Arch bridge construction support construction linear control method based on machine vision and point cloud technology

Abstract

The invention relates to the technical field of bridge engineering construction monitoring, and particularly discloses an arch bridge construction support construction linear control method based on machine vision and a point cloud technology. The method comprises the following steps: arranging a laser-structured light integrated point cloud acquisition device and a ground control point in an arch bridge construction area, and acquiring original point clouds on the surfaces of a bracket and an arch rib in real time; filtering, registering and projecting the point cloud to generate a multi-view Hill shade grayscale image sequence; sub-pixel displacement is calculated through digital image correlation analysis, and a three-dimensional displacement vector is inverted; the displacement field is visualized through a three-dimensional graphic engine, and the rate and the acceleration are calculated in combination with time sequence data; a displacement-rate-acceleration-trend four-level threshold model is established, and Bayesian updating is fused to carry out safety early warning; and parameter adjustment instruction closed-loop control is automatically generated during overrun. According to the invention, non-contact, full-field, high-precision and automatic three-dimensional deformation monitoring and construction linear closed-loop control are realized, and the arch bridge construction safety and control precision are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of bridge construction monitoring, and in particular to a method for controlling the construction alignment of arch bridge construction supports by combining machine vision, digital image correlation analysis (DIC), and point cloud 3D reconstruction technology. This method is used to achieve non-contact 3D displacement monitoring and structural deformation early warning during the pouring of concrete arch ribs and changes in the stress on the supports, and belongs to the field of civil engineering construction control and intelligent detection technology. Background Technology

[0002] In traditional arch bridge construction, the supports and arch ribs undergo complex deformations during concrete pouring, load transfer, and formwork removal. If the displacement of the supports or arch ribs exceeds the limits, it may lead to structural alignment deviations or early cracks, or even cause support instability and collapse.

[0003] Currently, the commonly used monitoring methods in the industry mainly include:

[0004] 1. Total station measurement method: The three-dimensional coordinates of key points are observed by setting up a measuring prism, but it requires manual operation, has a low monitoring frequency, and is difficult to capture continuous dynamic processes.

[0005] 2. Strain gauge, displacement gauge and inclinometer method: Although it can obtain local fine data, it is cumbersome to install, has complicated cables, is greatly affected by construction and has limited coverage.

[0006] 3. Laser scanning or photogrammetry point cloud method: It can reconstruct the surface morphology of the structure, but it is usually only used for stage inspection and lacks the ability to perform time continuity and automated analysis.

[0007] Traditional methods generally suffer from the following technical problems:

[0008] 1. Low monitoring frequency and insufficient real-time performance: unable to capture the transient deformation characteristics of the arch ribs under varying casting loads;

[0009] 2. Sparse measuring points and discontinuous information: It is difficult to form a complete deformation field;

[0010] 3. High environmental interference and strong reliance on manual labor: Manual measurement and calibration are required, resulting in poor data consistency;

[0011] 4. Difficulty in solving three-dimensional displacement: often requires multiple registrations or DEM differences, the calculation is complex and errors are superimposed.

[0012] Therefore, there is an urgent need for a three-dimensional deformation monitoring and support construction control method that can achieve high spatiotemporal resolution, non-contact, and automated operation during the construction of arch bridges, so as to improve construction safety and accuracy control. Summary of the Invention

[0013] The purpose of this invention is to provide a construction alignment control method for arch bridge support structures based on machine vision and point cloud technology. By constructing a high-precision temporal point cloud and combining it with digital image correlation (DIC) analysis and multi-angle projection fusion, the three-dimensional displacement field of the arch bridge support and arch ribs can be calculated in real time, realizing automated monitoring, early warning, and construction control of structural deformation during construction. This method can accurately acquire millimeter-level three-dimensional displacement changes without contacting the structure and is applicable to key construction stages such as arch bridge support loading, concrete pouring, and formwork unloading.

[0014] The above-mentioned objectives of the present invention are achieved through the following technical solutions:

[0015] The present invention provides a method for controlling the construction alignment of an arch bridge construction support based on machine vision and point cloud technology, comprising the following steps:

[0016] S1. Data Acquisition System Construction: Laser-structured light integrated fully automatic point cloud acquisition devices are deployed in the safe zone on both sides of the arch bridge construction area. The single-station range is ≥100 m, the spatial resolution is ≤2 mm, and the ranging error is ≤±1 mm, forming a closed observation network symmetrically along the arch axis. ≥12 stainless steel reflective GCP marker posts are deployed simultaneously and measured by total station + GNSS to serve as the benchmark for the unified construction coordinate system. The acquisition devices are triggered to scan at a cycle of 3 minutes during the pouring stage and 10 minutes during normal operation to acquire the original point cloud of the support and arch rib surface in real time.

[0017] S2. Point Cloud Processing and Image Projection: Perform Statistical Outlier Removal, radius filtering, and ICP fine registration sequentially on the original point cloud, with an iteration mean square error ≤ 0.1 mm, and unify to the same coordinate system; project the registered point cloud into a 2 mm grid DEM, and generate a 16-bit Hillshade grayscale image with a virtual light source azimuth angle of 45° and an elevation angle of 60°, containing at least two sets of perspectives: vertical 90° and tilt 45°, to form a time-series image sequence;

[0018] S3. Digital Image Correlation (DIC) Analysis: Using ZNCC as the criterion, correlation matching with a 150×150 pixel window and 50% overlap is performed on the Hillshade images of the same viewpoint at adjacent time points. Subpixel displacement is obtained by quadratic polynomial fitting. The planar displacement (Δx, Δy) is output from the vertical viewpoint, and the composite displacement ΔY′ is output from the tilted viewpoint. The vertical displacement is then inverted according to Δz=ΔY′ / tanθ to synthesize the three-dimensional displacement vector (Δx, Δy, Δz) of each node. Finally, Kalman filtering is used for time-series smoothing.

[0019] S4. 3D Displacement Calculation and Visualization: Map the DIC displacement back to the DEM mesh, calculate the combined displacement and direction, and use a 3D graphics rendering engine to generate continuous displacement cloud maps, deformation vector maps, and displacement-velocity-acceleration curves of key points; the system self-checks that the DIC-total station residual RMSE is ≤0.8 mm and outputs displacement results with confidence intervals.

