Bridge hydrodynamic load non-contact measurement method based on visual inversion

Abstract

The invention provides a visual inversion-based bridge hydrodynamic load non-contact measurement method. The method comprises the following steps of: realizing layered correction of jitter, drift and parallax error of image acquisition equipment by adopting multi-scale image stabilization processing; accurate mapping from pixel coordinates to physical coordinates is established on the basis of orthorectification, and it is guaranteed that speed and load inversion has real physical significance; artificial tracer particles are replaced by natural textures, so that the material cost and deployment complexity of detection are remarkably reduced; optical flow speed field calculation is performed on a dynamic video sequence through a multi-frame joint optimization model, so that particle-free mapping from visual brightness change to a fluid speed vector is realized, and the estimation precision of the speed field is remarkably improved; through carrying out local consistency weighted filtering processing on a velocity field, a false velocity field is substantially reduced and stability of load inversion is increased. And finally, inputting the filtered velocity field into a fluid dynamic pressure equation, calculating the instantaneous hydrodynamic load of the bridge pier, and realizing high-precision non-contact measurement of the hydrodynamic load of the bridge.

Description

Technical Field

[0001] This invention relates to the field of bridge engineering monitoring technology, and more specifically, to a non-contact measurement method for bridge hydrodynamic loads based on visual inversion. Background Technology

[0002] In natural environments, water flow and waves are among the main sources of dynamic loads on bridges, especially bridge piers. The impact of waves and water flow on bridge structures, particularly in aquatic environments, is complex and unpredictable. Accurately measuring and assessing the dynamic loads on bridge piers under the influence of water flow and waves is crucial for ensuring bridge structural safety, optimizing design, and developing maintenance strategies. Existing dynamic load measurement technologies are mainly divided into two categories: contact and non-contact. Contact measurement technologies, including strain gauges and accelerometers, provide accurate data, but require equipment installation on the bridge, which can compromise structural integrity and increase maintenance costs. Non-contact measurement technologies, including laser ranging and GPS, are significantly affected by environmental factors such as weather conditions and equipment installation location, and their real-time performance and accuracy still need improvement.

[0003] Particle image velocimetry (PIV) is a measurement technique widely used in fluid dynamics research. It tracks the movement of fluids by adding tracer particles to obtain information such as flow velocity and direction. Although PIV technology provides detailed flow field visualization data for fluid mechanics and related fields, it faces the following obvious technical defects and practical challenges when applied to the dynamic load measurement of large structures such as bridge piers: (1) It is difficult to ensure the uniformity of tracer particle distribution, which may have an impact on the ecosystem; (2) The application of PIV technology is limited in different aquatic environments, especially in scenarios with large areas or complex terrain. Summary of the Invention

[0004] The technical problem to be solved by this invention is how to overcome the technical limitations of traditional particle image velocimetry and realize environmentally friendly, high-precision, non-contact measurement of bridge hydrodynamic loads.

[0005] This invention provides a non-contact measurement method for bridge hydrodynamic loads based on visual inversion, comprising: Step 1: Acquire dynamic video sequences containing the bridge piers and the surrounding water flow using image acquisition equipment; Step 2: Perform multi-scale image stabilization and orthorectification on the dynamic video sequence to establish the mapping relationship between image pixel coordinates and actual physical coordinates, and obtain the orthoprojection image sequence. Step 3: Based on the orthophoto image sequence, the grayscale variance and spatial spectrum features of each frame of orthophoto image are calculated to identify natural texture regions, low texture regions, and high-brightness reflection regions. The low texture regions are enhanced and the high-brightness reflection regions are suppressed to generate a texture feature weight map. Step 4: Based on the texture feature weight map, construct a multi-frame joint optimization model that includes temporal consistency constraints and spatial smoothness constraints, perform optical flow calculation on the dynamic video sequence, and inversely determine the velocity field of the fluid surface; Step 5: Perform local consistency weighted filtering on the velocity field; Step 6: Input the filtered velocity field into the hydrodynamic pressure equation to calculate the hydrodynamic pressure distribution on the pier surface; and combine it with the geometric model of the pier to integrate the hydrodynamic pressure distribution on the flow-receiving surface to obtain the instantaneous hydrodynamic load acting on the pier. Step 7: Construct a self-supervised loss function, which includes at least a consistency constraint based on the time variation of dynamic pressure and a physical conservation constraint based on the velocity field divergence; and perform dynamic correction on the velocity field and instantaneous hydrodynamic load by minimizing the self-supervised loss function.

