Cable force rapid measurement method and system based on digital image correlation method

Abstract

The application discloses a kind of based on digital image correlation method's cable-stayed cable cable force fast measurement method and system, method includes: acquisition cable-stayed cable video data;With line segment measurement calibration method, with cable-stayed cable texture point as feature point, establish feature point selection mechanism;Based on redundant observation method, the coordinate mapping model of pixel-physical space is established;And through collaborative calibration method, model parameters are obtained;Based on digital image correlation method, displacement measurement is carried out;Establish double-layer feature extraction mechanism to complete local and global feature extraction;And through weight self-adapting fusion strategy, the weight distribution of local feature and global feature is dynamically adjusted;Finally, double-layer tracking and failure recovery mechanism is established to complete layered cooperation of coarse-grained global tracking and fine-grained local tracking;The displacement measurement result is carried out fast fourier transform, and cable force is calculated.The reliable technical support is provided for non-contact precision calculation measurement of cable force.

Description

Technical Field

[0001] This invention belongs to the technical field of digital image correlation methods, and more specifically, relates to a method and system for rapid measurement of cable force in cable-stayed bridges based on digital image correlation methods. Background Technology

[0002] As a core structural form of modern long-span bridges, cable-stayed bridges rely heavily on cable stress monitoring as a crucial component of bridge health monitoring systems. Changes in cable stress directly reflect the stress state of the bridge structure, and accurate dynamic displacement measurement is a fundamental prerequisite for cable stress calculation. However, current cable displacement measurement technology faces multiple technical bottlenecks in engineering applications, urgently requiring systematic breakthroughs.

[0003] Traditional contact measurement methods, such as strain gauges and servo displacement gauges, require physical bonding or mechanical connection to the cable surface. This not only interferes with the original vibration characteristics of the cable, leading to distorted measurement data, but also presents challenges such as difficult installation and maintenance, and short sensor lifespan in complex environments such as high altitude and high humidity during bridge operation, making it difficult to achieve long-term continuous monitoring.

[0004] In non-contact measurement technology, the digital image correlation (DIC) method is widely used due to its advantages of non-contact and full-field measurement. However, its matching mechanism, which relies on local texture features, has inherent defects: in long-distance monitoring scenarios, the surface texture of the cable-stayed bridge is reduced in resolution in the image, and areas with missing texture are prone to appear; large vibrations of the cable-stayed bridge caused by strong winds, vehicle loads, etc., will cause local feature points to exceed the search range of DIC; environmental factors such as sudden changes in illumination and atmospheric turbulence will interfere with the gray-scale distribution of the image, causing a decrease in matching quality, and ultimately leading to tracking failure and displacement data interruption.

[0005] Another core pain point of existing technologies lies in insufficient calibration accuracy. Traditional line segment calibration methods lack scientific screening and spatial distribution optimization for feature point selection, making it difficult to meet engineering requirements in terms of calibration accuracy. Single feature tracking modes are even more difficult to balance measurement accuracy and environmental adaptability—while relying solely on local DIC features can achieve high positioning accuracy, reliability drops sharply in scenarios with missing textures or large displacements; while relying solely on global features provides contour semantic robustness, spatial positioning accuracy is insufficient, and the fusion mechanism of global and local features lacks dynamic adjustment logic, making it difficult to intelligently switch according to real-time scenarios.

[0006] When displacement measurement data is erroneous or interrupted due to tracking failure, frequency shift will occur in the spectrum analysis, and cable force calculation will result in deviation due to distortion of input data, exceeding the allowable range of the specifications, which will seriously affect the accuracy of bridge safety assessment.

[0007] In summary, existing technologies face core problems in cable-stayed bridge displacement measurement, such as "insufficient accuracy of long-distance calibration" and "failure of tracking in complex environments." There is an urgent need to achieve high-precision monitoring through technological innovation to provide reliable technical support for accurate cable force calculation and bridge structural safety assessment. Summary of the Invention

[0008] This invention aims to provide an improved method for measuring the displacement of stay cables, addressing issues such as insufficient calibration accuracy during long-distance monitoring and easy tracking failure in complex environments in existing technologies. By establishing a precise pixel-physical space mapping through a line segment measurement calibration method, and combining a multi-scale feature fusion tracking method with a dual-layer tracking mechanism, it achieves synergy between high-precision measurement of local features and robust protection of global features, improving the accuracy and environmental adaptability of displacement measurement, and providing reliable data support for accurate calculation of stay cable forces.

[0009] To address the aforementioned deficiencies or improvement needs of existing technologies, as a first aspect of this invention, the present invention provides a method for rapid measurement of cable force in stay cables based on digital image correlation, comprising:

[0010] S1. Collect video data of the cable-stayed bridge;

[0011] S2. A feature point optimization selection mechanism is established using the line segment measurement calibration method with the cable-stayed bridge texture points as feature points; a pixel-physical space coordinate mapping transformation model is established based on the redundant observation method; and the model parameters are calculated through the collaborative calibration method.

[0012] S3. Displacement measurement is performed based on the Digital Image Correlation (DIC) method; a two-layer feature extraction mechanism is established to complete local and global feature extraction; the weight allocation of local and global features is dynamically adjusted through a weight adaptive fusion strategy; finally, a two-layer tracking and failure recovery mechanism is established to complete the hierarchical collaboration between "coarse-grained global tracking" and "fine-grained local tracking";

[0013] S4. Perform a fast Fourier transform on the dynamic displacement measurement results to calculate the cable force of the cable.