[0020] S5. Deformation Trend Analysis and Safety Early Warning: After smoothing the time-series displacement using a smoothing filter, the velocity v and acceleration a are calculated; a four-level threshold model of "displacement-velocity-acceleration-trend" is established, and the instability probability is corrected in real time by combining inverse velocity extrapolation and Bayesian updates; when the velocity at any monitoring point increases monotonically three times consecutively and exceeds the threshold, an audible and visual alarm is automatically triggered, and an electronic notification containing coordinates, the proportion of exceeding the limit, and handling suggestions is pushed to the construction terminal, with a false alarm rate ≤1%;

[0021] S6. Construction Control and Feedback: The measured three-dimensional displacement field is compared with the design alignment in real time through standard industrial communication protocols. If the displacement or rate exceeds the threshold, the system automatically generates digital instructions to adjust the concrete pouring rate, layer thickness, support unloading sequence, or tension force. After confirmation, these instructions are issued for execution. The adjustment effect is continuously monitored. If the displacement rate decreases, it is marked as "positive feedback"; otherwise, a review is prompted. All point clouds, DIC results, early warning logs, construction parameters, and control instructions are stored in an integrated database with timestamps and digital signatures to achieve traceable alignment control throughout the entire process.

[0022] Furthermore, step S1, the construction of the data acquisition system, specifically includes:

[0023] S101. Fully automatic point cloud acquisition instruments are deployed in the safe zones on both sides of the arch bridge construction area. The point cloud acquisition instruments are integrated laser scanning and structured light stereo imaging devices with a single scanning distance ≥100 m, spatial resolution better than 2 mm, and ranging error ≤±1 mm. The acquisition instruments are symmetrically arranged along the arch axis on the outer side of the arch foot, the middle of the main arch span, and both ends of the bridge deck to form a closed observation network. Each acquisition instrument is fixed to a steel base by expansion anchors or concrete embedded parts. The coordinates and attitude of the base are determined by a total station and unified to the construction control coordinate system, and remain stationary during construction.

[0024] S102. No fewer than 12 high-strength stainless steel reflective marker posts shall be set up around the construction area as ground control points (GCPs), and they shall be evenly distributed in a ring on both sides of the bridge and within the main span. Each marker post shall be rigidly connected to a fixed foundation. Its three-dimensional coordinates shall be determined by a total station and GNSS joint measurement, and used for point cloud spatial registration and time series unification.

[0025] S103. High color rendering index LED lights are arranged on both sides of the arch rib and the support. The brightness of the lights is controlled synchronously with the data acquisition instrument. A light shield or sunshade is installed above each data acquisition instrument to ensure consistent illuminance during day and night data acquisition and to avoid direct sunlight and noise.

[0026] S104. Set up a central control host, which connects to each data acquisition instrument via wired Ethernet or industrial-grade wireless network. The main control system triggers scanning at intervals of 3 minutes during the pouring stage and 10 minutes during the regular monitoring stage, and records the timestamp, equipment number and attitude parameters to achieve precise alignment of time-series point clouds.

[0027] S105. Each data acquisition unit integrates a temperature and humidity sensor and an automatic calibration unit in its casing. Before data acquisition, it performs equipment self-test, lens cleaning test, and laser intensity adjustment. The internal inertial measurement unit (UTC timestamp IMU) compensates for equipment micro-vibrations or attitude changes in real time, reducing environmental disturbance errors.

[0028] S106. After on-site debugging, the optimal observation range of the system was determined, and a local high-density scanning mode was enabled for the outer contour of the arch rib, the support columns, and the tie rod nodes, so that the original point cloud density was ≥1×10⁻⁶. 6 Points per square meter (m²) are transmitted in real time to a central server and automatically backed up, providing input for subsequent point cloud processing and digital image correlation analysis.

[0029] Further, step S2, point cloud processing and image projection, specifically includes:

[0030] S201. For the original point cloud of each acquisition cycle, execute the statistical outlier removal algorithm and radius neighborhood filtering in sequence: calculate the distance from the point to the neighborhood mean in the k-nearest neighbor. If the distance is greater than 2σ, delete it and remove isolated clusters with fewer than the set number of points, so that the point cloud density is uniform and the structural surface features are intact after cleaning.

[0031] S202. Using the point cloud at the first time step as the reference, the ground control point GCP is used as the initial transformation, and then the nearest point ICP is iterated for fine registration. The iteration continues until the mean square error between two adjacent points is ≤0.1 mm, thus completing the unification of the time-series point cloud coordinates.

[0032] S203. Project the registration point cloud along the normal of the outer surface of the structure onto a 2 mm × 2 mm regular grid, calculate the center elevation of each grid using the weighted average method, reconstruct the DEM of the arch rib, arch foot and support column in blocks, and form the overall DEM by splicing the boundaries.

[0033] S204. Set the virtual light source azimuth angle to 45°, elevation angle to 60°, and cosine attenuation light intensity model in the DEM coordinate system to generate a 16-bit grayscale Hillshade image;

[0034] S205. Project the same DEM twice, using a vertical viewing angle of 90° and an oblique viewing angle of 45° respectively. Use bilinear interpolation to maintain pixel equidistance and obtain two sets of images: one for planar displacement using Hillshade(90°) and the other for composite displacement using Hillshade(45°).

[0035] S206. Set the Hillshade ground pixel resolution to be consistent with the DEM grid spacing, that is, each pixel represents 2mm of actual length. The image size is automatically cropped so that the arch rib and the key area of ​​the support are located in the center, and the pixel grayscale depth is 16 bits.

[0036] S207. Using the sum of image grayscale variance and edge gradient as the evaluation function, automatically optimize the azimuth and elevation angles of the light source to maximize the evaluation function and enhance the contrast of surface micro-textures.

[0037] S208. All time-series Hillshade images use the same projection matrix, light source vector, and grayscale mapping scale, and the parameters are written to the metadata file for consistent use by the DIC program;

[0038] S209. For DEM holes caused by temporary occlusion or reflection, if the missing area is ≤50 mm×50 mm, the weighted average elevation value of adjacent valid points is used to fill the hole; if it is >50 mm×50 mm, it is marked as an occlusion area and excluded in subsequent DIC analysis, thereby obtaining a high-precision, low-noise, and time-consistent multi-angle Hillshade image sequence.

[0039] Further, step S3, digital image correlation (DIC) analysis, specifically includes:

[0040] S301. For Hillshade images at adjacent times with the same projection angle, a phase correlation algorithm is used for sub-pixel level registration. The mean square error test shows that the overall misalignment is ≤0.2 pixels, and the images are cropped to the same image size.

[0041] S302. Divide the registered image into 150×150 pixel correlation windows with a step size of 50 pixels and 50% overlap. The center of each window forms a measurement node.

[0042] S303. Using the zero-mean normalized cross-correlation (ZNCC) function as the matching criterion, the maximum value of the correlation coefficient is found within the search area to obtain the integer pixel displacement;

[0043] S304. For integer displacements, a quadratic polynomial surface fitting is used to fit the relevant peak neighborhood, and the extreme points are used as sub-pixel displacements.

[0044] S305. Bicubic spline interpolation is used to reconstruct the displacement of each node into a continuous planar displacement field, and median filtering is performed using the local mean ±3σ threshold to remove outlier vectors, thus obtaining (Δx,Δy) from the vertical perspective.

