Ancient bridge deflection non-contact measurement method based on binocular vision

Abstract

The invention discloses an ancient bridge deflection non-contact measurement method based on binocular vision. In order to solve the problem that the ancient bridge is made of a special material and cannot adopt a contact sensor due to the limitation of cultural relic protection, a high-contrast artificial vision mark point is arranged at a to-be-detected point, and laser irradiation is assisted to enhance image features; a binocular camera system is accurately calibrated through a Zhang Zhengyou calibration method, and the geometric accuracy is improved by combining distortion removal and epipolar correction; realizing three-dimensional reconstruction of the mark points by using a block matching algorithm, and extracting vertical displacement as an initial deflection value; and a dial indicator is synchronously introduced for calibration. Experiments show that after error calibration, the average relative error of the final deflection measurement result relative to the reference value of the dial indicator is stably controlled within 7%, the engineering precision requirements of safety preliminary evaluation and health monitoring of the ancient bridge structure are met, real non-destructive, low-cost and high-reliability ancient bridge deflection measurement is achieved, and the method is suitable for popularization and application. And a feasible technical path is provided for cultural relic bridge health monitoring.

Description

Technical Field

[0001] This invention belongs to the intelligent manufacturing equipment industry and relates to intelligent detection devices and methods, specifically to a non-contact measurement method for the deflection of ancient bridges based on binocular vision. Background Technology

[0002] Bridge deflection is a key indicator reflecting the stress state and safety performance of a bridge structure, and is widely used in load-bearing capacity assessment and structural health monitoring. Existing deflection measurement methods can be broadly classified into two categories: contact methods and non-contact methods.

[0003] For ordinary bridges, contact methods (such as displacement gauges, LVDTs, and strain gauges) are widely used in engineering practice due to their high accuracy and stability; non-contact methods (such as total stations, laser rangefinders, and photogrammetry) also provide effective supplements in specific scenarios. However, for ancient bridges, which are cultural relics with significant historical and cultural value, their structural materials are mostly stone, wood, or traditional masonry, and they are subject to strict regulations on cultural relic protection, prohibiting any form of drilling, pasting, bracket fixing, or other operations that may cause physical damage. This rigid restriction makes most contact measurement methods impractical in the context of ancient bridges.

[0004] Meanwhile, while some non-contact methods can theoretically avoid physical contact, they still face significant challenges in practical applications on ancient bridges. For example, some laser ranging or target-based photogrammetry schemes still require the installation of reflective targets or auxiliary devices on the bridge surface, which essentially constitutes an intervention in the cultural relic itself; high-precision equipment (such as total stations and 3D laser scanners) is expensive and complex to operate, making it difficult to widely apply in ancient bridge conservation projects with limited resources.

[0005] Against this backdrop, computer vision-based non-contact measurement technology is considered a potential solution due to its advantages such as low equipment cost, flexible deployment, and complete non-destructive nature. However, existing visual measurement methods face triple constraints in terms of accuracy, stability, and engineering reliability in the context of ancient bridges, specifically manifested as follows:

[0006] (1) Monocular vision methods are difficult to meet the requirements for three-dimensional deflection measurement of ancient bridges.

[0007] Monocular vision typically estimates structural displacement through pixel displacement or two-dimensional feature tracking, but it cannot directly obtain depth information of the measured point. However, the deflection assessment of ancient bridges relies on accurate vertical (Z-axis) displacement components. Due to the lack of depth constraints, the measurement results of monocular methods are easily affected by small changes in camera pose, fluctuations in shooting distance, and uncertainties in image scale. In complex on-site environments, it is difficult to guarantee the reliability of three-dimensional displacement calculations, and it is difficult to support quantitative safety assessments.

[0008] (2) There are systemic technical shortcomings in the application of binocular vision methods in ancient bridges.

[0009] Although binocular vision can reconstruct 3D coordinates through the principle of parallax, existing technical solutions still have the following shortcomings in the context of ancient bridges:

[0010] First, the accuracy of 3D reconstruction is significantly affected by the calibration and matching process: some methods do not fully perform high-precision camera calibration (such as not effectively correcting lens distortion), or use unoptimized stereo matching algorithms, which leads to large errors and poor repeatability in 3D coordinate reconstruction under typical ancient bridge site conditions such as long distance and low light.

[0011] Second, it relies too much on natural textures and lacks robustness: the surface of ancient bridges often shows features such as missing textures (e.g., smooth stone surface), repeated textures (e.g., regular masonry) or severe weathering. In addition, the dynamic changes in on-site lighting conditions (day and night, sunny and cloudy, shadows) make the matching process based on natural features prone to feature point extraction failure, mismatch or tracking loss, making it difficult to guarantee measurement continuity and stability.

[0012] Third, there is a lack of synchronous verification mechanism with high-precision benchmarks: most studies only conduct offline image analysis or compare with theoretical values, without introducing contact high-precision sensors for synchronous, same-location field measurement comparison and systematic error calibration. This results in the absolute accuracy of visual measurement results being unable to be effectively calibrated, insufficient engineering credibility, and difficulty in being adopted by ancient bridge safety assessment and cultural relic protection practices.

[0013] In summary, there is an urgent need for a truly non-destructive deflection measurement method that adapts to the material and structural characteristics of ancient bridges, possesses stable three-dimensional measurement capabilities, and can be verified through engineering-grade benchmarks, in order to fill the current technological gap in the field of health monitoring of cultural relic bridges. Summary of the Invention

[0014] To address the technical problems of existing technologies, such as the damage to the artifacts caused by contact measurement methods and the insufficient accuracy and lack of depth information in non-contact measurements, this invention provides a non-contact measurement method for the deflection of ancient bridges based on binocular vision. By deploying a binocular vision system, highly recognizable artificial markers are used as measurement targets without contact with the bridge itself. Systematic camera calibration and image correction ensure geometric accuracy, and a stereo matching algorithm reconstructs the three-dimensional coordinates of the measurement points. Deflection is calculated by comparing the coordinate changes under reference and loaded conditions. The key innovation lies in the simultaneous introduction of a high-precision dial gauge as a benchmark to calibrate and verify the visual measurement results, thus forming a complete measurement scheme suitable for ancient bridge artifact scenarios, combining the advantages of non-contact measurement with engineering-grade reliability.

