A high-precision bridge structure global strain measurement method and system

Abstract

The application discloses a kind of high-precision bridge structure global strain measurement method and system, and it is related to civil engineering-structural health monitoring technical field.The system includes: video acquisition module is arranged in bridge stable observation point, and standard frame rate video is collected;Pixel domain pretreatment module constructs space-time matrix, generates weight matrix and obscuration mask;Multi-scale space adjacency graph module is divided node according to one-level component-two-stage segmentation-superpixel clustering, and edge weight is defined and geometric priori constraint is applied;Variational displacement solving module obtains displacement field by multi-constrained objective function and ADMM iteration;Self-adapting baseline strain calculation module dynamically selects baseline length and calculates strain;Abnormal point processing module ensures data integrity by judging-eliminating-reestimating process;Cross-section synchronous constraint module reduces displacement deviation.The method solves the problems of large traditional calibration error, local measurement limitation and environmental interference, and guarantees the reliability of bridge health monitoring data.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of civil engineering-structure health monitoring, and particularly relates to a high-precision bridge structure global strain measurement method and system. BACKGROUND

[0002] Traditional bridge strain measurement technology mainly relies on physical calibration components (such as strain gauges and displacement meters) and local point measurement methods, which have significant limitations:

[0003] Calibration dependence and error problem: the measurement reference needs to be established through physical calibration components, which is complex to deploy on site and easy to introduce calibration errors, making it difficult to adapt to changes in the dynamic environment of the bridge.

[0004] Local measurement limitations: traditional methods can only obtain strain data at discrete points and cannot achieve global continuous measurement, making it difficult to reflect the overall stress distribution of the bridge.

[0005] Environmental interference sensitivity: environmental factors such as obstructions (vehicles / pedestrians) and changes in light can easily lead to measurement errors, and the noise suppression capability is weak, especially in the absence of calibration.

[0006] Baseline fixation defects: fixed baseline length leads to the contradiction between large noise for short baseline and low resolution for long baseline, making it difficult to adapt to the strain characteristics of different regions of the bridge (such as the flat region in the middle of the main beam and the complex curvature region at the bottom of the tower).

[0007] In the prior art, although computer vision has been used to optimize local measurement, it still faces problems such as poor global coordination, weak environmental adaptability, and unstable precision. For example, traditional optical flow or digital image correlation (DIC) technology requires pre-calibration and is sensitive to obstructions and changes in light. While multi-sensor fusion methods expand the measurement range, they have high calibration complexity and difficulty in data synchronization and consistency processing.

[0008] The present application proposes a high-precision global strain measurement system and method based on pixel domain video processing to address the above problems. SUMMARY

[0009] To overcome the shortcomings and deficiencies of the prior art, the first object of the present application is to provide a high-precision bridge structure global strain measurement method, and the second object of the present application is to provide a high-precision bridge structure global strain measurement system.

[0010] The first object of the present application adopts the following technical solutions:

[0011] A high-precision bridge structure global strain measurement method, comprising the following steps:

[0012] Standard frame rate video acquisition; space-time matrix, weight matrix and occlusion mask construction, extract video brightness channel reorganization into space-time matrix, calculate weight matrix W, generate occlusion mask O using GMM background modeling combined with optical flow vector amplitude;

[0013] Multi-scale spatial adjacency graph construction, node division according to primary component-secondary segmentation-superpixel clustering, determination of adjacency relationship and edge weight, generation of Laplace matrix by applying geometric prior constraint;

[0014] Variational displacement field solving, construction of variational objective function containing brightness consistency residual term, spatial Figure 1 consistency term, time polynomial constraint term and bandwidth constraint term, ADMM iteration is used for solving, and the convergence criterion is set as the relative decline of the objective function being lower than the set value, and the global displacement time history u(y, t) is output;

[0015] Adaptive baseline strain calculation, local SNR of different Lg is calculated for each spatial position of u(y, t), the Lg with the maximum SNR is selected, and the strain field ε(y, t)=du / dy is calculated by using first-order weighted orthogonal polynomial regression;

[0016] Abnormal point processing, after discriminating and removing abnormal points, the abnormal point strain is reconstructed based on the edge weight interpolation of the multi-scale adjacency graph;

[0017] Cross-section synchronous constraint, the cross-section synchronous constraint is selected, the main beam is uniformly distributed, the synchronous deviation penalty term is introduced into the variational objective function for iteration optimization, and the final global strain field ε(y, t) is output.

[0018] The second object of the application adopts the following technical scheme:

[0019] A high-precision bridge structure global strain measurement system is used to realize a high-precision bridge structure global strain measurement method, and the system comprises a video acquisition module, a pixel domain data preprocessing module, a multi-scale spatial adjacency graph construction module, a variational displacement solving module, an adaptive baseline strain calculation module, an abnormal point processing module and a cross-section synchronous constraint module.

[0020] The video acquisition module is arranged at a stable observation point around the bridge, acquires standard frame rate video covering the target area of the main beam, the bridge tower and the cable clamp of the bridge, and directly processes data in the pixel domain without relying on external physical calibration components.

[0021] The pixel domain data preprocessing module receives the image sequence output by the video acquisition module, constructs a space-time matrix M, generates a weight matrix W based on the texture stability of the bridge deck, the feature tracking confidence, the historical consistency weighted calculation, and the occlusion mask O obtained by background modeling combined with motion saliency detection.