[0006] Compared with existing technologies, this application has the following advantages: It employs multi-scale image stabilization processing to achieve layered correction of image acquisition device jitter, drift, and parallax errors; it establishes a precise mapping from pixel coordinates to physical coordinates based on orthorectification, achieving construction alignment and scale uniformity, ensuring that velocity and load inversion have real physical meaning; next, it uses natural textures instead of artificial tracer particles, significantly reducing the material cost and deployment complexity of detection; and it performs optical flow calculations on dynamic video sequences through a multi-frame joint optimization model to invert the velocity field of the fluid surface, achieving a particle-free mapping from visual brightness changes to fluid velocity vectors, significantly improving the estimation accuracy of the velocity field; then, it significantly reduces false velocity fields and improves the stability of load inversion by performing local consistency weighted filtering on the velocity field; finally, it inputs the filtered velocity field into the hydrodynamic pressure equation to calculate the instantaneous hydrodynamic load of the bridge pier, achieving high-precision non-contact measurement of the bridge's hydrodynamic load.

[0007] In one possible implementation, step 2 specifically includes: Step 201: Construct an image Gaussian pyramid for each video image in the input dynamic video sequence. , Indicates the first Frame video images are used as the lower layer of a Gaussian pyramid. After Gaussian blurring and downsampling, a sequence of images with progressively decreasing resolution is generated. ,in, Represents video images The original resolution, This represents the resolution after Gaussian blurring and downsampling operations applied through an L-layer pyramid. Step 202: Select a static structural region as a reference region in the dynamic video sequence; Step 203: Within the reference region, starting from the top layer of the image Gaussian pyramid, extract salient feature points using a feature point detection algorithm. ; Step 204, for video images Optical flow is used to track video images at the corresponding levels of the Gaussian pyramid. Significant feature points detected in , to obtain in video image The corresponding point set ; Step 205: Use an affine transformation model to describe the global motion between frames, expressed as: ; In the formula, Represents the original coordinates of the pixels in the image. Represents the coordinates of the transformed pixel. and Let be the transformation parameters to be solved; Step 206: Use a robust parameter estimation algorithm to estimate the parameters from the matched feature point set. The global motion model of the top layer of the Gaussian image pyramid is calculated. and the global motion model As The initial estimate of the layered pyramid, in When performing feature tracking using a pyramid structure, the positions of feature points are predicted, and steps 204-206 are repeated until the lower layers of the pyramid are reached, resulting in a representation of the video image. Global motion model that varies to the reference region ; Step 207: Perform an inverse transformation on the global motion change model to obtain... For video images Remapping is performed, and the output is a video sequence stable in the reference region. ; Step 208, based on video sequence Through perspective transformation matrix Perform orthorectification to obtain an orthoprojection image sequence. .

[0008] Compared to existing technologies, multi-scale image stabilization eliminates spurious displacements introduced by the motion of the acquisition device; orthorectification accurately maps image coordinates to physical coordinates. This step ensures that all subsequent velocity and load calculations are quantities with real physical units and geometric meanings, rather than remaining at the pixel level, which is a fundamental prerequisite for achieving quantitative mechanical inversion.

[0009] In one possible implementation, step 3 specifically includes: Step 301, define each pixel in the image Centered A local window; Step 302: Calculate the pixel grayscale value of the pixels within the local window. Gray-scale variance Spectral energy Brightness gradient amplitude ; Step 303: Determine whether the pixels within the local window satisfy the condition. and If yes, then the area within the local window is a high-brightness reflection area, which is suppressed using a brightness gradient mask to avoid light spot interference, and proceed to step 305; if no, then Proceed to step 304; This indicates the preset high brightness threshold for pixels. This indicates the preset high brightness gradient threshold; Step 304: Determine whether the grayscale variance is satisfied. and If yes, the region within the local window is a natural texture region used for tracking, and proceed to step 305; if no, the region within the local window is a low texture region, and guided filtering and histogram equalization are used for image enhancement operations, and proceed to step 305. Step 305: Assign corresponding weights to the amplitude of natural texture regions, enhanced low-texture regions, or suppressed highlight reflection regions; slide the local window and return to step 302 until the local window has traversed the entire image; output the texture feature weight map. .