[0015] The technical solution adopted in this invention is as follows:

[0016] A non-contact measurement method for the deflection of ancient bridges based on binocular vision includes the following steps:

[0017] S1. System setup and measurement point preparation: A binocular synchronous camera is set up in front of the area to be measured on the ancient bridge. High-contrast artificial visual markers are set at key measurement points on the ancient bridge. The artificial visual markers are illuminated by a laser light source to enhance their image recognizability.

[0018] S2. Basic calibration: Perform geometric calibration on the binocular synchronous camera to obtain the intrinsic parameter matrix of the left and right cameras, the lens distortion coefficient, and the rotation matrix and translation vector between the two cameras. Use the average reprojection error as the calibration quality criterion to verify the calibration results.

[0019] S3. Process calibration: Using the calibration parameters obtained in step S2, the left and right image pairs in the acquired reference state and loading state are subjected to distortion removal and epipolar correction to obtain the corrected image pairs and determine the disparity search range for stereo matching.

[0020] S4. Spatial calibration: In the calibrated image pair, the block matching algorithm is used to perform stereo matching on the artificial visual marker points, generate a disparity map, optimize and judge its quality, and reconstruct the three-dimensional spatial coordinates of the marker point in the reference state and the loading state based on the optimized disparity map.

[0021] S5. Result Calibration: Based on the three-dimensional spatial coordinates under the reference state and the loading state, calculate the difference to obtain the visually measured deflection value; synchronously install dial gauges near key measurement points to obtain the dial gauge reference deflection value under the same working condition; establish an error calibration model based on the correspondence between the visually measured deflection value and the dial gauge reference deflection value under different loading conditions; use the error calibration model to correct the visually measured deflection value; and output the final calibrated deflection measurement result.

[0022] Furthermore, in step S1, the binocular synchronous camera uses a USB interface camera module, with its left and right view imaging units integrated in the same module to achieve hardware synchronous acquisition of left and right images; the high-contrast artificial visual markers are circular or square-shaped reflective patches, black and white dots or checkerboard-style markers, or light spots formed by laser pointer projection, and their color or brightness forms a significant contrast with the background of the ancient bridge; the laser light source power is 3-8mW (too low a power will not enhance the effect, and too high a power will easily cause thermal damage to the stone or wood of the ancient bridge, violating the principle of cultural relic protection), and is only used to enhance the grayscale contrast of the markers, does not participate in the distance measurement calculation, and the laser irradiation position is consistent with the geometric center of the marker.

[0023] Furthermore, step S2 specifically includes:

[0024] S21. Using Zhang Zhengyou's planar calibration method, Q12-200-15 type glass substrate checkerboard is used as the calibration plate. At least 30 sets of synchronized images from the left and right cameras are collected from multiple angles within the common field of view of the binocular cameras.

[0025] S22. Solve for the intrinsic parameter matrices of the left and right cameras using the MATLAB calibration toolbox. , Distortion parameters and rotation matrix between the two cameras Translation vector ;

[0026] Among them, the intrinsic parameter matrix of the left camera Right camera intrinsic parameter matrix All follow the following intrinsic parameter matrix:

[0027] ,

[0028] In the formula, , These are the focal lengths along the x and y axes, respectively, in pixels; , () are the coordinates of the main point;

[0029] S23. Establish a unified reference system with the left camera coordinate system as the world coordinate system. The origin is located at the optical center of the left camera, the Z-axis is along its optical axis, and the X and Y axes are parallel to the imaging plane; spatial points (The homogeneous coordinates of the spatial point in the world coordinate system) are expressed in the left and right camera coordinate systems as follows:

[0030] ,

[0031] ,

[0032] in, , These represent the three-dimensional coordinates of the same point in the left and right camera coordinate systems, respectively. and These are the rotation matrix and translation vector from the world coordinate system to the left camera coordinate system, respectively. and These are the rotation matrix and translation vector from the world coordinate system to the right camera coordinate system, respectively.

[0033] Eliminating the world coordinate system yields the relative extrinsic parameters of the left and right cameras:

[0034] ,

[0035] in, ,

[0036] The rotation matrix is ​​in the form of:

[0037] ,

[0038] in, For elements of a rotation matrix, satisfying ;

[0039] S24. Using the average reprojection error as the calibration quality criterion, when the average reprojection error is controlled within the range of 0.09 to 0.12 pixels, the calibration result is considered to meet the accuracy requirements of subsequent 3D reconstruction; if the error exceeds this range, the calibration image is reacquired and the calibration process is repeated.

[0040] Further, step S3 includes:

[0041] S31. Collect left and right image pairs of the ancient bridge in the baseline state and the loading state respectively. Use the calibration parameters obtained in step S2 to perform distortion correction and stereo correction on the left and right image pairs so that the corrected images meet the epipolar constraint.

[0042] S32. Check the correction effect by generating an epipolar plot: if the lines connecting the corresponding feature points are strictly horizontally distributed, the stereo correction is considered to be qualified; otherwise, step S2 needs to be repeated.

[0043] S33. Based on the corrected image, the BM block matching algorithm is used to generate an initial disparity map, and the disparity search range is determined through multiple trials. The principle for determining the disparity search range is to cover the disparity values ​​of all marked points while avoiding the introduction of obvious noise or invalid matching areas.

[0044] Furthermore, in step S4, disparity map optimization and quality assessment include:

[0045] S41. If the initial disparity map contains noise, holes, or breaks, the stereo matching parameters are optimized. Optimization measures include adjusting the disparity search range, adjusting the matching block size, and filtering and hole filling of the disparity map.