[0022] The multi-scale spatial adjacency graph construction module divides nodes into three levels of primary components-secondary segmentation-superpixel clustering, defines edge weights based on geometric priori + texture consistency, and applies Figure 1 consistency constraints;

[0023] The variational displacement solving module constructs a variational objective function containing a brightness consistency residual term, a spatial Figure 1 consistency term, a time polynomial constraint term, and a bandwidth constraint term, and solves the global displacement time history u(y, t) using an iterative algorithm;

[0024] The adaptive baseline strain calculation module adaptively selects the baseline length Lg by maximizing the local signal-to-noise ratio or minimizing the information criterion, and calculates the strain by taking the spatial derivative of u(y, t) using weighted orthogonal polynomial regression;

[0025] The abnormal point processing module discriminates abnormal points based on spatiotemporal consistency, and reconstructs the displacement of abnormal points after removal through graph interpolation;

[0026] The cross-section synchronous constraint module selects bridge transverse uniformly distributed sections, introduces a synchronous deviation penalty term and integrates it into the variational objective function, to reduce the displacement deviation of different sections.

[0027] Preferably, the stable observation points of the video acquisition module are bridge piers or shore supports, the collected videos cover the key component areas of the bridge girder, bridge tower, and cable clamp, and the core features do not depend on external physical calibration elements. The image sequence data is directly processed in the pixel domain, reducing deployment complexity and calibration errors.

[0028] Preferably, in the pixel domain data preprocessing module, the calculation formula of the weight matrix W is ; wherein is the comprehensive weight value at pixel position y and time t; α, β, and γ are weight coefficients, ; is the bridge deck texture stability weight; is the feature tracking confidence weight; is the historical consistency weight; in the occlusion mask O, O(y, t) = 1 indicates unoccluded valid pixels, and O(y, t) = 0 indicates occluded invalid pixels. The background modeling uses a Gaussian Mixture Model (GMM), and the motion saliency detection uses optical flow anomaly area identification.

[0029] Preferably, in the node division of the multi-scale spatial adjacency graph construction module, the primary nodes are independent components of the bridge girder, bridge tower, and cable clamp, the secondary nodes are segmented based on the component axis and span or height, and the basic nodes are superpixels obtained by performing superpixel segmentation on the secondary node area. The adjacency relationship determination needs to satisfy that the two basic nodes belong to the same component and the spatial distance is ≤ a preset threshold. The edge weight calculation formula is Wherein, d is the node center distance, λ is the balance coefficient, the non-adjacent node edge weight ω=0; and the component axis geometry prior constraint is applied, the main beam node only communicates with the adjacent main beam node, and the main beam node displacement deviation in the axial direction is ≤ a set percentage.

[0030] Preferably, in the variational displacement solving module, the brightness consistency residual term is , wherein Ω (u; I) is the brightness consistency residual, indicating the deviation of the pixel brightness after displacement from the original brightness, Ω (u; I) = I (y + u (y, t), t) - I (y, t), W is the weight matrix generated by the pixel domain preprocessing module, O is the weight matrix and the occlusion mask, O (y, t) = 1 for effective pixels, O (y, t) = 0 for occluded pixels, and ρ is the Huber robust loss function.

[0031] Space Figure 1 The consistency term is , wherein , Laplace matrix of the pixel-level and region-level spatial adjacency graph respectively, The consistency constraint term is Figure 1 . , The weight coefficient of the spatial constraint is

[0032] The time polynomial constraint term is , wherein The time domain piecewise polynomial consistency constraint is deg≤2, and the displacement field u (y, t) in each time window is punished , wherein is a second-order polynomial, and β is the weight coefficient of the time constraint.

[0033] The bandwidth constraint term is , wherein The bandwidth selection constraint is represented by the spectral component in the target frequency band, , is the Fourier transform, and μ is the weight coefficient of the bandwidth constraint.

[0034] The iterative algorithm is the alternating direction multiplier method ADMM or the proximal gradient method, and the convergence criterion is that the relative decrease of the objective function is < a set threshold value a or the augmented Lagrangian residual is < a set threshold value b.

[0035] Preferably, in the adaptive baseline strain calculation module, when calculating the strain, the first-order orthogonal polynomial is adopted in the region with small bridge curvature and flat strain distribution, the strain formula is , the second-order orthogonal polynomial is adopted in the region with large bridge curvature and complex strain distribution, and the strain formula is wherein p(y) is a corresponding orthogonal polynomial, 、 、 is a coefficient of the orthogonal polynomial, and e(y, t) is a strain value at the position y and the time t, and is a coefficient solved by minimizing a weighted residual sum of squares.

[0036] Preferably, the outlier point discrimination condition of the outlier point processing module is that the displacement u(y, t) of a certain pixel deviates from the average displacement of a 3*3 pixel neighborhood by more than a set multiple of the standard deviation, or the strain e(y, t) exceeds the elastic strain limit value of the bridge material; the neighborhood re-estimation adopts a graph interpolation method, and the formula is 、 is an edge weight of the outlier point and the neighborhood node.

[0037] As described above, due to the adoption of the technical solutions described above, the present application has the following beneficial effects:

[0038] 1. The present application realizes full-domain strain continuous measurement without physical calibration components by using pixel domain video direct processing technology. Combined with spatial-time matrix reorganization and multi-scale spatial neighborhood graph constraint, the present application solves the problem of large calibration error and the problem that traditional local measurement cannot cover the full domain, and adapts to the demand of bridge overall stress distribution evaluation.