[0010] Compared with existing technologies, this technology overcomes the dependence of traditional PIV on artificial particles by adaptively identifying natural texture regions, high-brightness reflection regions, and low-texture regions. By intelligently identifying and processing natural textures, it can perform non-contact measurement under complex and ever-changing lighting and water quality conditions. By enhancing effective information (low-texture regions) and suppressing interference information (high-brightness reflection regions), it provides high-quality and highly reliable input for subsequent core calculations.

[0011] In one possible implementation, the grayscale variance of pixels within the local window in step 302 The calculation formula is: ; In the formula, Indicates the pixel position within a local window pixel grayscale values, This represents the average grayscale value of all pixels within a local window. This represents the total number of pixels within the local window; Spectral Energy The calculation is based on performing a two-dimensional discrete Fourier transform on the image patches within a local window to obtain the spectrum. ; then based on Calculate the spectral energy within the local window The calculation formula is: ; In the formula, Represents spatial frequency variables. Indicates the center of the spectrum. This represents the minimum frequency radius of a predefined spectrum. Indicates an indicator function; Brightness gradient amplitude The calculation formula is: ; In the formula, Represents the image brightness function. This represents the rate of change of image brightness in the x-direction. This represents the rate of change of image brightness in the y-direction. This represents the partial derivative.

[0012] In one possible implementation, the expression for the multi-frame joint optimization model in step 4 is: ; In the formula, This represents the velocity components of the fluid surface to be solved, and the velocity field of the fluid surface obtained by inversion. , Indicates the time consistency constraint coefficient, Indicates the spatial smoothing constraint coefficient; The brightness function of the image. Indicates image brightness Spatial gradient, ; Let represent the gradient of the velocity component field in the y-direction. ,in, Represents the velocity component field The rate of change in the x-direction, Represents the velocity component field The rate of change in the y-direction; Let represent the gradient of the velocity component field in the x-direction. ,in, Represents the velocity component field The rate of change in the x-direction, Represents the velocity component field Rate of change in the y-direction.

[0013] In one possible implementation, step 5 specifically includes: Step 501: Calculate adjacency weights based on spatial distance, flow direction similarity, and velocity amplitude differences. The calculation formula is: ; In the formula, Represents two vector points in the velocity field. yes The field point, , Representing vector points respectively Spatial location coordinates, Represents vector points The Euclidean distance between them Represents the radius of a spatial domain. Indicates directional tolerance. Represents the velocity amplitude scale, Representing vector points respectively The velocity vector at that point Represents velocity vector The angle between them Represents velocity vector The Euclidean norm of the difference; Step 502, based on adjacency weight Solve the energy minimization problem to obtain the reconstructed velocity field. The calculation formula is: ; In the formula, Represents the regularization weight coefficient. This represents a robust function used for soft suppression of outliers; Step 503: Calculate the degree of offset in the local velocity distribution using the optimal transmission distribution alignment theory. The calculation formula is: ; In the formula, This represents the velocity distribution within the current vector point's neighborhood. Indicates a robust reference distribution. Represent the Wasserstein-1 distance; if satisfying If so, then the vector point is an outlier, and for the vector point... After performing confidence attenuation correction, the filtered velocity field is obtained. : ; ; In the formula, This represents the dynamic attenuation factor based on optimal transmission distribution alignment. Represents the weighted average of the regions. This indicates a preset threshold.

[0014] Compared with existing technologies, this method strengthens the dominant flow direction through directional consistency constraints (equivalent to "modal weighting") and uses robust estimation and optimal transport distribution alignment theory to suppress outlier vectors at the statistical level, effectively eliminating abnormal velocities caused by environmental interference. It identifies and softly corrects outliers caused by floating objects, bubbles, etc. at the statistical distribution level, exhibiting strong robustness. Finally, it outputs a velocity field with high spatial continuity, high temporal smoothness, and strong physical consistency, significantly improving the stability of subsequent fluid field inversion and load calculation.

[0015] In one possible implementation, in step 6, the filtered velocity field is input into the hydrodynamic pressure equation, and the calculation formula for the hydrodynamic pressure distribution on the pier surface is as follows: ; In the formula, Indicates fluid density, The velocity field after filtering Normal velocity scalar field mapped onto the surface of the bridge pier.

[0016] In one possible implementation, in step 6, the dynamic pressure distribution on the flow-receiving surface is integrated using the geometric model of the bridge pier to obtain the instantaneous hydrodynamic load acting on the bridge pier. The calculation formula is as follows: ; In the formula, This represents the area of ​​the flow-receiving surface of the bridge pier. Represents the differential symbol.