[0014] Furthermore, the feature point optimization selection mechanism in S2 is specifically as follows:

[0015] The dense intersection points of the spiral lines on the surface of the cable-stayed bridge are selected as feature points. Their high texture contrast is used to ensure sub-pixel level positioning accuracy and avoid tracking errors caused by feature loss in smooth areas.

[0016] The Sobel operator is used to calculate the gradient response of the image, and a candidate set of feature points with gradient values ​​>128 is established to filter out low-contrast and noise-sensitive points, thereby improving the reliability of feature points.

[0017] A grid layout strategy is adopted to divide the image into multiple sub-regions. Feature points are selected evenly in each sub-region to avoid calibration parameter deviations caused by local clustering and ensure balanced global calibration accuracy.

[0018] Furthermore, the specific transformation model for the pixel-physical space coordinate mapping in S2 is as follows:

[0019] Establish a set of feature points Define pixel coordinates To physical coordinates The conversion relationship is as follows:

[0020] ,

[0021] ,

[0022] in, These are the local calibration coefficients for each point. and The offset is determined through subsequent collaborative calibration methods to achieve a precise mapping from pixel displacement to physical displacement.

[0023] Furthermore, the collaborative calibration method in S2 iteratively optimizes the initial calibration parameters through RANSAC robust estimation and weighted fusion:

[0024] The RANSAC robust estimation randomly selects no fewer than 4 feature points, calculates the transformation matrix H, counts the number of interior points with reprojection error < 2 pixels, and saves the optimal transformation parameters.

[0025] The weighted fusion is calculated by assigning weights based on the stability of feature points, using the following formula:

[0026] ,

[0027] in, For the first Standard deviation of tracking error for each feature point The corner response value ensures that feature points with high stability have a greater weight in the calibration, thus optimizing the overall accuracy. As a global error benchmark; For feature points; For the first Adaptive weights for each feature point.

[0028] Furthermore, the dual-layer feature extraction mechanism in S3 includes local feature layer extraction driven by DIC and global feature layer extraction driven by CNN;

[0029] The local feature layer extraction dynamically sets the subset size (2M+1)×(2M+1) based on the cable-stayed bridge texture density. Dense texture regions use small subsets (M not greater than 8) to ensure positioning accuracy; smooth regions use large subsets (M not less than 16) to enhance grayscale correlation. Texture richness is then selected by calculating the texture richness index of each subset, as follows:

[0030] ,

[0031] In the formula, The standard deviation of grayscale This is the grayscale average; only retain the grayscale value. A subset of features is used as tracking points to avoid tracking failure in smooth regions due to lack of texture; Texture richness index for subset; The set texture richness threshold;

[0032] The global feature layer extracts the input cable ROI image and outputs a cable contour probability map. It also extracts global contour features of the cable through semantic segmentation to compensate for the shortcomings of local features in DIC (Discrete Injection) under large-scale vibration or occlusion. A boundary enhancement loss function is used to simultaneously optimize contour pixel classification accuracy and boundary localization accuracy, ensuring the spatial localization accuracy of global features, as detailed below:

[0033] ,

[0034] in, For the total loss, The pixel classification loss measures the overlap between the predicted contour and the true contour. The boundary localization loss measures the distance between the predicted contour boundary and the true boundary.

[0035] Furthermore, the weight adaptive fusion strategy in S3 is as follows:

[0036] ,

[0037] in, The final feature weights after fusion; Weights for local features of DIC; The weights of the global features in the CNN; The weighting coefficients are calculated as follows:

[0038] ,

[0039] To match the quality of DIC, comprehensive consideration is needed. The correlation coefficient and the standard deviation of subregion displacement are calculated using the following formulas:

[0040] ,

[0041] In the formula, Indicates the first A subset Correlation coefficient; Indicates the first The standard deviation of displacement of each subset; This represents the total number of DIC subsets participating in the evaluation;

[0042] To improve the quality of CNN contour extraction, we evaluate it using contour integrity and image signal-to-noise ratio.

[0043] Furthermore, the two-layer tracking process in S3 is specifically as follows:

[0044] A lightweight semantic segmentation network is used to extract the complete outline of the cable and output a discrete set of points along the centerline to achieve robust estimation of large-scale displacement.

[0045] ,

[0046] In the formula, The discrete set of points along the global centerline of the cable; For the first The coordinates of discrete points along the centerline;

[0047] The global displacement field is obtained by differentiating the centerline point sets between frames:

[0048] ,

[0049] In the formula, For the global displacement field of the cable; for The set of points along the center line of the cable at any given moment; for The set of points along the center line of the cable at any given moment; For time frame indexing;

[0050] When texture is missing or lighting changes abruptly, the semantic robustness of CNN contours can maintain basic tracking and avoid "cliff-like failure" of local feature matching.

[0051] In global displacement Within a defined "small search window," the IC-GN method is applied to the DIC subset to compensate for the "coarse-grained error" of global positioning; by optimizing the gray-level correlation of pixels within the subset, sub-pixel-level displacement residuals are solved. The final output is the fused displacement:

[0052] ,

[0053] In the formula, This represents the final high-precision displacement output; This indicates a coarse estimate of displacement at the global layer. This indicates the local layer fine-tuning displacement.