[0039] 2. The present application adopts four-dimensional collaborative variation optimization of brightness consistency, spatial graph smoothing, time polynomial and bandwidth constraint, and cooperates with ADMM iterative solution to effectively suppress environmental interference such as occlusion and light change. Through dynamic weight matrix (fusing texture stability, tracking confidence, and historical consistency) and Huber loss function filtering of abnormal values, the stability and noise resistance of the displacement field under complex environment are ensured, and the robustness is enhanced.

[0040] 3. The present application solves the contradiction between large noise of short baseline and low resolution of long baseline by adaptive baseline length selection, and adapts to the strain characteristics of the flat area in the middle of the main beam and the complex curvature area of the bridge tower. The abnormal points are reconstructed by spatio-temporal consistency discrimination and graph interpolation method, combined with cross-section synchronous constraint, to ensure that the full-domain strain data is complete and meets the bridge mechanical characteristics (such as symmetric distribution of main beam transverse strain). BRIEF DESCRIPTION OF DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor based on these drawings.

[0042] Figure 1A module diagram of a high-precision bridge structure global strain measurement system is shown.

[0043] Figure 2 A flow chart of a high-precision bridge structure global strain measurement method is shown.

[0044] Figure 3 An effect comparison diagram of the technical scheme of the present application and prior art is shown.

[0045] Figure 4 An effect diagram of data processing of the present application is shown. DETAILED DESCRIPTION

[0046] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0047] In addition, the described features, structures or characteristics can be combined in any suitable manner in one or more example embodiments. In the following description, many specific details are provided to give a sufficient understanding of example embodiments of the present disclosure. However, one skilled in the art will realize that the technical solutions of the present disclosure can be practiced without one or more of the specific details, or can employ other methods, components, steps, etc. In other cases, well-known structures, methods, implementations or operations are not shown or described in detail to avoid obscuring aspects of the present disclosure.

[0048] Embodiment 1

[0049] Referring to Figure 1 The present application provides a high-precision bridge structure global strain measurement system, which aims to solve the problems of traditional bridge strain measurement, such as dependence on physical calibration, large limitation of local measurement, influence of environmental interference (shading / illumination) on precision, and realization of global strain accurate acquisition without external calibration. The system comprises:

[0050] Video acquisition module: deployed at stable observation points (such as piers, shore supports) around the bridge, acquires standard frame rate video covering target areas such as main girder, bridge tower and cable clamp, and the core feature is to directly process data in the pixel domain without relying on external physical calibration components (such as strain gauges and displacement meters), thereby reducing the complexity of field deployment and calibration error.

[0051] Pixel domain data preprocessing module: receives the image sequence output by the video acquisition module and performs two core processes:

[0052] 1. Constructing space-time matrix M: reorganize the time-series intensity values of each pixel in the spatial pixel coordinate (y) - time (t) dimension to form (H is the image height, W is the width, and T is the number of time frames);

[0053] 2. Generating weight matrix W and occlusion mask O:

[0054] Weight matrix W: weighted calculation based on bridge texture stability (the smaller the texture gradient variance, the higher the weight), feature tracking confidence (the smaller the optical flow tracking matching error, the higher the weight), and historical consistency (the smaller the deviation of the current pixel brightness from the historical same period, the higher the weight), which is expressed by the formula ; is the comprehensive weight value at pixel position y and time t, reflecting the reliability of the pixel for subsequent displacement calculation (the higher the value, the more reliable the pixel); α, β, γ are weight coefficients, reflecting the reliability of the pixel for subsequent displacement calculation (the higher the value, the more reliable the pixel), , dynamically adjusted according to the bridge scene (such as beam bridge / cable-stayed bridge, different component areas), balancing the influence weight of "texture stability, tracking confidence, and historical consistency"; is the texture stability weight, the smaller the texture gradient variance (the more stable the texture, the more reliable the displacement calculation), the higher the value; is the feature tracking confidence weight, the smaller the optical flow tracking matching error (the more reliable the tracking result), the higher the value; is the historical consistency weight, the smaller the deviation of the current pixel brightness from the historical same period (such as the same period the day before), the higher the value (the brightness change conforms to the historical law).

[0055] Occlusion mask O: static bridge background is extracted using background modeling (such as Gaussian Mixture Model GMM), and vehicle occlusion and pedestrian interference areas are marked by combining motion saliency detection (such as optical flow anomaly area identification), O(y, t) = 1 indicates no occlusion (valid pixel), and O(y, t) = 0 indicates occlusion (invalid pixel).

[0056] Multi-scale space adjacency graph construction module: used for constructing hierarchical adjacency graphs according to the geometric characteristics of different components of the bridge, realizing the spatial coherence constraint of the displacement field:

[0057] Node division: divided into three levels of "primary component - secondary segment - superpixel cluster":

[0058] Primary node: independent components such as main girder, tower, and cable clamp (such as the main girder divided into 1 primary node, and the tower divided into 2 primary nodes);

[0059] Secondary node: based on the component axis and span segmentation (e.g. the main beam is divided into three secondary nodes according to the span 1 / 4, 1 / 2, 3 / 4, and the bridge tower is divided into one secondary node every 10m in height);

[0060] Basic node: perform superpixel segmentation (e.g. SLIC algorithm) on each secondary node area, and each superpixel is a basic graph node.

[0061] Edge weight definition: determine the adjacent edge weight based on "geometric prior + texture consistency", if two basic nodes belong to the same secondary node and the distance is ≤ preset threshold (e.g. 5 pixels), the edge weight ; wherein d is the node center distance, λ is the balance coefficient, and the non-adjacent node edge weight ω = 0.