[0017] In one possible implementation, the formula for calculating the self-supervised loss function constructed in step 7 is: ; In the formula, , , Indicates the weighting coefficient. , Indicates the first Frame and the Dynamic pressure distribution in a frame image; Represents the divergence of the velocity field. This indicates the load predicted by the finite element method. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the process of this application. Detailed Implementation

[0019] First, those skilled in the art should understand that these embodiments are merely used to explain the technical principles of the embodiments of this application and are not intended to limit the scope of protection of the embodiments of this application. Those skilled in the art can make adjustments as needed to adapt to specific application scenarios.

[0020] In the description of the embodiments of this application, it should be noted that, unless otherwise explicitly specified and limited, the terms "connected" and "linked" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms in the embodiments of this application based on the specific circumstances.

[0021] In the embodiments of this application, unless otherwise expressly specified and limited, "above" or "below" the second feature can mean that the first feature is in direct contact with the second feature, or that the first feature is in indirect contact with the second feature through an intermediate medium. Furthermore, "above," "on top of," and "over" the second feature can mean that the first feature is directly above or diagonally above the second feature, or simply that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature can mean that the first feature is directly below or diagonally below the second feature, or simply that the first feature is at a lower horizontal level than the second feature.

[0022] The present application will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0023] See Figure 1 As shown in the figure, this application discloses a non-contact measurement method for bridge hydrodynamic loads based on visual inversion, including: Step 1: Collect a dynamic video sequence containing the bridge pier and the surrounding water flow based on the image acquisition device; the image acquisition device includes a drone, a shore-based camera or a mobile terminal; in this embodiment, a drone equipped with a high frame rate camera (1080p, 50fps) is hovered about 3 meters upstream of the bridge pier and the water surface video is captured at a downward angle of 30-45 degrees for about 60 seconds.

[0024] Step 2 involves performing multi-scale image stabilization and orthorectification on the dynamic video sequence to establish a mapping relationship between image pixel coordinates and actual physical coordinates, thereby obtaining an orthoprojected image sequence; specifically including: Step 201: Construct an image Gaussian pyramid for each video image in the input dynamic video sequence. , Indicates the first Frame video images are used as the lower layer of a Gaussian pyramid. After Gaussian blurring and downsampling, a sequence of images with progressively decreasing resolution is generated. ,in, Represents video images The original resolution, This represents the resolution after Gaussian blurring and downsampling operations on an L-layer pyramid.

[0025] Step 202: Select a static structural region as a reference region in the dynamic video sequence; in this embodiment, a clear bridge pier edge corner in the video frame is used as the reference region.

[0026] Step 203: Within the reference region, starting from the top layer of the image Gaussian pyramid, extract salient feature points using a feature point detection algorithm. .

[0027] Step 204, for video images Optical flow is used to track video images at the corresponding levels of the Gaussian pyramid. Significant feature points detected in , to obtain in video image The corresponding point set .

[0028] Step 205: Use an affine transformation model to describe the global motion between frames, expressed as: ; In the formula, Represents the original coordinates of the pixels in the image. Represents the coordinates of the transformed pixel. and The transformation parameters are to be solved.

[0029] Step 206: Use a robust parameter estimation algorithm to estimate the parameters from the matched feature point set. The global motion model of the top layer of the Gaussian image pyramid is calculated. and the global motion model As The initial estimate of the layered pyramid, in When performing feature tracking using a pyramid structure, the positions of feature points are predicted, and steps 204-206 are repeated until the lower layers of the pyramid are reached, resulting in a representation of the video image. Global motion model that varies to the reference region .

[0030] Step 207: Perform an inverse transformation on the global motion change model to obtain... For video images Remapping is performed, thereby stably aligning the background (static objects such as bridge piers) in the video sequence to the coordinate system of the reference frame, ultimately outputting a background-stabilized video sequence. The remaining significant motion in the image represents the actual motion of the water body.

[0031] Step 208, based on video sequence Through perspective transformation matrix Perform orthorectification to obtain an orthoprojection image sequence. .