[0054] Furthermore, the failure recovery process in S3 is specifically as follows:

[0055] When local DIC tracking faces the risk of failure due to poor matching quality or abnormal displacement residuals, the system activates a "global feature fallback + expanded range rematch" mechanism to ensure continuous and reliable displacement measurement, as detailed below:

[0056] The presence of any of the following warning signals indicates that local tracking is unreliable:

[0057] Gray-scale matching index of DIC subset Local displacement residuals ;

[0058] If any of the above conditions are met, local tracking is determined to have failed, triggering the following recovery process:

[0059] A lightweight semantic segmentation network is invoked to re-extract the complete outline of the cable and generate a new set of centerline points. Based on the semantic robustness of the contour, the tracking anchor points are quickly reconstructed to avoid direct data loss.

[0060] The search window for the DIC subset is expanded from a small, conventional range to the possible areas predicted by the global layer, and the texture is rematched using the IC-GN method. By "expanding the search radius," areas with abrupt displacements or missing textures are covered, thereby improving the success rate of DIC rematch.

[0061] As a second aspect of the present invention, a rapid measurement system for cable tension based on a digital image correlation method is also provided, comprising:

[0062] The video acquisition unit is used to acquire video data of the cable-stayed bridge.

[0063] The calibration modeling unit is used to establish a feature point optimization selection mechanism by adopting the line segment measurement calibration method and taking the cable-stayed bridge texture points as feature points; it establishes a pixel-physical space coordinate mapping transformation model based on the redundant observation method; and it calculates the model parameters through the collaborative calibration method.

[0064] The displacement measurement unit is used for displacement measurement based on the Digital Image Correlation (DIC) method; a two-layer feature extraction mechanism is established to complete local and global feature extraction; and the weight allocation of local and global features is dynamically adjusted through a weight adaptive fusion strategy; finally, a two-layer tracking and failure recovery mechanism is established to complete the hierarchical collaboration of "coarse-grained global tracking" and "fine-grained local tracking".

[0065] The cable force calculation unit is used to perform a fast Fourier transform on the dynamic displacement measurement results to calculate the cable force of the cable.

[0066] As a third aspect of the invention, a computer-readable storage medium is also provided, on which a computer program is stored, which is executed by a processor of any step of the method for rapid measurement of cable force of a cable-stayed bridge based on a digital image correlation method.

[0067] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects:

[0068] 1. This invention provides a rapid measurement method for cable tension in stay-stayed cables based on digital image correlation. By constructing a systematic calibration system, it achieves high-precision pixel-to-physical space conversion. Employing a line segment measurement calibration method, it uses cable texture points as feature points. Through a feature point optimization selection mechanism, it selects densely intersecting points of the cable threads and combines Sobel gradient filtering and gridded layout to ensure the reliability and balanced spatial distribution of feature points. A pixel-to-physical space coordinate mapping transformation model is established based on redundant observations. The model parameters are iteratively optimized using a RANSAC robust estimation and weighted fusion collaborative calibration method. This calibration system effectively overcomes the problems of arbitrary feature point selection and low calibration accuracy in traditional methods, establishing a precise correspondence between pixels and physical space, providing an accurate foundation for subsequent displacement measurements.

[0069] 2. The present invention provides a rapid method for measuring cable force in cable stays based on digital image correlation (DIC). This method improves the accuracy and stability of displacement measurement through an innovative dual-layer feature extraction and weighted adaptive fusion strategy. A dual-layer feature extraction mechanism is established: the local feature layer is driven by DIC, dynamically adjusting the subset size and selecting texture-rich regions based on texture density; the global feature layer is driven by CNN, extracting cable contour features. Simultaneously, a weighted adaptive fusion strategy is employed, dynamically adjusting the weights of DIC matching and CNN contour extraction based on their respective quality. This strategy overcomes the limitations of single feature tracking in complex environments, allowing for flexible switching of dominant features in scenarios with missing textures or changing illumination, ensuring the reliability and accuracy of displacement measurement under different conditions.

[0070] 3. The fast cable force measurement method for stay cables based on digital image correlation (DIC) of this invention ensures the continuity of displacement measurement through a dual-layer tracking and failure recovery mechanism. In the dual-layer tracking, a lightweight semantic segmentation network is first used to achieve global localization and obtain the cable centerline displacement as a coarse estimate. Then, based on the global displacement, the local DIC subset is refined using the IC-GN method to obtain a high-precision displacement. When a failure risk occurs in local DIC tracking, such as a low ZNCC correlation coefficient or abnormal displacement residuals, the failure recovery mechanism is triggered. This involves re-initializing the global features of a CNN and expanding the search range for re-matching. This mechanism effectively avoids data interruption caused by local tracking failures, ensuring that cable displacement measurement remains continuous and stable under complex conditions such as strong winds and vibrations, providing complete and reliable data for cable force calculation. Attached Figure Description

[0071] Figure 1 This is a flowchart of a rapid measurement method for cable tension based on digital image correlation according to an embodiment of the present invention;

[0072] Figure 2 This is an elevation layout diagram of a cable-stayed bridge according to an embodiment of the present invention;

[0073] Figure 3 This is a schematic diagram of camera calibration according to an embodiment of the present invention;

[0074] Figure 4 This is a system unit diagram of an embodiment of the present invention. Detailed Implementation

[0075] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0076] Example 1

[0077] Please refer to Figure 1 This embodiment 1 provides a rapid measurement method for cable tension based on digital image correlation, including:

[0078] S1. Collect video data of the cable-stayed bridge;

[0079] S2. A feature point optimization selection mechanism is established using the line segment measurement calibration method with the cable-stayed bridge texture points as feature points; a pixel-physical space coordinate mapping transformation model is established based on the redundant observation method; and the model parameters are calculated through the collaborative calibration method.