[0045] S306. Repeat S301-S305 for the Hillshade image with a 45° tilted viewpoint to obtain the composite displacement ΔY′, and invert the vertical displacement Δz according to Δz=ΔY′ / tan 45° to synthesize the three-dimensional displacement vector (Δx,Δy,Δz) of each pixel node;

[0046] S307. Kalman filtering is used to smooth the temporal three-dimensional displacement field to suppress instantaneous jumps and random noise;

[0047] S308. Real-time statistics of subset matching success rate. If <95%, automatically expand the window or search range and recalculate until the requirements are met. Finally, output continuous time-series 3D displacement cloud map and key point displacement curve.

[0048] Furthermore, step S4, three-dimensional displacement calculation and visualization, specifically includes:

[0049] S401. The planar displacement (Δx, Δy) and vertical displacement Δz output by DIC are transformed inversely according to the recorded projection matrix and light source vector to complete the unified mapping of pixel-geometric coordinates, so that each pixel displacement corresponds one-to-one with the DEM grid node.

[0050] S402. Calculate the three-dimensional displacement vector and its resultant displacement of each mesh node. And retain the component sign to determine the direction of upward arching, downward deflection or horizontal offset;

[0051] S403. Inverse distance weighting and bicubic spline hybrid interpolation are used to reconstruct the spatial continuity of discrete node displacements, forming a continuous three-dimensional displacement field covering the arch ribs, tie rods and support columns;

[0052] S404. Based on a 3D graphics rendering engine, it overlays the DEM before and after deformation, uses cool-warm color gradients and vector arrows to express the magnitude and direction of displacement, and supports real-time interactive rotation and scaling.

[0053] S405. Generate a deformation map of the difference between adjacent time points and a residual distribution map of the DIC-total station. The system verifies that the residual RMSE ≤ 0.8 mm, which meets the accuracy requirements for structural monitoring.

[0054] S406. Automatically extract the X, Y, Z time-series displacement curves and combined displacement curves of the arch mid-span, arch foot, hanger node and support column bottom, calculate displacement rate and acceleration, and trigger an early warning if the rate increases monotonically three times in a row and exceeds the preset threshold.

[0055] S407. An uncertainty assessment model is introduced, which integrates DIC matching, point cloud registration and sensor ranging errors, and calculates the overall uncertainty through error propagation analysis. The displacement results of each grid node are accompanied by a confidence interval label, which is convenient for use in subsequent safety assessments.

[0056] Furthermore, step S5, deformation trend analysis and safety early warning, specifically includes:

[0057] S501. The three-dimensional displacement time series data of the same monitoring point are sorted by timestamp, and the displacement curve is smoothed by smoothing filter. The filter window is automatically adjusted according to the sampling frequency to maintain the trend and filter out high-frequency noise.

[0058] S502. Calculate the rate of change of displacement between adjacent time points using the central difference method to obtain the velocity sequence; then differentiate the velocity sequence to obtain the acceleration curve; by analyzing the variation law of velocity and acceleration, identify the deformation characteristics of the arch rib and support at different stages of construction.

[0059] S503. Establish a four-level multi-threshold discrimination model:

[0060] ① Displacement threshold: set according to design specifications;

[0061] ② Rate threshold: If the displacement rate of a key point exceeds the warning value for three consecutive observations, it is determined to be an abnormal acceleration.

[0062] ③Acceleration threshold: Potential instability is determined when the acceleration suddenly jumps and continues to exceed twice the standard deviation of the mean;

[0063] ④ Trend threshold: Using regression analysis and inverse velocity method extrapolation, an early warning is issued when the extrapolated curve approaches the limit state in a certain period of time in the future;

[0064] The current structural state level is output by combining these parameters using a logical weighted algorithm.

[0065] S504. Introducing a Bayesian update mechanism to correct the risk probability distribution of each monitoring point in real time and integrate it with historical data to improve the robustness of early warning; when the status of any monitoring point reaches the warning level, the system highlights the area on the 3D interface and pops up the over-limit parameters. When the danger level is reached, an audible and visual alarm is immediately issued, and an electronic notification containing the monitoring point number, coordinates, displacement, rate, acceleration, over-limit ratio and suggested handling measures is pushed to the terminals of the on-site supervisor, general contractor and safety manager through the construction monitoring platform.

[0066] S505. Set up an automatic verification mechanism: For single-point anomalies, first check the correlation between the displacement of the neighborhood and the point cloud density. If there is no consistent deformation in the neighborhood or the data quality is low, it is judged as data anomaly and removed. Only after verification is it included in the early warning output, reducing the false alarm rate to ≤1%.

[0067] S506. Establish an alarm event database to automatically record alarm time, monitoring point number, displacement rate and acceleration value, handling measures and results, support historical playback and statistical analysis, and realize a safety assessment with full traceability.

[0068] Furthermore, step S6, construction control and feedback, specifically includes:

[0069] S601. A construction control interface module is set up at the main control terminal of the monitoring system, which is interconnected with the field automation control system and monitoring center through the standard industrial communication protocol to transmit three-dimensional displacement field, velocity curve and early warning level in real time; the communication link adopts bidirectional data verification and redundancy design.

[0070] S602. The monitoring area is divided into several control units, each unit corresponding to a unique construction parameter, namely concrete pouring rate, layer thickness, support unloading sequence or tension force; when the displacement or rate of a certain unit exceeds the threshold set in S503, the system automatically retrieves the corresponding parameter and generates a digital adjustment command, which is then sent to the construction dispatch terminal after confirmation by the supervisor or on-site personnel.

[0071] S603. After the control command is executed, the system continuously monitors the change in displacement rate: if the displacement rate decreases and then returns to stability after adjustment, the system automatically determines that the control is effective and records it as "positive feedback"; if the deformation continues to expand or the fluctuation intensifies after adjustment, the system determines that the control is ineffective and prompts for verification of construction measures.

[0072] S604. Establish an integrated construction monitoring and control database to uniformly store point clouds, DIC results, displacement fields, early warning logs, construction parameters, and control commands. All data are accompanied by timestamps and digital signatures to form a complete time-series archive. The database supports multi-dimensional retrieval and visualization playback by construction stage, monitoring area, and event type, providing a traceable basis for project acceptance, quality assessment, and subsequent maintenance.

[0073] The present invention has the following beneficial effects:

[0074] (1) Non-contact, full-field three-dimensional monitoring: By fusing point cloud with machine vision, the displacement field of the entire arch rib or support surface can be obtained without deploying physical sensors.

[0075] (2) High spatiotemporal resolution: It can continuously monitor at a sampling frequency of minutes to achieve dynamic tracking of the entire pouring process.

[0076] (3) Automation and real-time performance: Automatic data acquisition and algorithm processing are used to achieve unattended automated monitoring and real-time data updates.

[0077] (4) High precision and anti-interference: Combining multi-view Hillshade projection and DIC sub-pixel matching, the three-dimensional displacement accuracy can reach the millimeter level.