[0046] S42. After optimization, the following criteria shall be used to judge whether the disparity map quality is qualified: the disparity of the marked point area is continuous and the boundary is clear; the disparity change trend of the same marked point is consistent in the reference state and the loading state;

[0047] S43. Proceed to the 3D reconstruction step only if the disparity map meets the qualification criteria of step S42.

[0048] Furthermore, in step S4, the parallax 3D reconstruction specifically involves:

[0049] Based on the optimized disparity map, and utilizing the principle of binocular triangulation, the depth coordinates Z of the target point in the camera coordinate system are calculated according to the disparity value d, the camera focal length f (f is a virtual uniform focal length generated by the algorithm during calibration), and the binocular baseline distance B.

[0050] ,

[0051] in, , This represents the horizontal pixel coordinates of a spatial point in the left-corrected image. This represents the horizontal pixel coordinates of a spatial point in the right-corrected image.

[0052] Combine the pixel coordinates of the target point in the left camera image ( , ) and camera principal point coordinates ( , Calculate the plane coordinates X and Y:

[0053] ,

[0054] ,

[0055] This allows us to obtain the three-dimensional spatial coordinates (X, Y, Z) of the target point under both the baseline and loading states.

[0056] Further, step S5 includes:

[0057] S51. In the calibration images of the baseline state and the loaded state, extract the pixel coordinates of the same artificial visual marker point, and obtain the corresponding three-dimensional spatial coordinates based on the three-dimensional coordinate point cloud obtained from the three-dimensional reconstruction. The three-dimensional spatial coordinates of the baseline state are: The three-dimensional spatial coordinates of the loading state are Calculate the visually measured deflection value using the following formula. :

[0058] ;

[0059] S52. Simultaneously install dial indicators near the key measuring points, with the dial indicator probe in direct contact with the ancient bridge structure. The deviation of the measuring position from the artificial visual marker point in the vertical projection direction should not exceed 3 mm to obtain the dial indicator reference deflection value under the same loading condition. ;

[0060] S53. Repeat the test under various loading conditions and calculate the visually measured deflection value for each test. Compared with the reference deflection value of the dial gauge The absolute and relative errors between them;

[0061] S54. Establish an error calibration model based on comparative data under different loading conditions: If a stable proportional deviation or fixed offset exists, then establish a linear regression model. or proportional correction model The model parameters a, b, or k are determined using the least squares method to correct subsequent purely visual measurement results, thus obtaining the calibrated deflection value. .

[0062] Furthermore, in step S53, the different loading conditions include two loading levels, namely 414g and 614g, and repeated tests are conducted under each loading level; by statistically analyzing the distribution range and stability of visual measurement errors, the reliability of the visual measurement deflection value within the acceptable accuracy range for engineering is verified.

[0063] Furthermore, the baseline state is the unloaded or initially static state of the ancient bridge, while the loading state is the state of the ancient bridge under test load, vehicle traffic, or changes in ambient temperature.

[0064] Furthermore, the baseline distance of the binocular synchronous camera is selected based on the span of the ancient bridge and the measurement distance, with a working distance of 3 to 15 meters and a camera resolution of no less than 1920×1080 pixels. After error calibration, the average relative error of the final deflection measurement result relative to the dial gauge reference value is stably controlled within 7% (see Table 1 and Table 2), and the relative error of most measurement points does not exceed 10%, which meets the engineering accuracy requirements for the preliminary assessment of the structural safety and health monitoring of the ancient bridge.

[0065] Compared with the prior art, the present invention has the following technical effects:

[0066] (1) It realizes non-destructive measurement of ancient bridge cultural relics: the entire measurement process does not require the installation of any sensors or any physical contact on the ancient bridge structure, completely avoiding destructive operations such as drilling and pasting, and fully meeting the highest requirements for cultural relic protection.

[0067] (2) Provides true three-dimensional displacement measurement capability: Based on the principle of binocular stereo vision, it directly obtains the three-dimensional spatial coordinate changes of the measurement point. The calculation results are not affected by the shooting angle and scale scaling, and are more accurate and reliable than the monocular vision method.

[0068] (3) Significantly improved measurement robustness in complex environments: By using actively set marker points, the difficulties in feature extraction and matching caused by missing textures, weathering, and uneven lighting on the surface of ancient bridges were overcome, ensuring the stability and repeatability of the measurement process.

[0069] (4) An engineering-based reliability verification system was established: By introducing a dial gauge as a benchmark, a fusion measurement mode of non-contact main measurement and contact benchmark verification was formed. This mode not only provides a means of accuracy calibration, but more importantly, it provides a convincing and traceable engineering verification basis for the measurement results, greatly improving the acceptance and authority of the method in engineering practice. Experimental results show that after calibration by the method of this invention, the average relative error of each working condition decreased from 6.79%~6.85% before calibration to 3.98%~6.12% (see Table 1 and Table 2), and the measurement accuracy and reliability were effectively improved.

[0070] (5) It has good applicability and economy: The core of the system is a common binocular camera and computing equipment, and the cost is far lower than that of high-end equipment such as laser scanning. The method of this invention has a clear process and can be extended to the deformation monitoring of ancient bridges of different shapes and materials, and even other precious structures that are not allowed to be touched, and has broad prospects for promotion and application. Attached Figure Description

[0071] Figure 1 This is a schematic diagram of the overall process of the method of the present invention.

[0072] Figure 2 This is a schematic diagram of binocular camera calibration and epipolar correction.

[0073] Figure 3 This is a schematic diagram of measuring point marking and laser-assisted identification.

[0074] Figure 4 This is a schematic diagram of the stereo matching and 3D coordinate reconstruction process based on binocular vision.