[0062] Figure 1 Consistency constraint: limit the sudden change of displacement field in space through the connectivity of adjacent graph (e.g. main beam nodes only connect with adjacent main beam nodes, not with bridge tower nodes) and component axis geometric prior (e.g. the deviation of main beam node displacement in the axial direction needs to be ≤ 5%).

[0063] Variational displacement solving module: based on multiple constraint conditions, the displacement field u(y, t) (y is the spatial coordinate and t is the time) is optimized, and the core is to construct a target function containing multiple dimensional constraints:

[0064] Variational target function:

[0065] ; wherein, is the brightness consistency residual term; Ω(u; I) is the brightness consistency residual, which represents the deviation of pixel brightness after displacement from the original brightness, i.e. Ω(u; I) = I(y + u(y, t), t) - I(y, t); W is the weight matrix generated by the pixel domain preprocessing module; O is the weight matrix and the occlusion mask, O(y, t) = 1 for valid pixels and O(y, t) = 0 for occluded pixels; ρ is a robust loss function (e.g. Huber loss, which suppresses the influence of abnormal values in the occluded area), which is used to filter the occluded area and low confidence pixels, and suppress the interference of abnormal values on the solution;

[0066] is the spatial Figure 1 consistency term; , and are the Laplacian matrices of the pixel-level and regional-level spatial adjacency graph (generated by the multi-scale spatial adjacency graph construction module), which are used to describe the connectivity between nodes and smoothness constraints; is the Figure 1 consistency constraint term, which ensures the smoothness of the displacement field in space (avoids displacement sudden change) through the quadratic form of the Laplacian matrix; , For the weight coefficient of spatial constraint, it is dynamically adjusted according to the bridge type (beam bridge / cable-stayed bridge) or the component area (such as the cable clamp area which needs stronger spatial constraint), and the balance of different scales (pixel level, regional level) of spatial smoothing strength.

[0067] For the time polynomial constraint term; For the time domain piecewise polynomial consistency constraint, the displacement field u(y, t) in each time window is subjected to a deviation penalty not exceeding the second-order polynomial (deg≤2) (such as ), where is the second-order polynomial, which ensures the stability of the displacement field in time (adapted to the slow-changing characteristics of bridge vibration); β is the weight coefficient of time constraint, which balances the influence of time stability and other constraints.

[0068] For the bandwidth constraint term; For the bandwidth selection constraint, the formula is ; where is the Fourier transform, is the spectral component in the target frequency band (such as 0.1-5Hz, the main frequency range of bridge vertical vibration), which is used to suppress high-frequency noise outside the target frequency band (such as wind-induced interference); μ is the weight coefficient of bandwidth constraint, which adjusts the strength of frequency band constraint.

[0069] Solution method: use alternating direction multiplier method (ADMM) or proximal gradient method to solve iteratively, and the convergence criteria are "relative decrease of objective function " or "augmented Lagrangian residual ", and the final output is the global displacement time history u(y, t).

[0070] Adaptive baseline strain calculation module: strain field is solved by spatial derivative, and the core is to adaptively determine the baseline length Lg, avoiding the problem of "large noise with short baseline and low resolution with long baseline" caused by fixed baseline:

[0071] Adaptive selection of baseline length Lg:

[0072] Method 1: maximize the local signal-to-noise ratio (SNR), calculate the local SNR=10lg(signal power / noise power) under different Lg (such as 5-20 pixels), and select the Lg corresponding to the maximum SNR.

[0073] Method 2: minimize the information criterion, such as AIC criterion (AIC=2k-2lnL, k is the model parameter, and L is the likelihood function value) or SURE criterion (Stein unbiased risk estimate), select the Lg with the minimum criterion value.

[0074] Strain calculation: spatial derivatives of u(y, t) are calculated using weighted orthogonal polynomial regression. If first-order orthogonal polynomials are chosen (P1(y) = 1, P2(y) = y), then strain ; where p(y) is used to fit the first-order orthogonal polynomials of the displacement field u(y, t), which adapts to the region of the bridge with small curvature and smooth strain distribution (e.g., the midspan of the main girder); , are the coefficients of the orthogonal polynomials, which are solved by minimizing the weighted residual sum of squares (the weights are the W values of the neighboring pixels); and ε(y, t) is the strain value at position y and time t, which is obtained from the spatial first-order derivative of the displacement field. If second-order polynomials are chosen (P1(y) = 1, P2(y) = y, P3(y) = y2), then , which adapts to the region of the bridge with large curvature and complex strain distribution (e.g., the bottom of the tower); where p(y) is used to fit the second-order orthogonal polynomials of the displacement field u(y, t), which adapts to the region of the bridge with large curvature and complex strain distribution (e.g., the bottom of the tower); , , are the coefficients of the orthogonal polynomials, which are solved by minimizing the weighted residual sum of squares (the weights are the W values of the neighboring pixels); and ε(y, t) is the strain value at position y and time t, which is obtained from the spatial first-order derivative of the displacement field (the first-order derivative of the second-order polynomial).

[0075] Abnormal point processing module: for displacement / strain abnormal values caused by vehicle occlusion and sudden changes in light, the "discrimination-elimination-reestimation" process is performed:

[0076] Abnormal point discrimination: based on "spatiotemporal consistency", if the displacement u(y, t) of a pixel deviates from the average displacement of the 3x3 neighborhood by more than 3 times the standard deviation, or the strain ε(y, t) exceeds the elastic strain limit of the bridge material (e.g., ε>150με for concrete bridges and ε>300με for steel bridges), it is determined to be an abnormal point;

[0077] Abnormal point elimination: mark the abnormal point and set it as invalid;

[0078] Neighborhood reestimation: based on the connectivity of the multiscale spatial adjacency graph, the "graph interpolation method" is used to reconstruct the displacement of the abnormal point, such as , is the estimated value of the displacement of the abnormal point; is the edge weight between the abnormal point and the neighborhood nodes, ensuring the integrity of the global data; is the displacement of the neighborhood node j in the spatial coordinates y and time t.

