents the homography matrix. Indicates the high-quality current image of the th The original x-coordinates of the natural feature points Indicates the high-quality current image of the th The original ordinates of the natural feature points This indicates that the homography matrix is ​​mapped to the reference image coordinate system. The x-coordinates of natural feature points This indicates that the homography matrix is ​​mapped to the reference image coordinate system. The ordinates of the natural feature points;

[0146] Coordinates of natural feature points in the high-quality current image Mapping to the coordinate system of the reference image yields precise feature point coordinates. .

[0147] In practice, four pairs of matching points containing stable reference points are randomly selected from the precise matching point pairs (at least two of which are stable reference points). The initial homography matrix H is calculated using direct linear transformation. Then, the projection error of all matching point pairs is calculated, with an error threshold of 1.2 pixels. Points with projection errors exceeding this threshold are discarded, retaining only the inlier set. When the proportion of inlier sets exceeds a preset threshold of 85%, the homography matrix is ​​considered valid. At this point, the final matrix H is obtained through singular value decomposition optimization, with its condition number controlled within 10³ to ensure numerical stability. Coordinate transformation is then used. ;

[0148] This method unifies the coordinates of feature points in a high-quality current image to the coordinate system of a reference image, achieving coordinate system unification. Real-world testing data shows that this scheme controls the coordinate transformation residual to within 0.08 pixels (approximately 0.04 mm) in the monitoring of a suspension bridge, improving accuracy by 52% compared to the least squares method. It also maintains transformation stability (drift < 0.02 pixels) within a temperature range of -20℃ to 50℃. The entire algorithm is parallelized and accelerated on the Jetson AGX Xavier platform, with a single iteration taking 3.2 ms, meeting the real-time requirements of engineering projects.

[0149] In step S4, the precise feature point coordinates are compared with the natural feature point coordinates, and the two-dimensional pixel displacement is calculated. The processing logic includes:

[0150] The coordinate system obtains the precise feature point coordinates of natural feature points in the current high-quality image. Coordinates of natural feature points in the reference image Point-by-point comparison is performed to calculate the 2D pixel displacement. The specific formula is as follows:

[0151] ;

[0152] ;

[0153] in, Indicates the high-quality current image of the th The pixel coordinates of each natural feature point Indicating the first in the reference image The pixel coordinates of each natural feature point Indicates the first The two-dimensional pixel displacement of a natural feature point in the horizontal direction of the image. Indicates the first The two-dimensional pixel displacement of a natural feature point in the vertical direction of the image;

[0154] Based on the camera intrinsic and extrinsic parameters, the two-dimensional pixel displacement is projected and converted into the tower foundation settlement in the direction perpendicular to the ground. The formula for the tower foundation settlement is:

[0155] ;

[0156] in, Indicates the camera's focal length in the vertical direction. Indicates the optical center of the camera to the first The straight-line distance between each natural feature point on the monitoring surface This indicates the angle between the camera's optical axis and the normal to the monitoring surface. This indicates the amount of tower foundation settlement in the direction perpendicular to the ground.

[0157] In practice, the precise feature point coordinates after coordinate system one are compared with the reference image coordinates point by point to calculate the two-dimensional pixel displacement. ,(in As the reference coordinates, (Current coordinates); based on the camera intrinsic parameters (focal length) obtained from calibration. =3200px) and extrinsic parameters (distance from optical center to feature point L=80m±0.1m, angle between optical axis and normal to monitoring surface θ=2.17°±0.005°), are converted using the projection transformation formula. The vertical pixel displacement is converted into tower foundation settlement. Actual measurement data shows that this conversion model, at a monitoring distance of 100m, has a settlement measurement error of less than ±0.12mm, with a relative accuracy of 1 / 80000. In a cable-stayed bridge application, the system successfully identified a progressive settlement of 0.15mm / month in the tower foundation, with a correlation coefficient of 0.987 with traditional hydrostatic level data. The algorithm utilizes a temperature compensation module (compensation coefficient 1.2×10⁻⁶). -5 ( / ℃) Eliminates the influence of thermal deformation and maintains measurement stability within an ambient temperature range of -10℃ to 40℃.