[0078] (5) Early warning and decision support: By analyzing deformation rate and trend, potential instability can be predicted in advance, providing a quantitative basis for construction safety management.

[0079] (6) Wide applicability: In addition to arch bridges, it can also be extended to construction monitoring and deformation control of concrete structures such as arch culverts, tunnel entrances, and dam slopes. Attached Figure Description

[0080] Figure 1 This is a flowchart illustrating the overall process of the arch bridge construction support alignment control method based on machine vision and point cloud technology according to the present invention.

[0081] Figure 2 A schematic diagram of the monitoring system layout for a steel-concrete composite tied arch bridge (symmetrical arrangement).

[0082] Figure 3 This is a schematic diagram of the point cloud data preprocessing and image generation process.

[0083] Figure 4 Render images for shadows at 90-degree and 45-degree angles;

[0084] Figure 5 This is a schematic diagram of digital image correlation (DIC) analysis.

[0085] Figure 6 A time-series prediction diagram of displacement at key monitoring points during the arch rib casting process;

[0086] Figure 7 This is a flowchart for deformation trend analysis and safety early warning. Detailed Implementation

[0087] The following detailed description, with reference to the accompanying drawings, illustrates the implementation of the arch bridge construction support alignment control method based on machine vision and point cloud technology described in this invention. This embodiment uses the support construction of a large-span reinforced concrete arch bridge as an application scenario to specifically describe its operation process.

[0088] S1. Data Acquisition System Construction:

[0089] The system mainly consists of the following components: a fully automatic point cloud data acquisition device, a ground control point (GCP) network, a stable mounting bracket, an ambient lighting device, and a synchronous control module.

[0090] S101. Several fully automatic point cloud acquisition devices are deployed in the safe zone on both sides of the arch bridge construction area. The point cloud acquisition devices are preferably integrated devices with high-precision laser scanning and structured light stereo imaging capabilities, with a single scan distance of not less than 100m, a spatial resolution better than 2mm, and a ranging error not exceeding ±1mm. To ensure panoramic coverage of the arch ribs and supports, the acquisition devices are symmetrically arranged along the arch axis, with six observation stations set up at the outer side of the arch foot, the mid-span of the main arch, and both ends of the bridge deck, forming a closed observation network. Each acquisition device is fixedly installed on a dedicated steel base, which is connected to a stable platform via expansion bolts or concrete embedded parts to ensure no displacement during construction. The spatial position and installation attitude of each acquisition device are determined by a total station during the system initialization phase and unified to the construction control coordinate system, remaining stationary during construction.

[0091] S102. Establish a ground control point network (GCP) around the construction area. This network consists of high-strength stainless steel reflective marker posts, with no fewer than 12 control points distributed in a ring shape on both sides of the bridge and within the main span, evenly distributed longitudinally and laterally. Each control point is rigidly connected to a fixed foundation, and its coordinates are determined through combined measurements using a total station and GNSS. The control points are used for spatial registration and temporal series unification of point cloud data, providing a stable reference benchmark for subsequent three-dimensional displacement calculations.

[0092] S103. After the point cloud acquisition instrument and GCP are deployed, establish an environmental lighting and shading system. Since concrete pouring often takes place at night or on cloudy days, to ensure the quality of the acquired point cloud texture information, adjustable-angle high color rendering index LED lights are installed on both sides of the arch ribs and supports. The brightness of the lights is controlled synchronously with the acquisition system to maintain stable illumination. To reduce noise interference caused by direct sunlight or strong reflections, a shade or awning is installed above the equipment to ensure consistent lighting conditions for each acquisition.

[0093] S104. The fully automatic data acquisition and control module used in this embodiment consists of a central control host, a communication network, and a timing control program. Each data acquisition device is connected to the main control computer via wired Ethernet or an industrial-grade wireless network. The main control system automatically triggers point cloud acquisition according to the set time intervals. The sampling interval is set to 3 minutes during the pouring stage and 10 minutes during the regular monitoring stage. The main control system uniformly synchronizes the acquisition time of all devices and records the timestamp, device number, and attitude parameters of each scan to ensure that point cloud data at different times can be accurately aligned on the timeline.

[0094] S105. To improve the system's environmental adaptability, temperature and humidity sensors and an automatic calibration unit are installed on the housing of the data acquisition device to correct for the impact of the environment on ranging accuracy in real time. Before each data acquisition, the system automatically executes a calibration procedure, including device self-test, lens cleaning check, and laser intensity adjustment, to ensure data consistency. The data acquisition unit integrates an inertial measurement unit (IMU), which can compensate for minor vibrations or changes in posture, further reducing errors caused by external disturbances.

[0095] S106. After the data acquisition system is set up at the construction site, the optimal observation range is determined through on-site debugging. For key areas such as the outer contour of the arch rib, support columns, and tie rod nodes, a high-precision local scanning mode is set to improve local point cloud density. The average density of the raw point cloud obtained by the system is no less than 1 million points per square meter, which can completely cover the entire arch rib and support system. All raw data is transmitted to the central server in real time and automatically backed up, serving as input for subsequent point cloud processing and digital image correlation analysis.

[0096] Through the above-described deployment method, this embodiment forms a multi-view point cloud acquisition network covering the entire main arch and support, enabling high-precision and continuous acquisition of the structural surface morphology during the construction of the arch bridge.

[0097] S2. Point Cloud Processing and Image Projection

[0098] The point clouds from different time series are uniformly processed to form highly consistent input data that meets the requirements of digital image correlation (DIC) analysis. The point cloud processing and image projection steps in this embodiment mainly include five stages: data preprocessing, spatial registration, rasterization reconstruction, Hillshade image generation, and parameter optimization.

[0099] S201. Preprocess the raw point cloud data obtained in each acquisition cycle. The point cloud data may contain environmental noise points, construction machinery reflection points, and non-structural surface data during the acquisition process. To improve model purity, a combination of a statistical outlier removal algorithm and a radius-based neighborhood filtering algorithm is used to clean the point cloud. Specifically: the distance between each point and the neighborhood mean is calculated within the local neighborhood; if this distance exceeds twice the standard deviation σ, it is identified as an outlier and deleted; simultaneously, isolated clusters with fewer than a set number of points are removed entirely. After cleaning, the point cloud density remains uniform, and the structural surface features are intact.

[0100] S202. Perform spatial registration and coordinate unification. Since point clouds acquired at different times may exhibit slight attitude deviations, they need to be unified to a unified coordinate system. Using the point cloud at the first time step as a reference, fine registration is performed using ground control points (GCPs) and the Iterative Closest Point (ICP) algorithm. The ICP algorithm iteratively calculates the set of closest point pairs between point clouds at two time steps, minimizing the Euclidean distance error and achieving high-precision spatial overlap. The registration error is controlled within ±0.1 mm to ensure spatial consistency between time-series point clouds.