[0080] S3. Displacement measurement is performed based on the Digital Image Correlation (DIC) method; a two-layer feature extraction mechanism is established to complete local and global feature extraction; the weight allocation of local and global features is dynamically adjusted through a weight adaptive fusion strategy; finally, a two-layer tracking and failure recovery mechanism is established to complete the hierarchical collaboration between "coarse-grained global tracking" and "fine-grained local tracking";

[0081] S4. Perform a fast Fourier transform on the dynamic displacement measurement results to calculate the cable force of the cable.

[0082] This embodiment 1 further elaborates on the above steps.

[0083] (1) Video capture

[0084] To perform non-contact cable force measurement on cable-stayed bridges, it is necessary to first collect video data of the specific cables; please refer to... Figure 2 In a preferred embodiment, non-contact cable force measurement was performed on the #B-04 stay cable of the Yujiang Grand Bridge. The image acquisition device was an MV-CS050-10UM, the industrial lens was an HTFA12035-12MP, and a x2 teleconverter was used to increase the lens focal length. 160 seconds of video data was captured on the B-04 stay cable, with an image resolution of 2448×2048 and a video frame rate of 20Hz, totaling 3205 frames of image data.

[0085] (2) Calibration Modeling

[0086] Because the acquisition scene distance exceeds 150m, it is impossible to determine the distance between the object and the camera using a laser rangefinder. Furthermore, traditional targets cannot be attached to the surface of the cable-stayed bridge. Therefore, calibration methods relying on targets or rangefinders are abandoned, and a line segment measurement calibration method is adopted instead. This method is based on the natural texture features of the cable-stayed bridge surface (such as spiral lines), and achieves calibration through the mapping relationship between the pixel length of line segments in the image and their actual physical length, avoiding dependence on external targets and rangefinders.

[0087] In a preferred embodiment, a feature point optimization selection mechanism is established to further filter feature points; densely intersecting points of the spiral lines on the surface of the cable-stayed bridge are selected as feature points, utilizing their high texture contrast to ensure sub-pixel level positioning accuracy and avoid tracking errors caused by feature loss in smooth areas; the Sobel operator is used to calculate the image gradient response, establishing a candidate set of feature points with gradient values ​​>128, filtering out points with low contrast and susceptibility to noise interference, thus improving the reliability of feature points; a grid layout strategy is adopted to divide the image into multiple sub-regions, and feature points are uniformly selected in each sub-region to avoid calibration parameter deviations caused by local clustering and ensure balanced global calibration accuracy.

[0088] Simultaneously, based on the theory of redundant observation, by increasing the number of characteristic point observations (e.g., selecting n characteristic points), the number of system equations is increased to n times. The least squares method or weighted least squares method is then used to fit and solve multiple sets of observation data. This process is equivalent to "averaging" single-point noise, avoiding the impact of measurement biases caused by environmental interference (such as sudden changes in illumination or local vibrations) on the overall results. For example, when a characteristic point produces an outlier due to noise interference, the observation data of the other n-1 characteristic points can cancel out this outlier through redundant calculations, thereby effectively smoothing out random errors.

[0089] In long-distance measurements, images are susceptible to factors such as atmospheric turbulence and slight camera shake, leading to random errors in feature point localization. Redundant observations, by increasing the number of independent observations, make the calibration results more robust to these random errors. For example, traditional single-point calibration may result in deviations due to a single shake, while redundant observations can control the error to a very small range through cross-validation of multiple sets of data.

[0090] Furthermore, a transformation model for pixel-physical space coordinate mapping is established based on the redundant observation method, as follows:

[0091] Establish a set of feature points Define pixel coordinates To physical coordinates The conversion relationship is as follows:

[0092] ,

[0093] ,

[0094] in, These are the local calibration coefficients for each point. and The offset is determined through subsequent collaborative calibration methods to achieve a precise mapping from pixel displacement to physical displacement.

[0095] Redundant observations, through a multi-feature-point, multi-equation observation system, provide a large amount of low-noise pixel coordinate data for the mathematical model. This data, after being processed by least squares or weighted least squares, can lead to a more accurate solution. , , Parameters such as ) are used to avoid model fitting bias caused by insufficient data. For example, when there are 100 feature points, the model can solve for 3 unknown parameters through 100 sets of equations, forming an overdetermined system of equations, and then use the least squares method to obtain the optimal solution.

[0096] The mathematical model converts pixel coordinate data obtained from redundant observations into coordinate values ​​with actual physical meaning, making subsequent geometric constraints (such as line segment length consistency verification) and distortion correction (such as radial distortion compensation) possible. For example, after calculating the physical coordinates of feature points through the model, it is possible to verify whether the mapping relationship between the physical distance of a known length line segment on the cable-stayed bridge surface and the pixel distance conforms to the theoretical value, thereby further optimizing the calibration parameters.