[0158] In step S5, a time-series analysis of the tower foundation settlement is performed to obtain differential settlement values. The processing logic includes:

[0159] The settlement of the tower base at natural feature points during the monitoring period is stored in chronological order to form a settlement time series.

[0160] Based on the settlement time series, the difference between each natural feature point on the same tower base is calculated. The settlement difference between any two natural feature points at the same monitoring time is calculated. The absolute value of the settlement difference is taken to obtain the local differential settlement value of the corresponding tower base.

[0161] For each tower foundation of the bridge, the absolute value of the difference in settlement between any two natural feature points with the largest settlement at the same monitoring time is calculated to obtain the overall differential settlement value between the tower foundations.

[0162] In practice, the settlement of each natural feature point is stored in a time-series database at 30-minute intervals to form a standardized data sequence containing three elements: timestamp, point number, and settlement amount. For the six natural feature points on the same tower base (such as the No. 3 main tower), the settlement difference between any two points is calculated and the absolute value is taken to obtain 15 sets of local differential settlement values ​​(such as the difference between feature points A01 and A02 |0.12-0.08|=0.04mm). When any local differential settlement value exceeds the threshold of 2.0mm for three consecutive cycles, a local warning is triggered. At the same time, the maximum settlement feature point of each of the four tower bases of the whole bridge is selected (such as the settlement of point P12 of tower 1 is 0.15mm and the settlement of point P28 of tower 2 is 0.23mm), and the overall differential settlement value between the tower bases is calculated as |0.15-0.23|=0.08mm. When this value exceeds the safety threshold L / 2000 (corresponding to 4.0mm), the overall structural safety assessment is initiated. During a 12-month monitoring period of a cable-stayed bridge, the system successfully provided early warning of uneven settlement in the tower foundation by detecting local differential settlement trends (0.05 mm / month → 0.12 mm / month), identifying the anomaly 47 days ahead of the design specification limits. The algorithm, based on a Spark distributed architecture, enables parallel computing; a complete analysis of 300,000 settlement data points took only 8 seconds, achieving a differential settlement identification accuracy of 99.2%.

[0163] The processing logic for generating an early warning for uneven settlement based on a preset safety threshold includes:

[0164] A first-level safety threshold and a second-level safety threshold are set for the local differential settlement value of the tower base, with the second-level safety threshold being greater than the first-level safety threshold;

[0165] A third-level safety threshold and a fourth-level safety threshold are set for the overall differential settlement value between the tower bases, with the fourth-level safety threshold being greater than the third-level safety threshold;

[0166] When any local differential settlement value of the tower base exceeds the first-level safety threshold but does not exceed the second-level safety threshold, a local first-level early warning is generated;

[0167] When any local differential settlement value of the tower base exceeds the second-level safety threshold, a local second-level early warning is generated;

[0168] When any overall differential settlement value between tower bases exceeds the third-level safety threshold but does not exceed the fourth-level safety threshold, an overall first-level early warning is generated;

[0169] When the differential settlement value of any individual tower base exceeds the fourth-level safety threshold, an overall second-level early warning is generated.