[0101] S203. Perform point cloud rasterization and digital elevation model (DEM) reconstruction. To balance computational efficiency and detail preservation, this embodiment projects the point cloud onto a regular grid along the normal direction of the outer surface of the structure, with a grid spacing of 2 mm. For each grid cell, a weighted average method is used to calculate its center elevation value, generating the corresponding DEM. This process establishes independent DEM blocks in three-dimensional space for the outer surface of the arch rib, the arch foot node area, and the support column area, and finally forms the overall digital elevation model by boundary stitching. The DEM reconstruction results are saved in a unified coordinate system, providing a basis for subsequent image generation.

[0102] S204. To enhance the texture features of the structural surface and improve the matching accuracy of digital image correlation analysis, the generated DEM is subjected to Hillshade image projection processing. A Hillshade image is a grayscale shadow map formed by projecting simulated lighting onto a three-dimensional surface, reflecting the subtle undulations and roughness characteristics of the structural surface. A virtual light source position is set within the DEM coordinate system, with the azimuth angle set to 45° and the elevation angle set to 60°. A cosine attenuation model is used for the light intensity distribution, generating a 16-bit grayscale Hillshade image.

[0103] S205. To achieve three-dimensional displacement decomposition, two Hillshade images with different viewpoints need to be generated. One set of images, with a projection angle perpendicular to the structure surface and a viewpoint of 90°, is used to extract the planar displacement components (Δx, Δy); the other set, with a viewpoint set at a 45° tilt direction, is used to extract the composite directional displacement (ΔY′). To avoid non-uniform pixel scaling caused by projection distortion, a bilinear interpolation algorithm is used during the projection transformation process to maintain the equidistant relationship between pixels.

[0104] S206. Regarding image resolution, the ground pixel resolution of the Hillshade image is consistent with the DEM grid spacing, meaning each pixel represents a true length of 2mm. The image size is automatically adjusted according to the monitoring area to ensure that the arch ribs and key areas of the support are all within the center of the image field of view. The generated image uses 16-bit grayscale depth to retain more brightness levels and avoid detail loss caused by excessive compression.

[0105] S207. During the Hillshade generation process, the light source direction and brightness parameters are adaptively optimized. By calculating the image grayscale variance and edge gradient distribution, the lighting combination that maximizes surface texture contrast is automatically selected. This optimization process allows the image to maintain overall uniform illumination while highlighting subtle geometric differences on the surface, improving the recognizability of subsequent DIC matching.

[0106] S208. To ensure consistency of time-series data, the same projection parameters and grayscale mapping scale are used for all Hillshade images at all times. During the image generation stage, the system records the projection matrix, light source vector, and pixel spatial resolution of each frame and stores them in a metadata file for the DIC analysis program to read, thus achieving consistent parameter access.

[0107] S209. To prevent local data gaps caused by temporary shading or reflections during the construction phase, this embodiment introduces a local reconstruction algorithm based on neighborhood elevation interpolation. When data in a certain grid cell is missing, it is automatically filled in using the weighted average elevation value of adjacent valid points. If the missing area exceeds a set threshold (e.g., 50mm × 50mm), the area is marked as an shading area and excluded in subsequent DIC analysis.

[0108] Through the above processing, this embodiment obtains a high-precision, low-noise DEM sequence with good temporal consistency and a corresponding multi-angle Hillshade image sequence.

[0109] S3. Digital Image Correlation (DIC) Analysis

[0110] After obtaining a multi-time- and multi-view Hillshade image sequence, Digital Image Correlation (DIC) is performed on images from adjacent time points to calculate the displacement field of the arch bridge support and arch rib structure surface. The DIC analysis method used in this embodiment is a non-contact, full-field displacement calculation technology capable of achieving millimeter-level or even sub-millimeter-level displacement detection accuracy, making it particularly suitable for dynamic deformation monitoring during structural construction. The processing flow in this step mainly includes: image registration, correlation window division, grayscale correlation calculation, sub-pixel interpolation solution, displacement field reconstruction, and error control.

[0111] S301. Register and crop Hillshade images at adjacent time points under the same projection angle. Due to slight rotation or scale deviations during point cloud generation, a fast registration algorithm based on phase correlation is used to ensure one-to-one pixel correspondence. This algorithm quickly obtains translation and rotation parameters by calculating the Fourier transform of the two images and determining the phase difference, achieving sub-pixel-level alignment. The registration results are checked for mean square error to ensure that the overall misalignment does not exceed 0.2 pixels, thereby guaranteeing the accuracy of subsequent DIC calculations.

[0112] S302. The registered Hillshade image is divided into several overlapping correlation calculation windows. In this embodiment, the size of each window is set to 150×150 pixels, the step size between windows is 50 pixels, and the windows maintain an overlap rate of approximately 50% to achieve a balance between spatial resolution and noise resistance. The center point of each subset represents a measurement node, corresponding to a sampling area on the actual structural surface.

[0113] S303. In the correlation calculation stage, the zero-mean normalized cross-correlation (ZNCC) function is used as the matching criterion. The ZNCC function can effectively eliminate the influence of illumination changes and overall brightness shifts, ensuring stable and reliable matching results. Its correlation coefficient is defined as follows:

[0114]

[0115] in, and These are the grayscale values ​​of corresponding pixels in the reference image and the deformed image, respectively. , Let u and v be the mean grayscale values ​​within the window, and u and v be the displacement values. The integer pixel displacement of each subset is obtained by calculating the position of the maximum correlation coefficient within the defined search area.

[0116] S304. To further improve measurement accuracy, sub-pixel interpolation is performed on the integer displacement results. This embodiment employs a sub-pixel interpolation algorithm based on quadratic polynomial fitting, fitting a three-dimensional surface function within the neighborhood of relevant peaks and obtaining the positions of its extreme points as the sub-pixel-level displacement results. This method can achieve pixel-level resolution under conditions of high grayscale signal-to-noise ratio.

[0117] S305. After solving for the displacements of each subset, a continuous planar displacement field is formed through spatial interpolation and filtering. A smooth displacement distribution is reconstructed between the center points of the subsets using bicubic spline interpolation, while median filtering is applied to remove outlier vectors. For vectors exceeding three times the local mean standard deviation σ, the system automatically marks them as invalid and re-interpolates to avoid noise interference. At this point, the planar displacement components (Δx, Δy) under the vertical view projection can be obtained.

[0118] S306. To obtain the three-dimensional displacement information of the structural surface, this embodiment simultaneously performs the same DIC calculation process on the Hillshade image under an oblique viewing angle (45°) to obtain the composite directional displacement component (ΔY′). Combining the two sets of projection results, according to the geometric relationship:

[0119]

[0120] Where θ is the projection tilt angle (45°), the vertical displacement Δz can be inverted. This geometric solution process is completed automatically by the system, and the calculation results, together with the planar displacement components, constitute the three-dimensional displacement vector (Δx, Δy, Δz) of each pixel node.