[0079] Cross-section synchronous constraint module: eliminate local drift (e.g., cross-section displacement deviation caused by single sensor error) and improve overall strain distribution consistency:

[0080] ​​Cross-section selection: Select the cross-sections evenly distributed in the bridge transverse direction (e.g., one transverse cross-section every 5m of the main girder, including left, middle, and right three measuring points);

[0081] Synchronization constraint application: For the displacement time history of the ith transverse cross-section at the same time t (i=1,2,...,n, n is the number of transverse cross-sections), introduce a synchronization deviation penalty term:

[0082] ;

[0083] The synchronization deviation penalty term is used to reduce the displacement deviation of different transverse cross-sections, ensuring that the transverse strain distribution of the bridge conforms to the mechanical properties (e.g., the transverse strain of the main girder is symmetrically distributed).

[0084] By incorporating into the variational objective function (as an additional constraint term), the displacement deviation of different cross-sections is reduced through iterative optimization, ensuring that the transverse strain distribution conforms to the mechanical properties of the bridge (e.g., the transverse strain of the main girder should be symmetrically distributed).

[0085] The beneficial effects of the embodiment are: the system realizes precise global strain measurement without external calibration through pixel domain processing, multi-scale graph constraint, and adaptive strain calculation. It solves the problems of traditional methods relying on physical calibration, local measurement limitations, and environmental interference, improves the accuracy and reliability of bridge health monitoring, adapts to different bridge structures and environmental conditions, and ensures the accuracy of structural safety evaluation.

[0086] Embodiment 2:

[0087] Preferably, the core of the multi-scale space adjacency graph construction module is to ensure the spatial coherence of the displacement field through "hierarchical node division + geometric prior constraint", and the specific steps are as follows:

[0088] Step 1: Component region segmentation and superpixel clustering

[0089] Input: Bridge image output by the video acquisition module (including main girder, bridge tower, and cable clamp region);

[0090] Component segmentation: Use a semantic segmentation algorithm (such as U-Net) to identify and label different component regions (main girder in red, bridge tower in blue, and cable clamp in yellow);

[0091] Superpixel clustering: Perform SLIC superpixel segmentation on each component region (superpixel size: 5x5 pixels for the main girder region, 3x3 pixels for the bridge tower region, and 2x2 pixels for the cable clamp region, adapting to the size difference of components), and each superpixel is taken as a basic graph node, denoted as V={v1,v2,...,vm} (m is the total number of superpixels).

[0092] Step 2: Adjacency relationship and edge weight calculation.

[0093] Adjacency relationship determination: If two base nodes vi, vj satisfy the following conditions, it is determined to be adjacent: a. Belong to the same component (such as main beam component); b. Spatial distance d(vi, vj) ≤ preset threshold (main beam region d ≤ 8 pixels, bridge tower region d ≤ 5 pixels, cable clamp region d ≤ 3 pixels);

[0094] Edge weight calculation: adopt "texture consistency + geometric distance" weighting model: ; Wherein, , is the texture gradient of vi, vj (the smaller the gradient, the more stable the texture), is the maximum texture gradient of the region; is the distance threshold of the component, to ensure that the edge weight

[0095] , the more consistent the texture and the closer the distance, the greater the edge weight.

[0096] Step 3: Multi-scale graph construction and consistency constraint.

[0097] Pixel-level graph : Constructed on the basis of superpixels, Laplacian matrix ; (D1 is the degree matrix, ; A1 is the adjacency matrix, ), constraint pixel-level displacement smoothing;

[0098] Region-level graph : Cluster superpixels of the same secondary node (such as main beam 1 / 2 span) as "region node", construct region-level adjacency graph, Laplacian matrix (D2 is the region node degree matrix, A2 is the region node adjacency matrix), constraint inter-region displacement continuity;

[0099] Geometric prior constraint: for main beam region, force "node displacement deviation along the axis direction ≤ 5%" (i.e. if vi, vj are distributed along the main beam axis), , multiply the penalty coefficient of 0.1 to the edge weight , strengthen the geometric consistency.

[0100] Preferably, the core of the variational displacement solving module is to obtain a displacement field that takes into account "brightness consistency, spatial smoothness, temporal stability, and frequency band effectiveness" through multi-constraint collaborative optimization. The specific process is as follows:

[0101] Step 1: Time segmentation and polynomial modeling.

[0102] Input: space-time matrix M, weight matrix W, and occlusion mask O;

[0103] Space-time matrix M: a matrix that reorganizes the time-series intensity values of each pixel in the dimension of "spatial pixel coordinate (y) - time (t)", with the dimension of H x W x T (H is the image height, W is the image width, and T is the total number of frames of time).

[0104] Weight matrix W: reflects the reliability of pixels for "displacement calculation" (weighted by texture stability, feature tracking confidence, and historical consistency, with higher values indicating more reliable pixels).

[0105] Occlusion mask O: marks whether a pixel is occluded, with O(y, t) = 1 indicating "no occlusion (valid pixel)" and O(y, t) = 0 indicating "occlusion (invalid pixel)".