[0097] Based on the established conversion model, the initial calibration parameters are iteratively optimized using a collaborative calibration method (RANSAC robust estimation + weighted fusion).

[0098] In a preferred embodiment, the collaborative calibration method iteratively optimizes the initial calibration parameters through RANSAC robust estimation and weighted fusion:

[0099] The RANSAC robust estimation randomly selects no fewer than 4 feature points, calculates the transformation matrix H, counts the number of interior points with reprojection error < 2 pixels, and saves the optimal transformation parameters.

[0100] The weighted fusion is calculated by assigning weights based on the stability of feature points, using the following formula:

[0101] ,

[0102] in, For the first Standard deviation of tracking error for each feature point The corner response value ensures that feature points with high stability have a greater weight in the calibration, thus optimizing the overall accuracy. As a global error benchmark; For feature points; For the first Adaptive weights for each feature point.

[0103] For example: Obtain the pixel coordinates of 100 feature points through redundant observations, and substitute them into a mathematical model to obtain 100 sets of... Initial values, after RANSAC filtering out 20 outliers, the remaining 80 values ​​are used to calculate the final value using a weighted fusion formula. .

[0104] Meanwhile, the output of the collaborative calibration method (optimal) The applicability of the mathematical model can be verified in reverse: if the parameters after weighted fusion have a large deviation in multi-scale verification, the constraints of the mathematical model can be adjusted (such as adding nonlinear terms) to form a closed loop of "modeling-optimization-correction" and improve the overall calibration accuracy.

[0105] (3) Displacement measurement

[0106] Please refer to Figure 3 The displacement of test points is measured based on the digital image correlation (DIC) method; and by fusing multi-scale feature fusion method, the problem of insufficient accuracy in measuring the dynamic displacement of cable stays in long distances and complex environments is solved.

[0107] The DIC method selects a texture-rich region on the cable-stayed bridge surface (such as the intersection of the spiral lines) as a reference subset in the initial frame, and then tracks the displacement of the subset in subsequent frames through grayscale correlation analysis. The ZNCC (zero-mean normalized cross-correlation) method is used to calculate the pixel displacement of the subset between adjacent frames, and the pixel-to-physical conversion ratio (such as 0.852033 mm / px) obtained from camera calibration is combined to convert the pixel displacement into actual physical displacement.

[0108] The multi-scale feature fusion tracking method establishes a two-layer feature extraction mechanism, which includes local feature layer extraction driven by DIC and global feature layer extraction driven by CNN.

[0109] The local feature layer extraction dynamically sets the subset size (2M+1)×(2M+1) based on the cable-stayed bridge texture density. Dense texture regions use small subsets (M not greater than 8) to ensure positioning accuracy; smooth regions use large subsets (M not less than 16) to enhance grayscale correlation. Texture richness is then selected by calculating the texture richness index of each subset, as follows:

[0110] ,

[0111] In the formula, The standard deviation of grayscale This is the grayscale average; only retain the values ​​from the previous value. A subset of features is used as tracking points to avoid tracking failure in smooth regions due to lack of texture; Texture richness index for subset; The set texture richness threshold;

[0112] The global feature layer extracts the input cable ROI image and outputs a cable contour probability map. It also extracts global contour features of the cable through semantic segmentation to compensate for the shortcomings of local features in DIC (Discrete Injection) under large-scale vibration or occlusion. A boundary enhancement loss function is used to simultaneously optimize contour pixel classification accuracy and boundary localization accuracy, ensuring the spatial localization accuracy of global features, as detailed below:

[0113] ,

[0114] in, For the total loss, The pixel classification loss measures the overlap between the predicted contour and the true contour. The boundary localization loss measures the distance between the predicted contour boundary and the true boundary.

[0115] In a preferred embodiment, the multi-scale feature fusion tracking method employs a weighted adaptive fusion strategy, as follows:

[0116] ,

[0117] in, The final feature weights after fusion; Weights for local features of DIC; The weights of the global features in the CNN; The weighting coefficients are calculated as follows:

[0118] ,

[0119] To match the quality of DIC, comprehensive consideration is needed. The correlation coefficient and the standard deviation of subregion displacement are calculated using the following formulas:

[0120] ,

[0121] In the formula, Indicates the first A subset Correlation coefficient; Indicates the first The standard deviation of displacement of each subset; This represents the total number of DIC subsets participating in the evaluation;

[0122] To improve the quality of CNN contour extraction, we evaluate it using contour integrity and image signal-to-noise ratio.

[0123] In a preferred embodiment, the multi-scale feature fusion tracking method further includes a two-layer tracking and failure recovery mechanism, wherein the two-layer tracking process is as follows:

[0124] Extract the complete outline of the cable using a lightweight semantic segmentation network (such as a variant of U-Net), and output a discrete set of points along the centerline to achieve robust estimation of large-scale displacement:

[0125] ,

[0126] In the formula, The discrete set of points along the global centerline of the cable; For the first The coordinates of discrete points along the centerline;

[0127] The global displacement field is obtained by differentiating the centerline point sets between frames:

[0128] ,

[0129] In the formula, For the global displacement field of the cable; for The set of points along the center line of the cable at any given moment; for The set of points along the center line of the cable at any given moment; For time frame indexing;

[0130] When texture is missing (such as oil stains on the surface of cables) or there are sudden changes in lighting (such as strong light at the tunnel entrance), the semantic robustness of CNN contours can maintain basic tracking and avoid "cliff-like failure" of local feature matching.