[0121] S307. To ensure computational stability and temporal continuity, a temporal smoothing and noise suppression algorithm is introduced after the three-dimensional displacement field is generated. The system performs a time-weighted average of the displacement results at adjacent time steps and uses a Kalman filter model to eliminate instantaneous jumps and random noise. Comparative experiments show that after smoothing, the overall noise level is reduced and the displacement curve is more stable.

[0122] S308. During the DIC calculation process, the system records the correlation coefficient distribution, subset matching success rate, and anomaly detection statistics in real time for quality assessment. If the matching success rate is below 95%, the system automatically adjusts the subset size or search window range and recalculates to ensure overall reliability.

[0123] By following the DIC analysis steps described above, the three-dimensional displacement field of the arch bridge support and arch ribs can be obtained between any two time points. By superimposing the results from multiple time points, a continuous time-series displacement cloud map and displacement curves of key points can be formed.

[0124] S4. Three-dimensional displacement calculation and visualization

[0125] After completing the digital image correlation (DIC) analysis and obtaining the planar displacement components (Δx, Δy) and vertical displacement components (Δz), this embodiment further conducts three-dimensional displacement calculation and visualization processing to intuitively reflect the overall deformation state of the arch bridge support and arch ribs at each stage of construction.

[0126] S401. Project all component data output from the DIC analysis onto the same coordinate system. Since the planar and vertical displacements originate from Hillshade images projected at different angles, geometric registration and pixel alignment are required to establish their correspondence. Based on the projection matrix and light source vector parameters recorded in the preceding steps, the system establishes an inverse projection transformation matrix to uniformly map the coordinates of each pixel. The mapped pixels have unique coordinate indices in three-dimensional space, allowing direct correspondence with DEM grid nodes, thus achieving one-to-one spatial matching of displacement data.

[0127] S402. After data alignment, calculate the 3D displacement vector for each mesh node. The calculation formula is as follows:

[0128]

[0129] Where, Δr i This represents the resultant displacement of the i-th grid node, in millimeters. The system also records the sign information of each component's direction to determine the directional characteristics of the displacement (such as upward arching, downward deflection, horizontal offset, etc.).

[0130] S403. To facilitate overall structural analysis, the system performs spatial interpolation and mesh reconstruction on the three-dimensional displacement field. A hybrid interpolation algorithm combining inverse distance weighted (IDW) and bicubic spline interpolation is used to spatially continuousize the displacements at each node. This method ensures the fidelity of local deformation characteristics while smoothing out noise points and isolated outliers. The interpolated three-dimensional displacement field covers the entire arch rib, tie rod, and support column area, forming a continuous and smooth deformation distribution map.

[0131] S404. After the calculation is completed, the system automatically generates a 3D visualization model. The visualization module is developed based on a 3D graphics rendering engine (such as OpenGL or VTK), which overlays the DEM model before and after deformation, and displays the magnitude and direction of displacement in each region through color mapping and vector arrows. The color gradient from cool to warm colors indicates that the displacement amplitude is from small to large, and the length and direction of the vector arrows correspond to the displacement vector of each node. Users can freely rotate and scale the model in the visualization interface to observe the spatial deformation morphology of the arch ribs and supports in real time.

[0132] S405. The system can generate a differential deformation map and a residual displacement distribution map. The differential deformation map is used to represent the instantaneous deformation between adjacent time points, facilitating the identification of local stress concentrations or structural abrupt changes during the casting stage. The residual displacement distribution map is used to compare the differences between the DIC calculation results and the control point measurement data to evaluate the system accuracy. Verification shows that the root mean square error (RMSE) between the displacement results output by the system in this embodiment and the total station measured data does not exceed 0.8 mm, meeting the accuracy requirements for structural monitoring.

[0133] S406. After the three-dimensional displacement calculation is completed, the system automatically extracts the time-series displacement curves of several key monitoring points. Key points include locations such as the mid-span of the arch, the arch foot, the hanger nodes, and the bottom of the support columns. The system outputs the displacement curves and the combined displacement curves of each key point in the X, Y, and Z directions, and calculates the displacement rate and acceleration. If the displacement rate of a key point shows a monotonically increasing trend for three consecutive observations and exceeds a preset threshold, the system will trigger an early warning signal and generate an automatic report to be pushed to the monitoring and control terminal.

[0134] S407. To further improve the accuracy of the analysis, this embodiment introduces an uncertainty assessment model. This model comprehensively considers DIC matching error, point cloud registration error, and sensor ranging error, and calculates the overall uncertainty using the error propagation analysis method. The displacement result of each grid node is accompanied by a confidence interval label for easy use in subsequent safety assessments.

[0135] Through this step, the system ultimately achieves three-dimensional deformation visualization and quantitative assessment of the arch bridge support and arch rib structure.

[0136] S5. Deformation Trend Analysis and Safety Early Warning

[0137] After obtaining the three-dimensional displacement fields of the arch bridge support and arch ribs at multiple moments, this embodiment further establishes a deformation trend analysis and safety early warning module based on time series analysis to achieve trend identification of structural deformation development and real-time assessment of construction safety status. This module comprehensively considers displacement, displacement rate, acceleration, and structural geometric characteristics, and uses a multi-parameter joint discrimination model to achieve early identification and graded early warning of potential risks in the construction process.

[0138] S501. The three-dimensional displacement data at each monitoring time point are processed and smoothed over time. The system sorts the displacement data of the same monitoring point according to the acquisition timestamp, forming a continuous time-series displacement curve. To reduce the impact of measurement noise and construction transients, the Savitzky-Golay filtering method is used to smooth the original curve, maintaining the overall trend of the curve while filtering out high-frequency disturbances. The filtering window is automatically adjusted according to the sampling frequency.

[0139] S502. Calculate displacement rate and acceleration by differentiating the time-series displacement curve. The system uses the central difference method to calculate the rate of change of displacement at adjacent time points to obtain the velocity sequence; then, by differentiating the velocity sequence, the acceleration curve is obtained. By analyzing the variation law of velocity and acceleration, the deformation characteristics of the arch rib and support at different stages of construction can be identified.

[0140] S503. To further achieve quantitative early warning, this embodiment establishes a multi-threshold discrimination model. This model includes four levels: displacement threshold, velocity threshold, acceleration threshold, and trend threshold.

[0141] 1. Displacement threshold: determined according to design specifications and structural stress analysis;

[0142] 2. Rate threshold: When the displacement rate of a key point exceeds the warning value for three consecutive observations, it is determined to be an abnormal acceleration phase;

[0143] 3. Acceleration threshold: If the acceleration curve shows a sudden jump and continues to exceed twice the standard deviation of the mean, the system determines that there is a potential risk of instability;

[0144] 4. Trend threshold: Regression analysis and the inverse velocity method are used to predict the displacement trend. When the extrapolated curve approaches the limit state in a certain period of time in the future, an early warning is issued.