[0075] Figure 5 This is a comparison chart of the average relative error before and after calibration under different loading conditions. Detailed Implementation

[0076] The present invention will be further described in detail below with reference to specific embodiments, but the present invention is not limited thereto.

[0077] This invention presents a non-contact measurement method for the deflection of ancient bridges based on binocular vision measurement. Using a binocular camera as the core measuring device, it measures the displacement changes of key measuring points on an ancient bridge under both unloaded and loaded conditions through image acquisition, camera calibration, stereo matching, and 3D reconstruction. The method then calculates the bridge deflection. This invention eliminates the need to attach sensors to the surface of the ancient bridge structure or drill holes for installation, thus avoiding damage to the bridge's structural integrity. It is suitable for scenarios involving the safety assessment and structural health monitoring of ancient bridges.

[0078] Example 1

[0079] A schematic diagram of the overall process of the binocular vision-based non-contact measurement method for the deflection of ancient bridges in this invention is shown below. Figure 1 As shown, the specific steps include the following:

[0080] S1. System Setup and Measurement Point Preparation

[0081] like Figure 3 As shown, a binocular synchronous camera was stably erected in front of the area to be measured on the ancient bridge. The camera orientation was adjusted so that the overlapping area of ​​the left and right camera fields of view covered key measurement points in the middle of the bridge span. High-contrast artificial visual markers were set at the measurement points. In this embodiment, black and white circular reflective patches were used, whose colors formed a significant contrast with the background of the ancient bridge. A low-power laser pointer (5 mW) was used to illuminate the center of the marker, ensuring that the marker appeared as a clear and stable bright area in both left and right images. The laser was only used to enhance the grayscale contrast of the marker and did not participate in the distance measurement calculation. The illumination position was consistent with the geometric center of the marker.

[0082] S2. Basic Calibration: Camera Geometric Calibration

[0083] Before the formal measurement, the binocular synchronous camera was first geometrically calibrated.

[0084] The Zhang Zhengyou planar calibration method was adopted, and a Q12-200-15 type glass substrate checkerboard was used as the calibration plate. At least 30 sets of synchronous images from multiple angles were acquired within the common field of view of the binocular cameras.

[0085] Solving the intrinsic parameter matrices of the left and right cameras using the MATLAB Calibration Toolbox , Distortion parameters and rotation matrix between the two cameras Translation vector ;

[0086] Among them, the intrinsic parameter matrix of the left camera Right camera intrinsic parameter matrix All follow the following intrinsic parameter matrix:

[0087] ,

[0088] In the formula, , These are the focal lengths along the x and y axes, respectively, in pixels; , () are the coordinates of the main point;

[0089] S23. Establish a unified reference system with the left camera coordinate system as the world coordinate system. The origin is located at the optical center of the left camera, the Z-axis is along its optical axis, and the X and Y axes are parallel to the imaging plane; spatial points (The homogeneous coordinates of the spatial point in the world coordinate system) are expressed in the left and right camera coordinate systems as follows:

[0090] ,

[0091] ,

[0092] in, , These represent the three-dimensional coordinates of the same point in the left and right camera coordinate systems, respectively. and These are the rotation matrix and translation vector from the world coordinate system to the left camera coordinate system, respectively. and These are the rotation matrix and translation vector from the world coordinate system to the right camera coordinate system, respectively.

[0093] Eliminating the world coordinate system yields the relative extrinsic parameters of the left and right cameras:

[0094] ,

[0095] in, ,

[0096] The rotation matrix is ​​in the form of:

[0097] ,

[0098] in, For elements of a rotation matrix, satisfying ;

[0099] The average reprojection error is used as the calibration quality criterion. In this embodiment, the calculated average reprojection error is 0.11 pixels, which falls within the preferred range of 0.09 to 0.12 pixels, meeting the accuracy requirements for subsequent 3D reconstruction. If the error exceeds this range, the calibration image is reacquired and the calibration process is repeated.

[0100] S3. Process Calibration: Polar Correction and Parallax Range Determination

[0101] Left and right image pairs of the ancient bridge were acquired under both baseline (unloaded) and loaded conditions. Distortion and stereo corrections were performed on each image pair using the calibration parameters obtained from S2, ensuring the corrected images met epipolar constraints.

[0102] The correction effect is checked by generating an epipolar plot: several corresponding feature points are selected on the left and right corrected images. If the lines connecting them are strictly horizontal, the stereo correction is qualified; otherwise, step S2 needs to be repeated. In this embodiment, after correction, all feature point lines remain horizontal. The schematic diagram of binocular camera calibration and epipolar correction is shown below. Figure 2 As shown.

[0103] Based on the corrected image, an initial disparity map is generated using the BM block matching algorithm, and a reasonable disparity search range is determined through multiple trials. The optimal disparity search range determined in this embodiment is [d]. min ,d max [50, 90] pixels, this range covers the disparity values ​​of all marker points while avoiding the introduction of obvious noise or invalid matching areas.

[0104] S4. Spatial Calibration: Quality Control of 3D Reconstruction and Disparity Map

[0105] Within a defined disparity range, BM stereo matching is performed on the corrected images of the baseline and loaded states to generate an initial disparity map. If the disparity map contains noise, holes, or breaks, optimization is performed: the disparity search range is adjusted, the matching block size is adjusted, and mean filtering and hole filling are applied.

[0106] After optimization, the disparity map quality is judged according to the following criteria: the disparity of the marked point area is continuous and the boundaries are clear; the disparity change trend of the same marked point is consistent under both unloaded and loaded states. The disparity map in this embodiment meets the criteria and proceeds to the next step of 3D reconstruction.

[0107] A schematic diagram of the stereo matching and 3D reconstruction process is shown below. Figure 4 As shown. The 3D reconstruction specifically involves: based on the optimized disparity map, utilizing the principle of binocular triangulation, calculating the depth coordinates Z of the target point in the camera coordinate system according to the disparity value d, the camera focal length f (f is a virtual unified focal length generated by the algorithm during calibration), and the binocular baseline distance B.