[0131] In global displacement Within a defined "small search window" (e.g., ±5 pixels), the IC-GN (Iterative Closest-Gauss-Newton) method is applied to the DIC subset (texture-rich area) to compensate for the "coarse-grained error" of global localization; by optimizing the gray-level correlation of pixels within the subset, the sub-pixel-level displacement residual is solved. The final output is the fused displacement:

[0132] ,

[0133] In the formula, This represents the final high-precision displacement output; This indicates a coarse estimate of displacement at the global layer. This indicates the local layer fine-tuning displacement.

[0134] In a preferred embodiment, the failure recovery process specifically includes:

[0135] When local DIC tracking faces the risk of failure due to poor matching quality or abnormal displacement residuals, the system activates a "global feature fallback + expanded range rematch" mechanism to ensure continuous and reliable displacement measurement, as detailed below:

[0136] The presence of any of the following warning signals indicates that local tracking is unreliable:

[0137] Gray-scale matching index of DIC subset This means that interference such as texture blurring and sudden changes in lighting can cause the reliability of local feature matching to drop below 50%; local displacement residuals This includes abrupt shifts in displacement, feature mismatches, etc., which cause the residuals to exceed three times the standard deviation of the noise, deviating from the reasonable range.

[0138] If any of the above conditions are met, local tracking is determined to have failed, triggering the following recovery process:

[0139] A lightweight semantic segmentation network is invoked to re-extract the complete outline of the cable and generate a new set of centerline points. Based on the semantic robustness of the contour (independent of texture details), the tracking anchor points are quickly reconstructed to avoid direct data loss.

[0140] The search window for the DIC subset is expanded from a small, conventional range to the possible areas predicted by the global layer, and the texture is rematched using the IC-GN method. By "expanding the search radius," areas with abrupt displacements or missing textures are covered, thereby improving the success rate of DIC rematch.

[0141] The multi-scale feature fusion tracking method, through the collaborative mechanism of "local detail measurement + global structural constraints", not only ensures sub-pixel accuracy in texture-rich areas, but also makes up for the tracking shortcomings in complex environments through global features. This upgrades displacement measurement from a "microscopic mode" that relies solely on texture to a "macro-microscopic linkage mode" that integrates structural information, providing a reliable displacement data foundation for bridge health monitoring.

[0142] (4) Cable force calculation

[0143] After completing the acquisition of video data of the stay cables, high-precision calibration modeling, and stable displacement measurement, dynamic displacement measurement results reflecting the vibration of the stay cables were obtained. These results record the displacement changes of the stay cables at different times, containing rich time-domain information.

[0144] Subsequently, the dynamic displacement measurement results were subjected to Fast Fourier Transform (FFT). FFT is an efficient frequency domain analysis method that can convert time-domain displacement signals into frequency-domain signals, extracting the natural frequencies of the cable-stayed bridge's vibration. Through FFT processing, the originally complex time-domain displacement curve was decomposed into a superposition of different frequency components, clearly revealing the frequency characteristics of the cable-stayed bridge's vibration. These frequencies are closely related to the physical properties of the cable-stayed bridge.

[0145] Finally, based on the extracted natural frequency and combined with the structural parameters of the stay cable (such as length and linear density), the cable force is calculated using relevant mechanical formulas, as follows:

[0146] ,

[0147] ,

[0148] In the formula, Indicates tension; Indicates the linear density of the cable; Indicates the calculated length of the cable; Indicates the order of the natural frequency; Indicates the bending stiffness of the cable; These represent the vibration characteristic parameters of the stay cables.

[0149] Based on the principles of structural dynamics, this calculation process establishes a mathematical relationship between frequency and cable force. By accurately substituting parameters and performing formula calculations, the vibration frequency information is converted into specific cable force values, thereby achieving a quantitative assessment of the stress state of the stay cables and providing key data support for bridge structural health monitoring.

[0150] Example 2

[0151] Please refer to Figure 4 This embodiment 2 provides a rapid measurement system for cable tension based on digital image correlation method, including:

[0152] The video acquisition unit is used to acquire video data of the cable-stayed bridge.

[0153] The calibration modeling unit is used to establish a feature point optimization selection mechanism by adopting the line segment measurement calibration method and taking the cable-stayed bridge texture points as feature points; it establishes a pixel-physical space coordinate mapping transformation model based on the redundant observation method; and it calculates the model parameters through the collaborative calibration method.

[0154] The displacement measurement unit is used for displacement measurement based on the Digital Image Correlation (DIC) method; a two-layer feature extraction mechanism is established to complete local and global feature extraction; and the weight allocation of local and global features is dynamically adjusted through a weight adaptive fusion strategy; finally, a two-layer tracking and failure recovery mechanism is established to complete the hierarchical collaboration of "coarse-grained global tracking" and "fine-grained local tracking".

[0155] The cable force calculation unit is used to perform a fast Fourier transform on the dynamic displacement measurement results to calculate the cable force of the cable.

[0156] Example 3

[0157] This embodiment 3 also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it can implement any step of a method for rapid measurement of cable force based on digital image correlation.