[0106] Segment point detection: calculate the sliding window variance of the intensity consistency residual Ω(u; I) (window size = 5 frames), and if the variance mutation rate > 30% (e.g., from 0.1 to 0.4), set it as a time segmentation point, dividing the total time T into K time segments (e.g., T = 1000 frames, divided into K = 5 time segments, each with 200 frames).

[0107] Polynomial modeling: for each time segment , set the displacement field (second-order polynomial, suitable for the slow-changing characteristics of bridge vibration), and satisfy the C0 continuity between segments . Wherein, is the time range of the kth time segment, is the start frame of the segment, is the end frame of the segment; wherein , , are polynomial coefficients that vary with spatial coordinates y; C0 continuity: the displacement continuity constraint between time segments, i.e., the end displacement of the previous segment is equal to the start displacement of the next segment.

[0108] Step 2: Objective function construction and constraint integration.

[0109] Intensity consistency residual term: , where , I(y, t) is the intensity value of the original image at (y, t); Filter occluded areas and low-confidence pixels, and ρ is the Huber loss function (suppresses outliers), with the rule that when |x| ≤ δ, ρ(x) = x² / 2, and when |x| > δ, ρ(x) = δ|x| - δ² / 2, δ = 0.1 is a hyperparameter;

[0110] Spatial Figure 1 consistency term: , which constrains the spatial smoothness of the displacement field through the Laplacian matrix; is the spatial constraint weight coefficient; , is the Laplacian matrix of the pixel-level, region-level spatial adjacency graph (characterizing the connectivity between nodes and the smoothness constraint); is the vector representation of the displacement field.

[0111] Time polynomial term: , = 0.4, ensuring the displacement field complies with the time-domain polynomial variation rule; is the time constraint weight coefficient; is the value of the actual displacement field at time period tk.

[0112] is the second-order polynomial model of time period tk (same as in “Time segmentation and polynomial modeling”).

[0113] Bandwidth constraint term: , is the bandwidth constraint weight coefficient; is the spectral component within the 0.1-5 Hz frequency band (the main frequency of the vertical vibration of the bridge is usually in this range), which suppresses high-frequency noise; is the Fourier transform of the displacement field u (converts displacement from time domain to frequency domain).

[0114] Step 3: ADMM iterative solution.

[0115] Initialization: ; is the initial displacement field, assuming the displacement to be 0 before iteration starts; the initial value of the Lagrange multiplier is , which is used for the dual variable update of the ADMM algorithm; the penalty parameter ρ = 1;

[0116] Iteration steps: (1) u update: fixing , ρ, minimize the objective function to get ; (2) dual variable update: ; (3) convergence judgment: if , stop iteration; otherwise, return to step 1, where is the displacement field of the kth iteration, and if the relative change of the displacement fields of adjacent two iterations is less than , it is considered that the iteration has converged.

[0117] Output: global displacement time history u(y, t).

[0118] The beneficial effects of the embodiment are: the embodiment improves the spatial continuity of the displacement field by hierarchical node division and geometric constraints, and realizes collaborative and accurate solution of brightness, space, time and frequency band by combining multi-constraint variational optimization, enhances the precision, robustness and environmental adaptability of global strain measurement of bridges, and guarantees the reliability of health monitoring data and the accuracy of structural safety evaluation.

[0119] Embodiment 3:

[0120] The complete steps of the high-precision bridge structure global strain measurement method refer to Figure 2 , and in combination with the above system modules, the complete measurement method includes the following 7 steps:

[0121] Step 1: Standard frame rate video acquisition.

[0122] Deploy industrial cameras (resolution 1920x1080, frame rate 25fps) on bridge piers or shore supports, align the lens to the target areas such as main girder, bridge tower, cable clamp, etc., and continuously acquire video data for ≥30 minutes; without arranging physical calibration components (such as checkerboard calibration board), directly process the image sequence in the pixel domain to avoid the influence of calibration error.

[0123] Step 2: Construction of space-time matrix, weight matrix and occlusion mask.

[0124] Extract the brightness channel of each frame of image in the video, and reorganize it into a space-time matrix according to the "spatial coordinates (y)-time (t)". (30 minutes x 25fps = 7500 frames);

[0125] Calculate the weight matrix W: , wherein is the inverse of the texture gradient variance; is the optical flow tracking matching score (0-1), is the inverse of the brightness deviation of the current frame and the historical same period frame (such as the same period of the previous day);

[0126] Generate the occlusion mask O: use GMM background modeling to extract the static bridge deck, combine the optical flow vector amplitude (>5 pixels / frame to determine the moving target), mark the vehicle and pedestrian occlusion area, and O(y, t) = 1 (no occlusion) or 0 (occlusion).

[0127] Step 3: Construction of multi-scale space adjacency graph.

[0128] Divide the nodes according to "primary component (main girder / bridge tower / cable clamp)-secondary segmentation (main girder segmented by span, bridge tower segmented by height)-superpixel clustering", construct two layers of adjacency graph at pixel level and region level; apply geometric prior constraints (such as the displacement of main girder nodes deviating along the axis direction ≤5%), determine the adjacency edge weight, and generate the Laplacian matrix , .

[0129] Step 4: Variational displacement field solving.

[0130] Constructing variational objective function (including intensity consistency residual, multiscale graph constraint, temporal polynomial constraint, bandwidth constraint);

[0131] Solving by ADMM iteration, setting the convergence criterion as the relative decrease of the objective function , obtaining the global displacement time history u(y, t);

[0132] Setting the bandwidth constraint as 0.1-5Hz (adapted to the main frequency of concrete beam bridge), suppressing high-frequency noise (such as wind-induced high-frequency vibration).