[0170] In practice, the first-level safety threshold for local differential settlement of the tower base is set at 2.0 mm (corresponding to L / 5000), and the second-level threshold is 3.5 mm (L / 2857). The third-level threshold for overall differential settlement between tower bases is 4.0 mm (L / 2500), and the fourth-level threshold is 6.0 mm (L / 1667). When the monitoring system detects that the local differential settlement value between characteristic points A12 and A15 of tower base No. 3 reaches 2.3 mm (exceeding the first-level threshold but not exceeding the second-level threshold), a local first-level warning is automatically generated, triggering the monitoring frequency to increase to once every 15 minutes, and pushing the verification task to the handheld terminal of the maintenance personnel; when the value continues to develop to 3.8 mm (exceeding the second-level threshold), it is upgraded to a local second-level warning, and the drone aerial photography verification and expert consultation mechanism are launched simultaneously. In response to differential settlement between tower foundations, when the maximum settlement difference between towers 2 and 4 reaches 4.5mm (exceeding level 3 but not level 4), a level 1 overall warning is generated, traffic restrictions are implemented in adjacent lanes, and a structural safety assessment is initiated. When the difference exceeds 6.2mm (exceeding the level 4 threshold), a level 2 overall warning is immediately triggered, the entire bridge evacuation plan is activated, and the matter is reported to the provincial regulatory department. In a cross-sea bridge application, this system successfully issued a warning for a tower foundation uneven settlement development event (with a local differential settlement monthly increase of 0.4mm) through a multi-level warning and coordination mechanism. From the level 1 warning to the final handling, the time taken was only 2.3 hours, a 12-fold increase in efficiency compared to traditional manual inspections. All warning events are stored using blockchain technology to ensure data immutability, achieving a warning accuracy rate of 99.1% and a false alarm rate below 0.3%.

[0171] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0172] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for monitoring uneven settlement of a bridge tower foundation based on visual detection, characterized in that, Includes the following steps: Step S1: Select natural feature points and deploy the camera system to acquire a reference image, which includes natural feature points and stable reference points; Step S2: Acquire time-series images through the camera system, and preprocess the time-series images to obtain high-quality current images; Step S3 involves processing the baseline image and the high-quality current image using an improved SIFT feature extraction algorithm to obtain feature description vectors. The calculation logic includes: Centered on natural feature points, an arc window is used as the feature description region. The radius of the arc window is 24 pixels, and it is divided into 4 arc sub-regions from the inside out, with each arc differing by 6 pixels. Calculate the magnitude and orientation angle of the gradient of each pixel in each arc sub-region. Each arc sub-region is divided into 30° angles. Calculate the gradient histogram of each arc sub-region in 12 directions and generate a 48-dimensional feature description vector. The formula for calculating the magnitude of the pixel gradient is: ； The formula for calculating the direction angle of the pixel gradient is: ； in, Represents the value of the x-axis. This represents the value of the ordinate. Indicates coordinates grayscale value at that location Indicates in Gradient magnitude at point, Indicates coordinates The gradient direction angle; The pixels within the arc window are weighted using a Gaussian weighting function to generate a reduced-dimensional feature description vector. The formula for calculating the Gaussian weighting function is as follows: ； wherein, denotes a weighting coefficient at coordinates denotes a weighting coefficient at coordinates denotes a standard deviation of a Gaussian function, denotes a reference abscissa, denotes a reference ordinate, denotes a natural constant; The feature description vector matching algorithm is used to match natural feature points in the reference image to obtain matching point pairs. The matching point pairs are then filtered according to the Euclidean distance ratio and cosine similarity to obtain accurate matching point pairs. The formula for calculating cosine similarity is: ； in, Represents cosine similarity. The first feature descriptor vector representing the feature descriptor vector of natural feature points in the reference image. Each component value The first feature descriptor vector representing the feature description vector of natural feature points in a time series image. Each component value; Based on a stable reference point, the RANSAC algorithm is used to estimate the homography matrix, and coordinate system unification is performed to obtain accurate feature point coordinates. The precise coordinates of the feature points are obtained by using a coordinate system based on the homography matrix and stable reference points. Step S4 involves comparing the precise feature point coordinates with the natural feature point coordinates and calculating the two-dimensional pixel displacement. This two-dimensional pixel displacement is then converted to obtain the tower foundation settlement in the vertical direction. The processing logic is as follows: The coordinate system obtains the precise feature point coordinates of natural feature points in the current high-quality image. Coordinates of natural feature points in the reference image Point-by-point comparison is performed to calculate the 2D pixel displacement. The specific formula is as follows: ； ； in, Indicates the high-quality current image of the th The pixel coordinates of each natural feature point Indicating the first in the reference image The pixel coordinates of each natural feature point Indicates the first The two-dimensional pixel displacement of a natural feature point in the horizontal direction of the image. Indicates the first The two-dimensional pixel displacement of a natural feature point in the vertical direction of the image; Based on the camera intrinsic and extrinsic parameters, the two-dimensional pixel displacement is projected and converted into the tower foundation settlement in the direction perpendicular to the ground. The formula for the tower foundation settlement is: ； in, Indicates the camera's focal length in the vertical direction. Indicates the optical center of the camera to the first The straight-line distance between each natural feature point on the monitoring surface This indicates the angle between the camera's optical axis and the normal to the monitoring surface. This indicates the amount of tower foundation settlement in the direction perpendicular to the ground. Step S5: Perform time-series analysis on the settlement amount to obtain differential settlement values, determine the threshold based on a preset safety threshold, and generate an uneven settlement early warning.