[0158] The computer-readable storage medium may include various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0159] For a description of the computer-readable storage medium provided in this application, please refer to the above method embodiments; further details will not be repeated here.

[0160] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A rapid measurement method for cable force of a stay cable based on digital image correlation, characterized in that, include: S1. Collect video data of the cable-stayed bridge; S2. A feature point optimization selection mechanism is established using the line segment measurement calibration method with the cable-stayed bridge texture points as feature points; a pixel-physical space coordinate mapping transformation model is established based on the redundant observation method; and the model parameters are calculated through the collaborative calibration method. S3. Displacement measurement is performed based on the Digital Image Correlation (DIC) method; a two-layer feature extraction mechanism is established to complete local and global feature extraction; the weight allocation of local and global features is dynamically adjusted through a weight adaptive fusion strategy; finally, a two-layer tracking and failure recovery mechanism is established to complete the hierarchical collaboration between "coarse-grained global tracking" and "fine-grained local tracking"; S4. Perform a fast Fourier transform on the dynamic displacement measurement results to calculate the cable force of the stay cables; The feature point optimization selection mechanism in S2 is as follows: The dense intersection points of the spiral lines on the surface of the cable-stayed bridge are selected as feature points in the cable-stayed bridge texture points. Their high texture contrast is used to ensure sub-pixel level positioning accuracy and avoid tracking errors caused by feature loss in smooth areas. The Sobel operator is used to calculate the gradient response of the image, and a candidate set of feature points with gradient values ​​>128 is established to filter out low-contrast and noise-sensitive points, thereby improving the reliability of feature points. A grid layout strategy is adopted to divide the image into multiple sub-regions. Feature points are selected evenly in each sub-region to avoid calibration parameter deviations caused by local clustering and ensure balanced global calibration accuracy. The collaborative calibration method in S2 iteratively optimizes the initial calibration parameters through RANSAC robust estimation and weighted fusion. The RANSAC robust estimation randomly selects no fewer than 4 feature points, calculates the transformation matrix H, counts the number of interior points with reprojection error < 2 pixels, and saves the optimal transformation parameters. The weighted fusion is calculated by assigning weights based on the stability of feature points, using the following formula: ， in, For the first Standard deviation of tracking error for each feature point The corner response value ensures that feature points with high stability have a greater weight in the calibration, thus optimizing the overall accuracy. As a global error benchmark; For feature points; For the first Adaptive weights for each feature point; The dual-layer tracking process in S3 is specifically as follows: A lightweight semantic segmentation network is used to extract the complete outline of the cable and output a discrete set of points along the centerline to achieve robust estimation of large-scale displacement. ， In the formula, The discrete set of points along the global centerline of the cable; For the first The coordinates of discrete points along the centerline; The global displacement field is obtained by differentiating the centerline point sets between frames: ， In the formula, For the global displacement field of the cable; for The set of points along the center line of the cable at any given moment; for The set of points along the center line of the cable at any given moment; For time frame indexing; When texture is missing or lighting changes abruptly, the semantic robustness of CNN contours can maintain basic tracking and avoid "cliff-like failure" of local feature matching. In global displacement Within a defined "small search window," the IC-GN method is applied to the DIC subset to compensate for the "coarse-grained error" of global positioning; by optimizing the gray-level correlation of pixels within the subset, the sub-pixel-level displacement residual is solved. The final output is the fused displacement: ， In the formula, This represents the final high-precision displacement output; This indicates a coarse estimate of displacement at the global layer. This indicates the local layer fine-tuning displacement.

2. The method for rapid measurement of cable force in a stay cable based on digital image correlation according to claim 1, characterized in that, The specific transformation model for pixel-physical space coordinate mapping in S2 is as follows: Establish a set of feature points Define pixel coordinates To physical coordinates The conversion relationship is as follows: ， ， in, These are the local calibration coefficients for each point. and The offset is determined through subsequent collaborative calibration methods to achieve a precise mapping from pixel displacement to physical displacement.

3. The method for rapid measurement of cable force in a stay cable based on digital image correlation according to claim 1, characterized in that, The dual-layer feature extraction mechanism in S3 includes local feature layer extraction driven by DIC and global feature layer extraction driven by CNN. The local feature layer extraction dynamically sets the subset size (2M+1)×(2M+1) based on the cable-stayed bridge texture density. Dense texture regions use small subsets (M not greater than 8) to ensure positioning accuracy; smooth regions use large subsets (M not less than 16) to enhance grayscale correlation. Texture richness is then selected by calculating the texture richness index of each subset, as follows: ， In the formula, The standard deviation of grayscale This is the grayscale average; only retain the grayscale value. A subset of features is used as tracking points to avoid tracking failure in smooth regions due to lack of texture; Texture richness index for subset; The set texture richness threshold; The global feature layer extracts the input cable ROI image and outputs a cable contour probability map. It also extracts global contour features of the cable through semantic segmentation to compensate for the shortcomings of local features in DIC (Discrete Injection) under large-scale vibration or occlusion. A boundary enhancement loss function is used to simultaneously optimize contour pixel classification accuracy and boundary localization accuracy, ensuring the spatial localization accuracy of global features, as detailed below: ， in, For the total loss, The pixel classification loss measures the overlap between the predicted contour and the true contour. The boundary localization loss measures the distance between the predicted contour boundary and the true boundary.