[0145] The above discrimination model is based on multi-parameter joint logic and uses a logical weighting algorithm to comprehensively determine the current structural state level.

[0146] S504. To improve the robustness of the model, this embodiment introduces a probability correction mechanism based on Bayesian updates. The system updates the risk probability distribution of each monitoring point in real time and dynamically adjusts it in conjunction with historical monitoring data, making the early warning results more reliable.

[0147] Regarding the output of early warning results, the system provides a multi-level information display and linkage mechanism. When any monitoring point reaches the early warning state, the system automatically highlights the corresponding area in the 3D visualization interface and pops up a prompt message explaining the specific parameters and values ​​exceeding the limits. When a dangerous state is reached, the system immediately issues an audible and visual alarm signal and sends an electronic notification to the terminals of the site supervisor, general contractor, and safety manager through the construction monitoring platform. The alarm information includes: monitoring point number, location coordinates, current displacement, velocity, acceleration, exceedance ratio, and suggested handling measures.

[0148] S505. To prevent false alarms, the system employs an automatic verification mechanism. When a single-point anomaly occurs, the system automatically checks the displacement correlation and point cloud density distribution in the surrounding area. If there are no similar changes in the neighboring area or the data quality is low, it is determined to be a data anomaly rather than a true deformation. Only after successful verification can the anomaly be included in the early warning output. This mechanism significantly reduces the false alarm rate caused by noise or local occlusion.

[0149] S506. To achieve full-process recording and traceability, the system establishes a data log and event database. Each alarm event automatically records the alarm time, monitoring point number, parameter value, handling measures, and results. The system supports historical data playback and statistical analysis, facilitating safety assessments and experience summaries after construction.

[0150] In summary, the deformation trend analysis and safety early warning method of this embodiment can transform complex temporal point cloud and image-related data into operable safety indicators.

[0151] S6. Construction Control and Feedback

[0152] After completing the three-dimensional displacement monitoring, deformation trend analysis, and safety early warning of the arch bridge support and arch rib structure, this embodiment further establishes a construction control and feedback mechanism to realize the closed-loop transmission of monitoring data to the construction scheduling and control system, thereby enabling real-time optimization and dynamic safety management of the arch bridge construction process.

[0153] The S601 system features a construction control interface module at the main monitoring control terminal. This module interconnects with the automated control system and monitoring center at the construction site via standard industrial communication protocols (such as OPC UA or Modbus TCP). The three-dimensional displacement field, velocity curves, and early warning level information generated by the monitoring system are transmitted to the control terminal in real time, where the intelligent scheduling program adjusts the construction equipment accordingly. The communication process employs bidirectional data verification and redundant link design to ensure the reliability and timeliness of data transmission.

[0154] S602. Establish construction adjustment decision logic based on monitoring results. The system divides the monitoring area into multiple control units, each corresponding to a construction operation parameter (such as concrete pouring rate, layer thickness, support unloading sequence, or tension force). When an abnormal displacement or rate exceeds the threshold set in S503 in a certain unit area, the system automatically retrieves its control parameters and generates adjustment suggestions. All suggestions are submitted to the construction scheduling terminal in the form of digital commands or manual confirmation.

[0155] S603. After the implementation of construction control, the system monitors the adjustment effect in real time and performs feedback evaluation. If the displacement rate decreases and then stabilizes after adjustment, the system automatically determines that the control is effective and records it as a "positive feedback event"; if the deformation continues to expand or the fluctuation intensifies after adjustment, the system determines that the control is ineffective and prompts for review of construction measures. The feedback evaluation results will be updated to the control strategy database to optimize subsequent decision-making models, enabling the system to self-learn and continuously improve.

[0156] S604. To ensure the traceability of monitoring and control data, this embodiment establishes an integrated database for construction monitoring and control. This database uniformly stores point cloud data, DIC analysis results, displacement field information, early warning logs, construction parameters, and control command execution records. All data is accompanied by timestamps and digital signatures, forming a complete time-series archive. The database supports multi-dimensional retrieval and visual queries, and can quickly locate historical information by construction stage, monitoring area, or event type, providing a traceable basis for project acceptance, quality assessment, and subsequent maintenance.

[0157] like Figure 6The figure shows the displacement time-series prediction curves of key monitoring points during the arch rib casting process. As can be seen from the figure, after adopting the method of this invention, the system detected a sudden increase in the vertical displacement rate of the arch top monitoring point from 0.08 mm / 10 min to 0.32 mm / 10 min approximately 45 minutes after the start of casting, accompanied by the acceleration exceeding the 2σ threshold twice consecutively. The system immediately triggered a "yellow warning" and simultaneously pushed digital instructions to the construction terminal to "reduce the third-stage pumping rate by 20% and temporarily suspend formwork removal." After on-site execution, the displacement rate dropped back to 0.12 mm / 10 min after 20 minutes, the acceleration returned to a stable level, and the curve slope decreased significantly, successfully avoiding displacement exceeding the design limit (design limit 8 mm). This figure visually demonstrates that this invention, through a "point cloud-DIC-trend prediction" closed loop, can provide quantifiable control suggestions at least 10 minutes before the displacement reaches its limit value, verifying the millimeter-level pre-control capability of this invention for the construction alignment of arch bridges.

[0158] In summary, the construction control and feedback method in this embodiment achieves intelligent dynamic management of the entire construction process of the arch bridge support by establishing a closed loop of "monitoring-analysis-early warning-control-feedback".