2. The method for monitoring uneven settlement of a bridge tower foundation based on visual detection according to claim 1, characterized in that: The reference image in step S1 includes natural feature points and stable reference points; The natural feature points include the surface features of the tower base concrete, structural joint features, and features of ancillary facilities; The surface features of the tower base concrete include high-contrast textured areas, durable stain patches, and intersections of native cracks; The structural joint features include the edge of the construction joint, the anchor point of the expansion joint, and the corner point of the embedded part outline; The features of the ancillary facilities include permanent maintenance ladder connecting bolts, safety railing mounting bases, and original monitoring point markings; Based on preset camera parameters, this integrated device acquires reference images and time-series images through a camera system and performs image coordinate to spatial coordinate transformation. The camera parameters include intrinsic camera parameters and extrinsic camera parameters; The camera intrinsic parameters include focal length, principal point coordinates, and distortion coefficients, while the camera extrinsic parameters include camera mounting position and attitude parameters. The attitude parameters include the azimuth, pitch, and roll angles of the camera beam relative to the world coordinate system. The stable reference points include fixed points on natural exposed rock bases or hard concrete benchmark piers, artificial marker points formed by pre-embedded steel nails, bolts, and measuring piles, and permanently installed reflective sheets and prism positioning marker points.

3. The method for monitoring the uneven settlement of a bridge tower foundation based on visual detection according to claim 2, characterized in that: In step S2, the time-series image is preprocessed to obtain a high-quality current image. The processing logic includes: Distortion correction processing is performed on the acquired time-series images. The focal length, principal point coordinates, and distortion coefficients obtained from camera calibration are used to recover the distortion coordinates of each pixel in the time-series image to the true position through inverse distortion mapping, thus obtaining the distortion-corrected image. The distortion-corrected image is converted into a grayscale image and the grayscale distribution probability of the entire image is statistically analyzed. The grayscale levels are then redistributed using the histogram equalization method to obtain a brightness-equalized image. Noise filtering is performed on the brightness-equalized image, and the image is smoothed by convolution using a Gaussian filtering algorithm to obtain a high-quality current image; The Gaussian filtering formula is as follows: ； wherein, represents a Gaussian filter weight, represents a filter window center offset, represents a filter standard deviation, represents a natural constant.

4. The method for monitoring uneven settlement of bridge tower foundations based on visual inspection as described in claim 3, characterized in that: The Euclidean distance ratio and cosine similarity are used to filter the matching point pairs to obtain exact matching point pairs. The processing logic includes: The matching distance between the feature description vector and the corresponding feature description vector of the reference image is calculated according to the Euclidean distance, and the first minimum distance and the second minimum distance are respectively selected for the matching point pair and the second minimum distance ​ when When the value is less than the first threshold, the matching point pairs are retained to form the first set of matching point pairs, which is the set of all retained matching points. When when greater than or equal to a first threshold, discarding the matching point pair; After filtering by Euclidean distance ratio, the cosine similarity of the feature description vectors is calculated for the matching point pairs in the first set of matching point pairs. When the cosine similarity is less than or equal to the second threshold, the matching point pair is deemed invalid, and the matching point pair is removed and deleted. When the cosine similarity is greater than the second threshold, the matching point pair is determined to be valid, and an exact matching point pair is obtained.