4. The method for rapid measurement of cable force in a stay cable based on digital image correlation as described in claim 1, characterized in that, The weight adaptive fusion strategy in S3 is as follows: ， in, The final feature weights after fusion; Weights for local features of DIC; These are the weights of the global features in the CNN; The weighting coefficients are calculated as follows: ， To match the quality of DIC, comprehensive consideration is needed. The correlation coefficient and the standard deviation of subregion displacement are calculated using the following formulas: ， In the formula, Indicates the first A subset Correlation coefficient; Indicates the first The standard deviation of displacement of each subset; This represents the total number of DIC subsets participating in the evaluation; To improve the quality of CNN contour extraction, we evaluate it using contour integrity and image signal-to-noise ratio.

5. The method for rapid measurement of cable force in a stay cable based on digital image correlation according to claim 1, characterized in that, The failure recovery process in S3 is specifically as follows: When local DIC tracking faces the risk of failure due to poor matching quality or abnormal displacement residuals, the system activates a "global feature fallback + expanded range rematch" mechanism to ensure continuous and reliable displacement measurement, as detailed below: The presence of any of the following warning signals indicates that local tracking is unreliable: Gray-scale matching index of DIC subset Local displacement residuals ; If any of the above conditions are met, local tracking is determined to have failed, triggering the following recovery process: A lightweight semantic segmentation network is invoked to re-extract the complete outline of the cable and generate a new set of centerline points. Based on the semantic robustness of the contour, the tracking anchor points are quickly reconstructed to avoid direct data loss. The search window for the DIC subset is expanded from a small, conventional range to the possible areas predicted by the global layer, and the texture is rematched using the IC-GN method. By "expanding the search radius," areas with abrupt displacements or missing textures are covered, thereby improving the success rate of DIC rematch.

6. A rapid measurement system for cable tension in a stay cable based on digital image correlation, characterized in that, include: The video acquisition unit is used to acquire video data of the cable-stayed bridge. The calibration modeling unit is used to establish a feature point optimization selection mechanism by adopting the line segment measurement calibration method and taking the cable-stayed bridge texture points as feature points; it establishes a pixel-physical space coordinate mapping transformation model based on the redundant observation method; and it calculates the model parameters through the collaborative calibration method. The displacement measurement unit is used for displacement measurement based on the Digital Image Correlation (DIC) method; a two-layer feature extraction mechanism is established to complete local and global feature extraction; and the weight allocation of local and global features is dynamically adjusted through a weight adaptive fusion strategy; finally, a two-layer tracking and failure recovery mechanism is established to complete the hierarchical collaboration of "coarse-grained global tracking" and "fine-grained local tracking". The cable force calculation unit is used to perform a fast Fourier transform on the dynamic displacement measurement results to calculate the cable force of the stay cables; The feature point optimization selection mechanism in the calibration modeling unit is as follows: The dense intersection points of the spiral lines on the surface of the cable-stayed bridge are selected as feature points in the cable-stayed bridge texture points. Their high texture contrast is used to ensure sub-pixel level positioning accuracy and avoid tracking errors caused by feature loss in smooth areas. The Sobel operator is used to calculate the gradient response of the image, and a candidate set of feature points with gradient values ​​>128 is established to filter out low-contrast and noise-sensitive points, thereby improving the reliability of feature points. A grid layout strategy is adopted to divide the image into multiple sub-regions. Feature points are selected evenly in each sub-region to avoid calibration parameter deviations caused by local clustering and ensure balanced global calibration accuracy. The collaborative calibration method in the calibration modeling unit iteratively optimizes the initial calibration parameters through RANSAC robust estimation and weighted fusion. The RANSAC robust estimation randomly selects no fewer than 4 feature points, calculates the transformation matrix H, counts the number of interior points with reprojection error < 2 pixels, and saves the optimal transformation parameters. The weighted fusion is calculated by assigning weights based on the stability of feature points, using the following formula: ， in, For the first Standard deviation of tracking error for each feature point The corner response value ensures that feature points with high stability have a greater weight in the calibration, thus optimizing the overall accuracy. As a global error benchmark; For feature points; For the first Adaptive weights for each feature point; The dual-layer tracking process in the displacement measurement unit is specifically as follows: A lightweight semantic segmentation network is used to extract the complete outline of the cable and output a discrete set of points along the centerline to achieve robust estimation of large-scale displacement. ， In the formula, The discrete set of points along the global centerline of the cable; For the first The coordinates of discrete points along the centerline; The global displacement field is obtained by differentiating the centerline point sets between frames: ， In the formula, For the global displacement field of the cable; for The set of points along the center line of the cable at any given moment; for The set of points along the center line of the cable at any given moment; For time frame indexing; When texture is missing or lighting changes abruptly, the semantic robustness of CNN contours can maintain basic tracking and avoid "cliff-like failure" of local feature matching. In global displacement Within a defined "small search window," the IC-GN method is applied to the DIC subset to compensate for the "coarse-grained error" of global positioning; by optimizing the gray-level correlation of pixels within the subset, the sub-pixel-level displacement residual is solved. The final output is the fused displacement: ， In the formula, This represents the final high-precision displacement output; This indicates a coarse estimate of displacement at the global layer. This indicates the local layer fine-tuning displacement.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, The computer program is executed by a processor as described in any one of claims 1-5: a method for rapid measurement of cable force based on digital image correlation.

Images