[0108] ,

[0109] in, , This represents the horizontal pixel coordinates of a spatial point in the left-corrected image. This represents the horizontal pixel coordinates of a spatial point in the right-corrected image.

[0110] Combine the pixel coordinates of the target point in the left camera image ( , ) and camera principal point coordinates ( , Calculate the plane coordinates X and Y:

[0111] ,

[0112] ,

[0113] This allows us to obtain the three-dimensional spatial coordinates (X, Y, Z) of the target point under both the baseline and loading states.

[0114] In the corrected images of the baseline and loaded states, the pixel coordinates of the same artificial visual marker point are extracted, and the corresponding three-dimensional spatial coordinates are obtained from the three-dimensional coordinate point cloud obtained by 3D reconstruction. The three-dimensional spatial coordinates of the baseline state are: The three-dimensional spatial coordinates of the loading state are Calculate the visually measured deflection value using the following formula. : ;

[0115] S5. Result Calibration: Error Modeling and Correction Based on Dial Gauge

[0116] Dial gauges were installed synchronously near the same measuring points, with the dial gauge probes in direct contact with the ancient bridge structure. The deviation of the measuring position from the manually marked point in the vertical projection direction did not exceed 3mm, thus obtaining the dial gauge reference deflection value under the same loading condition. ;

[0117] S53. Repeat the test under various loading conditions and calculate the visually measured deflection value for each test. Compared with the reference deflection value of the dial gauge The absolute and relative errors between them;

[0118] S54. Establish an error calibration model based on comparative data under various loading conditions: If a stable proportional deviation or fixed offset exists, then establish a linear regression model. or proportional correction model The model parameters a, b, or k are determined using the least squares method to correct subsequent purely visual measurement results, thus obtaining the calibrated deflection value. .

[0119] The calculated values ​​are a = 0.956 and b = 0.089. A proportional correction model can also be used. In this embodiment, a linear model is selected based on goodness of fit. When establishing the calibration model, this embodiment simultaneously attempted linear regression, quadratic polynomial regression, and proportional models. The goodness of fit R0 of each model was compared. 2 And the residual distribution, revealing the R-value of the linear regression model. 2 R of the quadratic polynomial model 2 The differences between the two are not significant, and the coefficient of the quadratic term is close to zero. Considering the simplicity of the linear model, the clear physical meaning of the parameters, and the low risk of overfitting, this embodiment selects the linear regression model as the final calibration model. In practical applications, the most suitable model form can be selected according to the data characteristics, which does not affect the creativity of the core method of "establishing a calibration model based on a dial gauge benchmark" proposed in this invention.

[0120] The model was used to correct all visual measurement results to obtain calibrated deflection values. The comparison of some data before and after calibration is shown in Tables 1 and 2. The data are derived from the statistical averages of Tables 1 and 2. It can be seen that after correction by the calibration model of this invention, the average relative error of each operating condition has decreased significantly.

[0121] Table 1 Comparison of Deflection Measurement Results and Calibration Results under 414g Working Condition

[0122]

[0123] Table 2 Comparison of Deflection Measurement Results and Calibration Results under 614g Working Condition

[0124]

[0125] The average relative errors before and after calibration for each of the above operating conditions are plotted as follows: Figure 5 The improvement in error can be seen intuitively. Figure 5 The black dotted line represents the average relative error before calibration, and the red dotted line represents the average relative error after calibration. (From Tables 1 and 2...) Figure 5 It can be seen that the relative error between the visual measurement results before calibration and the dial gauge reference value averages approximately 6.8% to 6.9% under various operating conditions, with some data points approaching 10%, and exhibiting certain fluctuations. The error calibration model established by this invention... After correction, the average relative error for each operating condition decreased significantly: from 6.79% to 3.98% for the 414g condition and from 6.85% to 6.12% for the 614g condition. Overall, the mean relative error after calibration was stably controlled within 7%, a decrease of approximately 0.7 to 2.9 percentage points compared to before calibration, effectively improving measurement accuracy and stability.

[0126] Error Source Analysis: Before calibration, the errors mainly came from: (1) pixel quantization error of the binocular camera and residual lens distortion; (2) interpolation error in the stereo matching process; and (3) the influence of ambient light changes on the imaging of marker points. The above errors have both systematic and random characteristics. This invention introduces dial gauge reference measurement and establishes a multi-condition statistical calibration model, which effectively compensates for systematic deviations and makes the final results meet the engineering accuracy requirements for ancient bridge deflection measurement.

[0127] Calibration Results: It should be noted that this invention uses a standard industrial-grade binocular camera to measure the deflection of ancient bridges under strict non-destructive testing conditions. After calibration, the average relative error stabilized within 7%, with slightly higher errors at individual data points, mainly due to random errors (such as fluctuations in illumination and differences in the imaging quality of marker points). However, from an overall statistical perspective, the calibration model significantly improved the systematic bias of the measurement results, enhanced the consistency and reliability of the data, and can meet the engineering requirements for health monitoring and safety assessment of ancient bridge structures.

[0128] In summary, this embodiment fully verifies the feasibility and superiority of the non-contact measurement method for ancient bridge deflection based on binocular vision proposed in this invention. Under the premise of ensuring that the ancient bridge is undamaged, it achieves deflection measurement with controllable accuracy and high reliability.

[0129] Comparative Example 1

[0130] The same ancient bridge model or actual bridge measurement points as in Embodiment 1 of this invention are selected, and the same binocular camera system, calibration process (Zhang Zhengyou method), and image correction steps are used. The difference is that no artificial visual markers are set, and no laser illumination is used. Feature extraction (e.g., using SIFT or SURF algorithms) and stereo matching are performed directly on the natural textures of the ancient bridge surface (such as stone seams, brick seams, and wood grain).

[0131] Data acquisition: Simultaneously acquire image pairs of the bridge in both baseline and loaded states.