[0032] In this embodiment, at least three 1-meter-long calibration poles are deployed on-site as scale references. By detecting the positions of these calibration poles in the image, the perspective transformation matrix is ​​calculated, and the image is corrected into an orthographic projection image, so that one pixel in the image corresponds to a fixed size in the physical world (e.g., 0.01 meters). Using calibration objects of known physical size deployed on-site (e.g., calibration poles with a length of 1.0 meter), at least four (more are recommended to improve accuracy) control points are selected in the image. Based on the image pixel coordinates (x, y) of these control points and their corresponding real-world coordinates (X, Y) (usually Z=0), the perspective transformation matrix H is solved; this can be accomplished using algorithms such as Direct Linear Transformation (DLT). Through multi-scale image stabilization processing and orthographic correction, parallax errors caused by slight drift of the UAV platform are eliminated, achieving a one-to-one correspondence between pixels and physical coordinates, providing a reliable spatial reference for subsequent velocity inversion.

[0033] Step 3: Based on the orthophoto image sequence, by calculating the grayscale variance and spatial spectral features of each frame of the orthophoto image, natural texture regions, low-texture regions, and high-brightness reflection regions are identified. Image enhancement is performed on low-texture regions, and suppression is applied to high-brightness reflection regions, generating a texture feature weight map; specifically including: Step 301, define each pixel in the image Centered A local window.

[0034] Step 302: Calculate the pixel grayscale value of the pixels within the local window. Gray-scale variance Spectral energy Brightness gradient amplitude Among them, the grayscale variance of pixels within a local window The calculation formula is: ; In the formula, Indicates the pixel position within a local window pixel grayscale values, This represents the average grayscale value of all pixels within a local window. This represents the total number of pixels within a local window.

[0035] Spectral Energy The calculation is based on performing a two-dimensional discrete Fourier transform on the image patches within a local window to obtain the spectrum. ; then based on Calculate the spectral energy within the local window The calculation formula is: ; In the formula, Represents spatial frequency variables. Indicates the center of the spectrum. This represents the minimum frequency radius of a predefined spectrum. This indicates an indicator function.

[0036] Brightness gradient amplitude The calculation formula is: ; In the formula, Represents the image brightness function. This represents the rate of change of image brightness in the x-direction. This represents the rate of change of image brightness in the y-direction. This represents the partial derivative.

[0037] Step 303: Determine whether the pixels within the local window satisfy the condition. and If yes, then the area within the local window is a high-brightness reflection area, which is suppressed using a brightness gradient mask to avoid light spot interference, and proceed to step 305; if no, then Proceed to step 304; This indicates the preset high brightness threshold for pixels. This indicates the preset high brightness gradient threshold.

[0038] Step 304: Determine whether the grayscale variance is satisfied. and If yes, the region within the local window is a natural texture region used for tracking, and proceed to step 305; if no, the region within the local window is a low texture region, and guided filtering and histogram equalization are used for image enhancement operations, and proceed to step 305. The original image is smoothed by using guided filtering to preserve edges. The output of the guided filter is... It is a local linear model that can be viewed locally as the guide image (usually the original image itself) and the input image; for the low-texture region image after the guide filter, contrast-limited adaptive histogram equalization (CLAHE) is applied; CLAHE effectively enhances the local contrast of low-texture regions by dividing the image into small blocks and performing histogram equalization within each block, while limiting the magnitude of contrast amplification to avoid amplifying noise.

[0039] Step 305: Assign corresponding weights to the amplitude of natural texture regions, enhanced low-texture regions, or suppressed highlight reflection regions; slide the local window and return to step 302 until the local window has traversed the entire image; output the texture feature weight map. .

[0040] By adaptively identifying natural texture regions (such as natural floating objects, water ripple interference fringes, light spot reflections, etc.) used for tracking velocity, suppressing high-brightness reflection regions and enhancing low-texture regions, the system can adaptively lock the most reliable visual features under different lighting and water quality conditions, improve robustness, and lay the foundation for high-precision inversion of fluid velocity fields.

[0041] Step 4: Based on the texture feature weight map, construct a multi-frame joint optimization model that includes temporal consistency constraints and spatial smoothness constraints. Perform optical flow calculation on the dynamic video sequence to retrieve the velocity field of the fluid surface. The expression of the multi-frame joint optimization model is: ; In the formula, This represents the velocity components of the fluid surface to be solved, and the velocity field of the fluid surface obtained by inversion. , Indicates the time consistency constraint coefficient, Indicates the spatial smoothing constraint coefficient; The brightness function of the image. Indicates image brightness Spatial gradient, ; Let represent the gradient of the velocity component field in the y-direction. ,in, Represents the velocity component field The rate of change in the x-direction, Represents the velocity component field The rate of change in the y-direction; Let represent the gradient of the velocity component field in the x-direction. ,in, Represents the velocity component field The rate of change in the x-direction, Represents the velocity component field The rate of change in the y-direction; this embodiment of the application introduces a texture feature weight map. This transforms traditional global optical flow calculation into adaptive calculation based on natural texture confidence; it introduces... The traditional two-frame optical flow is extended to multi-frame spatiotemporal joint optimization, which significantly improves noise resistance and temporal consistency. By minimizing the total energy function, a velocity field that is both consistent with image observation data and spatiotemporally smooth and physically reasonable is obtained.