[0133] Step 5: Adaptive baseline strain calculation.

[0134] For each spatial position y of u(y, t), calculate the local SNR of different baseline lengths Lg (5-20 pixels), select the Lg with the maximum SNR (such as Lg=15 pixels in the main beam midspan area and Lg=8 pixels in the cable clamp area); take the spatial derivative of u(y, t) by first-order weighted orthogonal polynomial regression, obtaining the strain field ε(y, t)=du / dy.

[0135] Step 6: Abnormal point processing.

[0136] Distinguishing abnormal points: if ε(y, t)>150με (concrete bridge) or the deviation from the average strain of the neighborhood is >3 times the standard deviation, mark it as an abnormal point; after removing the abnormal points, interpolate and reconstruct the abnormal point strain based on the edge weight of the multiscale adjacency graph, ensuring the integrity of the global strain data.

[0137] Step 7: Cross-sectional synchronous constraint.

[0138] Selecting 5 cross sections (5m apart) in the main beam, imposing a synchronous deviation penalty on the cross-sectional displacement time history at each time t; after iterative optimization, output the final global strain field ε(y, t), with a strain measurement accuracy of ±5με (verified on a 30m simply supported beam bridge, with a deviation of <3% from the strain gauge measured value).

[0139] The embodiment has the beneficial effects that: the embodiment realizes high-precision measurement of global bridge strain through non-calibration video acquisition, multi-constraint variational optimization, adaptive baseline strain calculation, and intelligent processing of abnormal points, solves the problems of large calibration error, local measurement limitations, and environmental interference of traditional methods, improves the reliability of health monitoring data, and ensures the accuracy of structural safety evaluation.

[0140] Figure 3The effect comparison chart of the technical scheme of the present application and the prior art is shown: in the laboratory strain test, the strain gauge has high precision, and is used as a true value comparison, the patent method and the strain gauge are basically consistent in the figure, and the error of the traditional method is large.

[0141] Figure 4 The effect chart of the data processing of the present application is shown: the two laboratory strain tests of instantaneous impact in the figure are almost consistent, and the strain gauge is used as a reference.

[0142] The formulas of the present application are dimensionless values, and the preset parameters in the formulas are set by the person skilled in the art according to the actual situation.

[0143] The weight coefficient of the present application is used to measure the influence degree of different factors or variables on a certain result or decision. The definition of the weight coefficient is that a value is assigned to each factor when comparing and evaluating multiple factors to reflect its importance or priority. These weight coefficients can be determined according to specific circumstances and needs, and are usually formulated and confirmed by professionals or relevant parties. By reasonably setting the weight coefficient, the program or system can make more accurate decisions or predictions.

[0144] The above is only the preferred embodiment of the present application, but the protection scope of the present application is not limited to this. Any person skilled in the art can make equivalent replacements or changes within the technical scope disclosed by the present application according to the technical scheme and the inventive concept of the present application, which should be covered within the protection scope of the present application.

[0145] The preferred embodiments of the present application disclosed above are only used to help explain the present application. The preferred embodiments do not describe all the details and do not limit the present application to the specific embodiments. Obviously, many modifications and changes can be made according to the content of the present application. The present application selects and describes these embodiments in order to better explain the principles and practical applications of the present application, so that the person skilled in the art can well understand and utilize the present application. The present application is limited by the claims and their entire scope and equivalents.

Claims

1. A high-precision bridge structure global strain measurement method, characterized in that, The method comprises the following steps: Standard frame rate video acquisition; weight matrix and occlusion mask construction, extracting the luminance channel data of the video sequence, and reorganizing to generate a space-time matrix M, and generating a weight matrix W based on the matrix calculation; at the same time, an occlusion mask O is generated by using GMM background modeling combined with optical flow vector amplitude analysis; Multi-scale space adjacency graph construction, node division according to primary component-secondary segmentation-superpixel clustering, determination of adjacency relationship and edge weight, and application of geometric prior constraint to generate Laplace matrix; Variational displacement field solving, construction of a variational objective function containing a luminance consistency residual term, a spatial graph consistency term, a time polynomial constraint term and a bandwidth constraint term, iterative solving by using ADMM, and setting the convergence criterion as the relative decrease of the objective function being lower than a set value, and outputting the global displacement time history u(y,t); Adaptive baseline strain calculation, calculation of the local SNR of different Lg for each spatial position of u(y,t), selection of the Lg with the maximum SNR, and calculation of the strain field ε(y,t)=du / dy by using first-order weighted orthogonal polynomial regression; Abnormal point processing, discrimination and elimination of abnormal points, and reconstruction of the abnormal point strain based on the edge weight interpolation of the multi-scale adjacency graph; Cross-section synchronous constraint, selection of uniformly distributed cross sections of the main beam, introduction of a synchronous deviation penalty term into the variational objective function for iterative optimization, and output of the final global strain field ε(y,t).