[0132] Results processing: Attempt to reconstruct the 3D coordinates of the measurement points and calculate the deflection. Record the matching success rate (number of successfully matched image frames / total number of frames) and the calculated deflection value.

[0133] Experimental results show that in areas with relatively good lighting conditions and clear surface textures, the binocular matching method based on natural textures can only obtain matching results in some image frames, with a low overall matching success rate, and the matching results show large fluctuations with frame changes.

[0134] In areas with poor lighting, shadow coverage, or excessively smooth or weathered surfaces, feature point extraction fails or mismatches occur at extremely high rates, resulting in most image frames failing to complete effective matching. This makes it difficult to perform deflection measurement continuously and stably.

[0135] Even in the few successfully matched frames, the accuracy of the 3D reconstruction is significantly lower than that of the artificially set high-contrast markers because the positions and scales of natural feature points are not fixed. This results in a significant increase in the error of the deflection calculation results, making it difficult to meet the requirements of engineering measurement in most cases.

[0136] The above comparison results show that in structural scenarios such as ancient bridges with complex surface textures and high degrees of weathering, it is difficult to achieve stable and reliable feature matching and 3D reconstruction by directly using general binocular vision methods that rely on natural texture features, thus failing to meet the accuracy and continuity requirements of deflection measurement. Therefore, this invention effectively overcomes the matching difficulties caused by the instability of the natural surface texture of ancient bridges by deploying high-contrast artificial visual markers in the measurement area, which is a key technical feature for achieving high-precision and stable deflection measurement.

[0137] Comparative Example 2

[0138] Using the same ancient bridge model or actual bridge measurement point locations as in Embodiment 1 of this invention, Comparative Example 2 employs the traditional contact deflection measurement method commonly used in engineering practice. This traditional method measures the deflection changes of the bridge under reference and loading conditions by deploying contact displacement measuring devices (such as mechanical dial gauges or electrical displacement gauges) at key measurement points on the bridge and using rigid supports or external reference structures as references.

[0139] During implementation, support devices need to be erected on or near the bridge structure to fix the measuring instruments, and the measuring points need to be manually installed and leveled. Before measurement, the instruments need to be zero-point calibrated, and displacement changes are read in real time during loading to obtain bridge deflection data.

[0140] This traditional contact measurement method can obtain high-precision single-point deflection measurement results, but it has the following problems in practical applications:

[0141] (1) The measuring device needs to be in direct contact with the bridge structure, the installation process is complicated, and it may cause potential disturbance or risk to the ancient bridge and other cultural relics protection structures.

[0142] (2) The number of measuring points is limited, making it difficult to achieve simultaneous measurement of deflection at multiple measuring points or across the entire field;

[0143] (3) The measurement preparation time is relatively long and it is highly dependent on the on-site conditions, making it unsuitable for long-term or frequent monitoring scenarios;

[0144] (4) In complex terrain or bridge environments with limited access, it is difficult to deploy instruments and build support structures.

[0145] Comparing the traditional contact deflection measurement method of Comparative Example 2 with the non-contact deflection measurement method based on binocular vision of Embodiment 1 of the present invention, it can be seen that there are significant differences between the two in terms of measurement method, engineering applicability and impact on structure.

[0146] In terms of installation and implementation, traditional contact measurement methods require the deployment of measurement devices and support structures on the bridge structure, which is time-consuming and highly dependent on site conditions. In contrast, the method of this invention adopts a non-contact binocular vision measurement method, which only requires the deployment of artificial visual markers in the measurement area and the completion of camera installation and calibration. It has high installation efficiency, does not have direct contact with the bridge structure, and is particularly suitable for scenarios such as ancient bridges where it is not appropriate to deploy sensors.

[0147] In terms of measurement range and scalability, traditional methods can usually only achieve deflection measurement at a limited number of measurement points, making it difficult to simultaneously acquire deflection information at multiple points or across the entire field; while the method of this invention can simultaneously acquire three-dimensional displacement information of multiple measurement points through image processing, and has good measurement point expansion capability.

[0148] Regarding the impact on the structure, traditional contact measurement may cause some disturbance to the bridge structure, which is especially unfavorable for the long-term monitoring of cultural relic protection structures; while the method of this invention uses high-contrast artificial visual markers and non-contact measurement methods, which have virtually no physical impact on the bridge structure.

[0149] In summary, Comparative Example 2 shows that although traditional contact measurement methods have certain advantages in terms of single-point accuracy, they have significant shortcomings in terms of installation convenience, measurement point expansion, and applicability to ancient bridge structures. In contrast, this invention, by introducing high-contrast artificial visual markers and binocular visual measurement technology, significantly improves engineering applicability and structural friendliness while ensuring measurement accuracy.

[0150] Comparative Example 3

[0151] Comparative Example 3 uses pure visual measurement without dial gauge calibration. It employs the exact same experimental equipment, measurement environment, and operating procedures as Example 1. Specifically, the model, resolution, baseline distance, and measurement distance of the binocular synchronous camera are the same as in Example 1. The artificial visual markers are also set at the same measurement points on the ancient bridge, and the marker type, size, and installation method are the same as in Example 1. Laser-assisted illumination uses a laser source of the same power to illuminate the center of the markers. Image acquisition and processing follow steps S1 to S4 of Example 1, including system setup, camera calibration, epipolar correction, disparity map optimization, and 3D reconstruction.

[0152] The only difference is that step S5 is not performed, meaning that dial gauges are not introduced for synchronous measurement, and no form of error calibration or model correction is performed. The deflection is calculated directly using the three-dimensional coordinates reconstructed in S4.