5. The method for monitoring the uneven settlement of a bridge tower foundation based on visual detection according to claim 4, characterized in that: Geometric constraints are applied to precisely matched point pairs based on stable reference points. The homography matrix is ​​estimated using the RANSAC algorithm. The coordinates of precisely matched point pairs in the high-quality current image are then unified to the coordinate system of the reference image to obtain the precise feature point coordinates.

6. The method for monitoring the differential settlement of a bridge tower foundation based on visual detection according to claim 5, characterized in that: The coordinate unification processing logic includes: Using stable reference points already contained in the baseline image, geometric constraints are applied to the precisely matched point pairs. The homography matrix is ​​estimated using the RANSAC algorithm. The processing logic includes: Iterative calculations are performed using the RANSAC algorithm. At least four pairs of matching points containing stable reference points are arbitrarily selected to calculate the homography matrix. The projection error is calculated for all matching point pairs, and pairs with projection errors greater than a preset error threshold are discarded. Matching point pairs with projection errors not exceeding the preset error threshold are considered interior points. When the proportion of interior points to the total number of matching point pairs exceeds a preset proportion threshold, the estimated homography matrix is ​​deemed valid. The formula for the homography matrix is: ； in, Represents the homography matrix. Indicates the high-quality current image of the th The original x-coordinates of the natural feature points Indicates the high-quality current image of the th The original ordinates of the natural feature points This indicates that the homography matrix is ​​mapped to the reference image coordinate system. The x-coordinates of natural feature points This indicates that the homography matrix is ​​mapped to the reference image coordinate system. The ordinates of the natural feature points; Mapping natural feature point coordinates in a high-quality current image to a coordinate system of a reference image to obtain accurate feature point coordinates Mapping natural feature point coordinates in a high-quality current image to a coordinate system of a reference image to obtain accurate feature point coordinates .

7. The method for monitoring the differential settlement of a bridge tower foundation based on visual detection according to claim 6, characterized in that: In step S5, a time-series analysis is performed on the tower foundation settlement to obtain differential settlement values. The processing logic includes: The settlement of the tower base at natural feature points during the monitoring period is stored in chronological order to form a settlement time series. Based on the settlement time series, the difference between each natural feature point on the same tower base is calculated. The settlement difference between any two natural feature points at the same monitoring time is calculated. The absolute value of the settlement difference is taken to obtain the local differential settlement value of the corresponding tower base. For each tower foundation of the bridge, the absolute value of the difference in settlement between any two natural feature points with the largest settlement at the same monitoring time is calculated to obtain the overall differential settlement value between the tower foundations.

8. The method for monitoring the differential settlement of a bridge tower foundation based on visual detection according to claim 7, characterized in that: The processing logic for generating an early warning for uneven settlement based on a preset safety threshold includes: A first-level safety threshold and a second-level safety threshold are set for the local differential settlement value of the tower base, wherein the second-level safety threshold is greater than the first-level safety threshold; A third-level safety threshold and a fourth-level safety threshold are set for the overall differential settlement value between the tower bases, wherein the fourth-level safety threshold is greater than the third-level safety threshold; When any local differential settlement value of the tower base exceeds the first-level safety threshold but does not exceed the second-level safety threshold, a local first-level early warning is generated; When any local differential settlement value of the tower base exceeds the second-level safety threshold, a local second-level early warning is generated; When any overall differential settlement value between tower bases exceeds the third-level safety threshold but does not exceed the fourth-level safety threshold, an overall first-level early warning is generated; When the differential settlement value of any individual tower base exceeds the fourth-level safety threshold, an overall second-level early warning is generated.

Images