2. A high-precision bridge structure global strain measurement system for implementing a high-precision bridge structure global strain measurement method as claimed in claim 1, characterized in that, The system comprises a video acquisition module, a pixel domain data preprocessing module, a multi-scale space adjacency graph construction module, a variational displacement solving module, an adaptive baseline strain calculation module, an abnormal point processing module and a cross-section synchronous constraint module; The pixel domain data preprocessing module receives the image sequence output by the video acquisition module, constructs a space-time matrix M, generates a weight matrix W based on the bridge deck texture stability, feature tracking confidence and historical consistency weighted calculation, and generates an occlusion mask O by using background modeling combined with motion saliency detection; The multi-scale space adjacency graph construction module divides nodes according to three levels of primary component-secondary segmentation-superpixel clustering, defines edge weight based on geometric prior+texture consistency, and applies graph consistency constraint; The variational displacement solving module constructs a variational objective function containing a luminance consistency residual term, a spatial graph consistency term, a time polynomial constraint term and a bandwidth constraint term, and solves and outputs the global displacement time history u(y,t) by using an iterative algorithm; The adaptive baseline strain calculation module adaptively selects the baseline length Lg by maximizing the local signal-to-noise ratio or minimizing the information criterion, and calculates the strain by taking the spatial derivative of u(y,t) by using weighted orthogonal polynomial regression; The abnormal point processing module discriminates abnormal points based on spatiotemporal consistency, eliminates the abnormal points, and reconstructs the displacement of the abnormal points by using graph interpolation method.

3. The high precision bridge structure global strain measurement system of claim 2, wherein, The video acquisition module is arranged at a stable observation point around the bridge, acquires a standard frame rate video covering a target area of the main beam, the tower and the cable clamp of the bridge, and directly processes data in the pixel domain without relying on external physical calibration components; the stable observation point of the video acquisition module is a bridge pier or a shore support, the acquired video covers the key component area of the main beam, the tower and the cable clamp of the bridge, the core features do not rely on external physical calibration components, the image sequence data is directly processed in the pixel domain, and the deployment complexity and the calibration error are reduced.

4. The high precision bridge structure global strain measurement system of claim 2, wherein, The calculation formula of the weight matrix W in the pixel domain data preprocessing module is ; wherein, is a comprehensive weight value at a pixel position y and a time t; a, β, and γ are weight coefficients, ; is a bridge texture stability weight; is a feature tracking confidence weight; is a history consistency weight; O(y, t) = 1 in the occlusion mask O indicates a non-occluded valid pixel, O(y, t) = 0 indicates an occluded invalid pixel, a Gaussian mixture model GMM is used for background modeling, and an optical flow abnormal region identification is used for motion saliency detection.

5. The high precision bridge structure global strain measurement system of claim 2, wherein, In the node division of the multi-scale spatial adjacency graph construction module, the first-level nodes are the independent components of the bridge main beam, the bridge tower and the cable clamp, the second-level nodes are segmented based on the component axis and the span or height, and the basic nodes are superpixels obtained by performing superpixel segmentation on the second-level node region; the adjacency relationship determination needs to meet the condition that two basic nodes belong to the same component and the spatial distance is less than or equal to a preset threshold, and the edge weight calculation formula is wherein d is the node center distance, λ is the balance coefficient, and the edge weight ω of the non-adjacent node is 0; and the component axis geometric prior constraint is applied, the main beam node is only connected with the adjacent main beam node, and the displacement deviation of the main beam node in the axial direction is less than or equal to a certain percentage.

6. The high precision bridge structure global strain measurement system of claim 2, wherein, The brightness consistency residual term in the variational displacement solving module is Wherein, Ω (u; I) is a brightness consistency residual, indicating the deviation of pixel brightness after displacement from original brightness, Ω (u; I) = I (y + u (y, t), t) - I (y, t), W is a weight matrix generated by the pixel domain preprocessing module, O is a weight matrix and an occlusion mask, O (y, t) = 1 is a valid pixel, O (y, t) = 0 is an occluded pixel, and p is a Huber robust loss function. The spatial graph consistency term is wherein, , are Laplacian matrices of the pixel-level and region-level spatial adjacency graph, respectively, is a graph consistency constraint term; , is a weight coefficient of the spatial constraint; The time polynomial constraint term is wherein, is a time domain piecewise polynomial consistency constraint, which imposes a deviation penalty not exceeding a second order polynomial deg < 2 on the displacement field u(y, t) within each time window wherein is a second order polynomial, and β is a weight coefficient of the time constraint; The bandwidth constraint term is wherein, is a bandwidth selection constraint, representing spectral components within a target frequency band, , is a Fourier transform, and μ is a weight coefficient for the bandwidth constraint. The iterative algorithm is an alternating direction multiplier method ADMM or a proximal gradient method, and the convergence criterion is that a relative decrease of the objective function is less than a set threshold a or an augmented Lagrangian residual is less than a set threshold b.

7. The high precision bridge structure global strain measurement system of claim 2, wherein, In the adaptive baseline strain calculation module, when calculating the strain, a first-order orthogonal polynomial is adopted for a small bridge curvature and a gentle strain distribution region , and a strain formula is , a second-order orthogonal polynomial is adopted for a large bridge curvature and a complex strain distribution region , and a strain formula is , wherein p(y) is a corresponding orthogonal polynomial, 、 、 is a coefficient of the orthogonal polynomial, ε(y, t) is a strain value at a position y and a time t, and is a coefficient solved by minimizing a weighted residual square sum.

8. The high precision bridge structure global strain measurement system of claim 2, wherein, The abnormal point discrimination condition of the abnormal point processing module is that the displacement u(y, t) of a pixel and the average displacement deviation of the neighborhood 3*3 pixels are greater than the standard deviation of the set multiple, or the strain e(y, t) exceeds the elastic strain limit value of the bridge material; the neighborhood re-estimation adopts a graph interpolation method, and the formula is , is the edge weight of the abnormal point and the neighborhood node.

9. The high precision bridge structure global strain measurement system of claim 2, wherein, The cross-section synchronous constraint module selects bridge transverse uniformly distributed cross sections, introduces a synchronous deviation penalty term and integrates it into a variational objective function, and reduces displacement deviation of different cross sections.

Images