[0153] Comparative Example 3 was repeated under two loading conditions (414g and 614g), and the visually measured deflection values ​​were recorded. and the dial gauge reference deflection value (The dial gauge in this comparative example is only used for post-hoc verification.) The experimental results are statistically summarized below:

[0154] Table 3. Comparative Example 3: Visual Measurement Error Statistics for Various Working Conditions

[0155]

[0156] Table 3 shows that under both operating conditions, the visual measurement results generally tend to be smaller, meaning the measured values ​​are typically lower than the dial indicator's reference value. This indicates that visual measurements suffer from systematic errors caused by factors such as imaging principles, calibration residuals, and matching errors. The relative errors for each operating condition range from 3.65% to 9.54%, with some data points approaching 10%, and the error distribution is inconsistent across different operating conditions, exhibiting a certain degree of random fluctuation. Since dial indicator calibration was not introduced, these systematic biases and random fluctuations cannot be effectively compensated for, making it difficult to guarantee the absolute accuracy of the measurement results.

[0157] Analysis of the reasons for the smaller size: The main reasons for the overall smaller size of the visual measurement results are: (1) The stereo matching algorithm has sub-pixel interpolation error at the edge of the marker point, and tends to locate the matching point in the middle of the area where the gray level changes slowly, resulting in a slightly larger disparity value. According to the formula (1) Large parallax leads to a smaller calculated depth Z value, which in turn leads to a smaller deflection (depth change); (2) Although the residual lens distortion of the camera calibration has been corrected, there are still slight errors, resulting in a systematic shift in the 3D reconstruction coordinates; (3) Changes in ambient lighting cause asymmetry in the gray-scale distribution of the marker points, resulting in a slight shift in the matching center. The combined effect of these factors makes the visual measurement results generally smaller.

[0158] The experimental results of Comparative Example 3 and Example 1 (using the complete method of this invention) are compared as shown in Table 4:

[0159] Table 4 Comparison of results between Comparative Example 3 and Example 1

[0160]

[0161] Comparative analysis shows that while relying solely on a binocular vision system and high-precision calibration can achieve deflection measurement, it cannot eliminate systematic errors caused by imaging principles, hardware limitations, and environmental factors. This invention, by introducing a dial gauge reference measurement and establishing a multi-condition statistical calibration model, can effectively compensate for systematic deviations and significantly improve measurement accuracy and stability. The calibration results of Example 1 are significantly better than those of Comparative Example 3 in terms of average value, indicating that step S5 plays a core and crucial role in this invention.

[0162] It should be noted that after calibration in Example 1, there were still a few outliers, mainly due to random errors (such as sudden changes in illumination, uneven reflection on the surface of the markers). However, from a statistical perspective, the average error under each operating condition decreased significantly (by about 0.7 to 2.9 percentage points), and the standard deviation of the error distribution decreased, indicating that the calibration model effectively compensated for systematic biases and improved the overall measurement reliability.

[0163] The results of Comparative Example 3 show that, under the same hardware conditions and measurement environment, relying solely on visual measurement and geometric calibration cannot completely eliminate systematic errors, and the absolute accuracy of the measurement results is difficult to guarantee. The dial indicator result calibration step (S5) proposed in this invention is a key step in achieving stable and controllable accuracy, which can effectively improve the engineering reliability of the measurement method and has significant creative and practical value.

Claims

1. A non-contact measurement method for the deflection of ancient bridges based on binocular vision, characterized in that, Includes the following steps: S1. System setup and measurement point preparation: A binocular synchronous camera is set up in front of the area to be measured of the ancient bridge. High-contrast artificial visual markers are set at key measurement points of the ancient bridge. The artificial visual markers are illuminated by a laser light source to enhance their image recognizability. S2. Basic calibration: Perform geometric calibration on the binocular synchronous camera to obtain the intrinsic parameter matrix of the left and right cameras, the lens distortion coefficient, and the rotation matrix and translation vector between the two cameras. Use the average reprojection error as the calibration quality criterion to verify the calibration results. S3. Process calibration: Using the calibration parameters obtained in step S2, the left and right image pairs in the acquired reference state and loading state are subjected to distortion removal and epipolar correction to obtain the corrected image pairs and determine the disparity search range for stereo matching. S4. Spatial calibration: In the calibrated image pair, the block matching algorithm is used to perform stereo matching on the artificial visual marker points, generate a disparity map, optimize and judge its quality, and reconstruct the three-dimensional spatial coordinates of the marker point in the reference state and the loading state based on the optimized disparity map. S5. Result Calibration: Based on the three-dimensional spatial coordinates under the reference state and the loading state, calculate the difference to obtain the visually measured deflection value; synchronously install dial gauges near the key measuring points to obtain the dial gauge reference deflection value under the same working condition; establish an error calibration model based on the correspondence between the visually measured deflection value and the dial gauge reference deflection value under different loading conditions; use the error calibration model to correct the visually measured deflection value; and output the final calibrated deflection measurement result.

2. The non-contact measurement method for ancient bridge deflection based on binocular vision as described in claim 1, characterized in that, In step S1, the binocular synchronous camera uses a USB interface camera module, and its left and right view imaging units are integrated in the same module to achieve hardware synchronous acquisition of left and right images; the high-contrast artificial visual markers are circular or square-shaped reflective patches, black and white dots or checkerboard-shaped markers, or light spots formed by laser pointer projection, and their color or brightness is significantly contrasted with the background of the ancient bridge; the laser light source power is 3-8mW, which is only used to enhance the grayscale contrast of the markers and does not participate in the distance measurement calculation, and the laser illumination position is consistent with the geometric center of the markers.