[0042] Step 5: Perform local consistency weighted filtering on the velocity field; specifically including: Step 501: Calculate adjacency weights based on spatial distance, flow direction similarity, and velocity amplitude differences. The calculation formula is: ; In the formula, Represents two vector points in the velocity field. yes The field point, , Representing vector points respectively Spatial location coordinates, Represents vector points The Euclidean distance between them Represents the radius of a spatial domain. Indicates directional tolerance. Represents the velocity amplitude scale, Representing vector points respectively The velocity vector at that point Represents velocity vector The angle between them Represents velocity vector The difference in Euclidean norm.

[0043] Step 502, based on adjacency weight Solve the energy minimization problem to obtain the reconstructed velocity field. The calculation formula is: ; In the formula, Represents the regularization weight coefficient. This represents a robust function used for soft suppression of outliers.

[0044] To further remove local outliers caused by floating objects, bubbles, etc., optimal transport theory is introduced for statistical outlier suppression. Step 503: Calculate the degree of offset in the local velocity distribution using the optimal transmission distribution alignment theory. The calculation formula is: ; In the formula, This represents the velocity distribution within the current vector point's neighborhood. Indicates a robust reference distribution. Represent the Wasserstein-1 distance; if satisfying If so, then the vector point is an outlier, and for the vector point... After performing confidence attenuation correction, the filtered velocity field is obtained. : ; ; In the formula, This represents the dynamic attenuation factor based on optimal transmission distribution alignment. Represents the weighted average of the regions. This indicates a preset threshold.

[0045] Step 6: Input the filtered velocity field into the hydrodynamic pressure equation to calculate the hydrodynamic pressure distribution on the pier surface; and combine it with the geometric model of the pier to integrate the hydrodynamic pressure distribution on the flow-receiving surface to obtain the instantaneous hydrodynamic load acting on the pier; wherein, the calculation formula for the hydrodynamic pressure distribution on the pier surface by inputting the filtered velocity field into the hydrodynamic pressure equation is as follows: ; In the formula, Indicates fluid density, The velocity field after filtering Normal velocity scalar field mapped onto the surface of the bridge pier.

[0046] Based on the geometric model of the bridge pier, the dynamic pressure distribution on the flow-receiving surface is integrated to obtain the instantaneous hydrodynamic load acting on the bridge pier. The calculation formula is as follows: ; In the formula, This represents the area of ​​the flow-receiving surface of the bridge pier. Represents the differential symbol.

[0047] This enables a direct conversion from visual observation to mechanical quantities, allowing the dynamic load on bridge piers to be obtained without contact sensors, providing crucial input for structural safety assessment.

[0048] Step 7: Construct a self-supervised loss function, which includes at least a consistency constraint based on the time variation of dynamic pressure and a physical conservation constraint based on the velocity field divergence; the formula for calculating the self-supervised loss function is: ; In the formula, , , Indicates the weighting coefficient. , Indicates the first Frame and the Dynamic pressure distribution in a frame image; Represents the divergence of the velocity field. Indicates the load predicted by the finite element method; Next, by minimizing the self-supervised loss function, the velocity field and instantaneous hydrodynamic load are dynamically corrected to obtain the corrected velocity field and instantaneous hydrodynamic load.

[0049] In the description of the embodiments of this application, it should be noted that the terms "inner" and "outer" and other terms indicating direction or positional relationship are based on the direction or positional relationship shown in the drawings. This is only for the convenience of description and does not indicate or imply that the device or component must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this application.

[0050] In the description of this application, the references to terms such as "an embodiment," "some embodiments," "in this embodiment," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in a suitable manner in any one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0051] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A non-contact measurement method for bridge hydrodynamic loads based on visual inversion, characterized in that, include: Step 1: Acquire dynamic video sequences containing the bridge piers and the surrounding water flow using image acquisition equipment; Step 2: Perform multi-scale image stabilization and orthorectification on the dynamic video sequence to establish the mapping relationship between image pixel coordinates and actual physical coordinates, and obtain the orthoprojection image sequence. Step 3: Based on the orthophoto image sequence, the grayscale variance and spatial spectrum features of each frame of orthophoto image are calculated to identify natural texture regions, low texture regions, and high-brightness reflection regions. The low texture regions are enhanced and the high-brightness reflection regions are suppressed to generate a texture feature weight map. Step 4: Based on the texture feature weight map, construct a multi-frame joint optimization model that includes temporal consistency constraints and spatial smoothness constraints, perform optical flow calculation on the dynamic video sequence, and inversely determine the velocity field of the fluid surface; Step 5: Perform local consistency weighted filtering on the velocity field; Step 6: Input the filtered velocity field into the hydrodynamic pressure equation to calculate the hydrodynamic pressure distribution on the pier surface; and combine it with the geometric model of the pier to integrate the hydrodynamic pressure distribution on the flow-receiving surface to obtain the instantaneous hydrodynamic load acting on the pier. Step 7: Construct a self-supervised loss function, which includes at least a consistency constraint based on the time variation of dynamic pressure and a physical conservation constraint based on the velocity field divergence; and perform dynamic correction on the velocity field and instantaneous hydrodynamic load by minimizing the self-supervised loss function.

2. The non-contact measurement method for bridge hydrodynamic loads based on visual inversion according to claim 1, characterized in that, Step 2 specifically includes: Step 201: Construct an image Gaussian pyramid for each video image in the input dynamic video sequence. , Indicates the first Frame video images are used as the lower layer of a Gaussian pyramid. After Gaussian blurring and downsampling, a sequence of images with progressively decreasing resolution is generated. ,in, Represents video images The original resolution, This represents the resolution after Gaussian blurring and downsampling operations applied through an L-layer pyramid. Step 202: Select a static structural region as a reference region in the dynamic video sequence; Step 203: Within the reference region, starting from the top layer of the image Gaussian pyramid, extract salient feature points using a feature point detection algorithm. ; Step 204, for video images Optical flow is used to track video images at the corresponding levels of the Gaussian pyramid. Significant feature points detected in , to obtain in video images The corresponding point set ; Step 205: Use an affine transformation model to describe the global motion between frames, expressed as: ； In the formula, Represents the original coordinates of the pixels in the image. Represents the coordinates of the transformed pixel. and Let be the transformation parameters to be solved; Step 206: Use a robust parameter estimation algorithm to estimate the parameters from the matched feature point set. The global motion model of the top layer of the Gaussian image pyramid is calculated. and the global motion model As The initial estimate of the layered pyramid, in When performing feature tracking using a pyramid structure, the positions of feature points are predicted, and steps 204-206 are repeated until the lower layers of the pyramid are reached, resulting in a representation of the video image. Global motion model that varies to the reference region ; Step 207: Perform an inverse transformation on the global motion change model to obtain... For video images Remapping is performed, and the output is a video sequence stable in the reference region. ; Step 208, based on video sequence Through perspective transformation matrix Perform orthorectification to obtain an orthoprojection image sequence. .

3. The non-contact measurement method for bridge hydrodynamic loads based on visual inversion according to claim 2, characterized in that, Step 3 specifically includes: Step 301, define each pixel in the image Centered A local window; Step 302: Calculate the pixel grayscale value of the pixels within the local window. Gray-scale variance Spectral energy Brightness gradient amplitude ; Step 303: Determine whether the pixels within the local window satisfy the condition. and If yes, then the area within the local window is a high-brightness reflection area, which is suppressed using a brightness gradient mask to avoid light spot interference, and proceed to step 305; if no, then Proceed to step 304; This indicates the preset high brightness threshold for pixels. This indicates the preset high brightness gradient threshold; Step 304: Determine whether the grayscale variance is satisfied. and If yes, the region within the local window is a natural texture region used for tracking, and proceed to step 305; if no, the region within the local window is a low texture region, and guided filtering and histogram equalization are used for image enhancement operations, and proceed to step 305. Step 305: Assign corresponding weights to the amplitude of natural texture regions, enhanced low-texture regions, or suppressed highlight reflection regions; slide the local window and return to step 302 until the local window has traversed the entire image; output the texture feature weight map. .