3. The non-contact measurement method for ancient bridge deflection based on binocular vision as described in claim 1, characterized in that, Step S2 specifically includes: S21. Using Zhang Zhengyou's planar calibration method, Q12-200-15 type glass substrate checkerboard is used as the calibration plate. At least 30 sets of synchronized images from the left and right cameras are collected from multiple angles within the common field of view of the binocular cameras. S22. Solve for the intrinsic parameter matrices of the left and right cameras using the MATLAB calibration toolbox. , Distortion parameters and rotation matrix between the two cameras Translation vector ; Among them, the intrinsic parameter matrix of the left camera Right camera intrinsic parameter matrix All follow the following intrinsic parameter matrix: ， In the formula, , These are the focal lengths along the x and y axes, respectively, in pixels; , () are the coordinates of the main point; S23. Establish a unified reference system with the left camera coordinate system as the world coordinate system. The origin is located at the optical center of the left camera, the Z-axis is along its optical axis, and the X and Y axes are parallel to the imaging plane; spatial points In the left and right camera coordinate systems, this is expressed as follows: ， ， in, , These represent the three-dimensional coordinates of the same point in the left and right camera coordinate systems, respectively. and These are the rotation matrix and translation vector from the world coordinate system to the left camera coordinate system, respectively. and These are the rotation matrix and translation vector from the world coordinate system to the right camera coordinate system, respectively. Eliminating the world coordinate system yields the relative extrinsic parameters of the left and right cameras: ， in, , The rotation matrix is ​​in the form of: ， in, For elements of a rotation matrix, satisfying ; S24. Using the average reprojection error as the calibration quality criterion, when the average reprojection error is controlled within the range of 0.09 to 0.12 pixels, the calibration result is considered to meet the accuracy requirements of subsequent 3D reconstruction; if the error exceeds this range, the calibration image is reacquired and the calibration process is repeated.

4. The non-contact measurement method for ancient bridge deflection based on binocular vision according to claim 1, characterized in that, Step S3 includes: S31. Collect left and right image pairs of the ancient bridge in the baseline state and the loading state respectively. Use the calibration parameters obtained in step S2 to perform distortion correction and stereo correction on the left and right image pairs so that the corrected images meet the epipolar constraint. S32. Check the correction effect by generating an epipolar plot: if the lines connecting the corresponding feature points are strictly horizontally distributed, the stereo correction is considered to be qualified; otherwise, step S2 needs to be repeated. S33. Based on the corrected image, the BM block matching algorithm is used to generate an initial disparity map, and the disparity search range is determined through multiple trial calculations. The principle for determining the disparity search range is to cover the disparity values ​​of all marked points while avoiding the introduction of obvious noise or invalid matching areas.

5. The non-contact measurement method for ancient bridge deflection based on binocular vision according to claim 1, characterized in that, In step S4, disparity map optimization and quality assessment include: S41. If the initial disparity map has noise, holes or breaks, the stereo matching parameters are optimized. Optimization measures include adjusting the disparity search range, adjusting the matching block size, and filtering and hole filling of the disparity map. S42. After optimization, the following criteria shall be used to judge whether the disparity map quality is qualified: the disparity of the marked point area is continuous and the boundary is clear; the disparity change trend of the same marked point is consistent in the reference state and the loading state; S43. Proceed to the 3D reconstruction step only if the disparity map meets the qualification criteria of step S42.

6. The non-contact measurement method for ancient bridge deflection based on binocular vision according to claim 1, characterized in that, In step S4, the disparity map 3D reconstruction specifically involves: Based on the optimized disparity map, and using the principle of binocular triangulation, the depth coordinates Z of the target point in the camera coordinate system are calculated according to the disparity value d, the camera focal length f, and the binocular baseline distance B. ， in, , This represents the horizontal pixel coordinates of a spatial point in the left-corrected image. This represents the horizontal pixel coordinates of a spatial point in the right-corrected image. Combine the pixel coordinates of the target point in the left camera image ( , ) and camera principal point coordinates ( , Calculate the plane coordinates X and Y: ， ， This allows us to obtain the three-dimensional spatial coordinates (X, Y, Z) of the target point under both the baseline and loading states.

7. The non-contact measurement method for ancient bridge deflection based on binocular vision as described in claim 1, characterized in that, Step S5 includes: S51. In the calibration images of the baseline state and the loaded state, extract the pixel coordinates of the same artificial visual marker point, and obtain the corresponding three-dimensional spatial coordinates based on the three-dimensional coordinate point cloud obtained from the three-dimensional reconstruction. The three-dimensional spatial coordinates of the baseline state are: The three-dimensional spatial coordinates of the loading state are Calculate the visually measured deflection value using the following formula. : ； S52. Simultaneously install dial indicators near the key measuring points, with the dial indicator probe in direct contact with the ancient bridge structure. The deviation of the measuring position from the artificial visual marker point in the vertical projection direction should not exceed 3 mm to obtain the dial indicator reference deflection value under the same loading condition. ; S53. Repeat the test under various loading conditions and calculate the visually measured deflection value for each test. Compared with the reference deflection value of the dial gauge The absolute and relative errors between them; S54. Establish an error calibration model based on comparative data under different loading conditions: If a stable proportional deviation or fixed offset exists, then establish a linear regression model. or proportional correction model The model parameters a, b, or k are determined using the least squares method to correct subsequent purely visual measurement results, thus obtaining the calibrated deflection value. .

8. The non-contact measurement method for ancient bridge deflection based on binocular vision according to claim 7, characterized in that, In step S53, the different loading conditions include two loading levels, namely 414g and 614g, and repeated tests are conducted under each loading level; by statistically analyzing the distribution range and stability of visual measurement errors, the reliability of the visual measurement deflection value within the acceptable accuracy range for engineering is verified.

9. The non-contact measurement method for ancient bridge deflection based on binocular vision as described in any one of claims 1 to 8, characterized in that, The baseline state is the unloaded or initially stationary state of the ancient bridge, while the loading state is the state of the ancient bridge under test load, vehicle traffic, or changes in ambient temperature.

10. The non-contact measurement method for ancient bridge deflection based on binocular vision as described in any one of claims 1 to 8, characterized in that, The baseline distance of the binocular synchronous camera is selected based on the span of the ancient bridge and the measurement distance, with a working distance of 3 to 15 meters and a camera resolution of no less than 1920×1080 pixels.

Images