4. The non-contact measurement method for bridge hydrodynamic loads based on visual inversion according to claim 3, characterized in that, Gray-level variance of pixels within the local window in step 302 The calculation formula is: ； In the formula, Indicates the pixel position within a local window pixel grayscale values, This represents the average grayscale value of all pixels within a local window. This represents the total number of pixels within the local window; Spectral Energy The calculation is based on performing a two-dimensional discrete Fourier transform on the image patches within a local window to obtain the spectrum. ; then based on Calculate the spectral energy within the local window The calculation formula is: ； In the formula, Represents spatial frequency variables. Indicates the center of the spectrum. This represents the minimum frequency radius of a predefined spectrum. Indicates an indicator function; Brightness gradient amplitude The calculation formula is: ； In the formula, Represents the image brightness function. This represents the rate of change of image brightness in the x-direction. This represents the rate of change of image brightness in the y-direction. This represents the partial derivative.

5. The non-contact measurement method for bridge hydrodynamic loads based on visual inversion according to claim 4, characterized in that, The expression for the multi-frame joint optimization model in step 4 is: ； In the formula, This represents the velocity components of the fluid surface to be solved, and the velocity field of the fluid surface obtained by inversion. , Indicates the time consistency constraint coefficient, Indicates the spatial smoothing constraint coefficient; The brightness function of the image. Indicates image brightness Spatial gradient, ; Let represent the gradient of the velocity component field in the y-direction. ,in, Represents the velocity component field The rate of change in the x-direction, Represents the velocity component field The rate of change in the y-direction; Let represent the gradient of the velocity component field in the x-direction. ,in, Represents the velocity component field The rate of change in the x-direction, Represents the velocity component field Rate of change in the y-direction.

6. The non-contact measurement method for bridge hydrodynamic loads based on visual inversion according to claim 5, characterized in that, Step 5 specifically includes: Step 501: Calculate adjacency weights based on spatial distance, flow direction similarity, and velocity amplitude differences. The calculation formula is: ； In the formula, Represents two vector points in the velocity field. yes The field point, , Representing vector points respectively Spatial location coordinates, Represents vector points The Euclidean distance between them Represents the radius of a spatial domain. Indicates directional tolerance. Indicates the velocity amplitude scale, Representing vector points respectively The velocity vector at that point Represents velocity vector The angle between them Represents velocity vector The Euclidean norm of the difference; Step 502, based on adjacency weight Solve the energy minimization problem to obtain the reconstructed velocity field. The calculation formula is: ； In the formula, This represents the regularization weight coefficient. This represents a robust function used for soft suppression of outliers; Step 503: Calculate the degree of offset in the local velocity distribution using the optimal transmission distribution alignment theory. The calculation formula is: ； In the formula, This represents the velocity distribution within the current vector point's neighborhood. Indicates a robust reference distribution. Represent the Wasserstein-1 distance; if satisfying If so, then the vector point is an outlier, and for the vector point... After performing confidence attenuation correction, the filtered velocity field is obtained. : ； ； In the formula, This represents the dynamic attenuation factor based on optimal transmission distribution alignment. Represents the weighted average of the regions. This indicates a preset threshold.

7. The non-contact measurement method for bridge hydrodynamic loads based on visual inversion according to claim 6, characterized in that, In step 6, the filtered velocity field is input into the hydrodynamic pressure equation, and the calculation formula for the hydrodynamic pressure distribution on the pier surface is obtained as follows: ； In the formula, Indicates fluid density, The velocity field after filtering Normal velocity scalar field mapped onto the surface of the bridge pier.

8. The non-contact measurement method for bridge hydrodynamic loads based on visual inversion according to claim 7, characterized in that, In step 6, based on the geometric model of the bridge pier, the dynamic pressure distribution on the flow-receiving surface is integrated to obtain the instantaneous hydrodynamic load acting on the bridge pier. The calculation formula is as follows: ； In the formula, This represents the area of ​​the flow-receiving surface of the bridge pier. Represents the differential symbol.

9. The non-contact measurement method for bridge hydrodynamic loads based on visual inversion according to claim 8, characterized in that, The formula for calculating the self-supervised loss function constructed in step 7 is as follows: ； In the formula, , , Indicates the weighting coefficient. , Indicates the first Frame and the Dynamic pressure distribution in a frame image; Represents the divergence of the velocity field. This indicates the load predicted by the finite element method.

Images