Claims

1. A method for controlling the construction alignment of arch bridge construction supports based on machine vision and point cloud technology, characterized in that, Includes the following steps: S1. Data Acquisition System Construction: Laser-structured light integrated fully automatic point cloud acquisition devices are deployed in the safe zone on both sides of the arch bridge construction area. The single-station range is ≥100 m, the spatial resolution is ≤2 mm, and the ranging error is ≤±1 mm, forming a closed observation network symmetrically along the arch axis. ≥12 stainless steel reflective GCP marker posts are deployed simultaneously and measured by total station + GNSS to serve as the benchmark for the unified construction coordinate system. The acquisition devices are triggered to scan at a cycle of 3 minutes during the pouring stage and 10 minutes during normal operation to acquire the original point cloud of the support and arch rib surface in real time. S2. Point Cloud Processing and Image Projection: Perform Statistical Outlier Removal, radius filtering, and ICP fine registration sequentially on the original point cloud, with an iteration mean square error ≤ 0.1 mm, and unify to the same coordinate system; project the registered point cloud into a 2 mm grid DEM, and generate a 16-bit Hillshade grayscale image with a virtual light source azimuth angle of 45° and an elevation angle of 60°, containing at least two sets of perspectives: vertical 90° and tilt 45°, to form a time-series image sequence; S3. Digital Image Correlation (DIC) Analysis: Using ZNCC as the criterion, correlation matching with a 150×150 pixel window and 50% overlap is performed on the Hillshade images of the same viewpoint at adjacent time points. Subpixel displacement is obtained by quadratic polynomial fitting. The planar displacement (Δx, Δy) is output from the vertical viewpoint, and the composite displacement ΔY′ is output from the tilted viewpoint. The vertical displacement is then inverted according to Δz=ΔY′ / tanθ to synthesize the three-dimensional displacement vector (Δx, Δy, Δz) of each node. Finally, Kalman filtering is used for time-series smoothing. S4. 3D Displacement Calculation and Visualization: Map the DIC displacement back to the DEM mesh, calculate the combined displacement and direction, and use a 3D graphics rendering engine to generate continuous displacement cloud maps, deformation vector maps, and displacement-velocity-acceleration curves of key points; the system self-checks that the DIC-total station residual RMSE is ≤0.8 mm and outputs displacement results with confidence intervals. S5. Deformation Trend Analysis and Safety Early Warning: After smoothing the time-series displacement using a smoothing filter, the velocity and acceleration are calculated; a four-level threshold model of "displacement-velocity-acceleration-trend" is established, and the instability probability is corrected in real time by combining inverse velocity extrapolation and Bayesian updates; when the velocity at any monitoring point increases monotonically three times consecutively and exceeds the threshold, an audible and visual alarm is automatically triggered, and an electronic notification containing coordinates, the proportion of exceeding the limit, and handling suggestions is pushed to the construction terminal, with a false alarm rate of ≤1%; S6. Construction Control and Feedback: The measured three-dimensional displacement field is compared with the design alignment in real time through standard industrial communication protocols. If the displacement or rate exceeds the threshold, the system automatically generates digital instructions to adjust the concrete pouring rate, layer thickness, support unloading sequence, or tension force. After confirmation, the instructions are issued for execution. The adjustment effect is continuously monitored. If the displacement rate decreases, it is marked as "positive feedback"; otherwise, a review is prompted. All point clouds, DIC results, early warning logs, construction parameters, and control instructions are stored in an integrated database with timestamps and digital signatures to achieve full-process traceable alignment control.

2. The method according to claim 1, characterized in that, Step S1, data acquisition system construction, specifically includes: S101. Fully automatic point cloud acquisition instruments are deployed in the safe zones on both sides of the arch bridge construction area. The point cloud acquisition instruments are integrated laser scanning and structured light stereo imaging devices with a single scanning distance ≥100 m, spatial resolution better than 2 mm, and ranging error ≤±1 mm. The acquisition instruments are symmetrically arranged along the arch axis on the outer side of the arch foot, the middle of the main arch span, and both ends of the bridge deck to form a closed observation network. Each acquisition instrument is fixed to a steel base by expansion anchors or concrete embedded parts. The coordinates and attitude of the base are determined by a total station and unified to the construction control coordinate system, and remain stationary during construction. S102. No fewer than 12 high-strength stainless steel reflective marker posts shall be set up around the construction area as ground control points (GCPs), and they shall be evenly distributed in a ring on both sides of the bridge and within the main span. Each marker post shall be rigidly connected to a fixed foundation. Its three-dimensional coordinates shall be determined by a total station and GNSS joint measurement, and used for point cloud spatial registration and time series unification. S103. High color rendering index LED lights are arranged on both sides of the arch rib and the support. The brightness of the lights is controlled synchronously with the data acquisition instrument. A light shield or sunshade is installed above each data acquisition instrument to ensure consistent illuminance during day and night data acquisition and to avoid direct sunlight and noise. S104. Set up a central control host, which connects to each data acquisition instrument via wired Ethernet or industrial-grade wireless network. The main control system triggers scanning at intervals of 3 minutes during the pouring stage and 10 minutes during the regular monitoring stage, and records the timestamp, equipment number and attitude parameters to achieve precise alignment of time-series point clouds. S105. Each data acquisition unit integrates a temperature and humidity sensor and an automatic calibration unit in its casing. Before data acquisition, it performs equipment self-test, lens cleaning test, and laser intensity adjustment. The internal inertial measurement unit (UTC timestamp IMU) compensates for equipment micro-vibrations or attitude changes in real time, reducing environmental disturbance errors. S106. After on-site debugging, the optimal observation range of the system was determined, and a local high-density scanning mode was enabled for the outer contour of the arch rib, the support columns, and the tie rod nodes, so that the original point cloud density was ≥1×10⁻⁶. 6 Points per square meter (m²) are transmitted in real time to a central server and automatically backed up, providing input for subsequent point cloud processing and digital image correlation analysis.

3. The method according to claim 1, characterized in that, Step S2, point cloud processing and image projection, specifically includes: S201. For the original point cloud of each acquisition cycle, execute the statistical outlier removal algorithm and radius neighborhood filtering in sequence: calculate the distance from the point to the neighborhood mean in the k-nearest neighbor. If the distance is greater than 2σ, delete it and remove isolated clusters with fewer than the set number of points, so that the point cloud density is uniform and the structural surface features are intact after cleaning. S202. Using the point cloud at the first time step as the reference, the ground control point GCP is used as the initial transformation, and then the nearest point ICP is iterated for fine registration. The iteration continues until the mean square error between two adjacent points is ≤0.1 mm, thus completing the unification of the time-series point cloud coordinates. S203. Project the registration point cloud along the normal of the outer surface of the structure onto a 2 mm × 2 mm regular grid, calculate the center elevation of each grid using the weighted average method, reconstruct the DEM of the arch rib, arch foot and support column in blocks, and form the overall DEM by splicing the boundaries. S204. Set the virtual light source azimuth angle to 45°, elevation angle to 60°, and cosine attenuation light intensity model in the DEM coordinate system to generate a 16-bit grayscale Hillshade image; S205. Project the same DEM twice, using a vertical viewing angle of 90° and an oblique viewing angle of 45° respectively. Use bilinear interpolation to maintain pixel equidistance and obtain two sets of images: one for planar displacement using Hillshade(90°) and the other for composite displacement using Hillshade(45°). S206. Set the Hillshade ground pixel resolution to be consistent with the DEM grid spacing, that is, each pixel represents 2mm of actual length. The image size is automatically cropped so that the arch rib and the key area of ​​the support are located in the center, and the pixel grayscale depth is 16 bits. S207. Using the sum of image grayscale variance and edge gradient as the evaluation function, automatically optimize the azimuth and elevation angles of the light source to maximize the evaluation function and enhance the contrast of surface micro-textures. S208. All time-series Hillshade images use the same projection matrix, light source vector, and grayscale mapping scale, and the parameters are written to the metadata file for consistent use by the DIC program; S209. For DEM holes caused by temporary occlusion or reflection, if the missing area is ≤50 mm×50 mm, the weighted average elevation value of adjacent valid points is used to fill the hole; if it is >50 mm×50 mm, it is marked as an occlusion area and excluded in the subsequent DIC analysis, thereby obtaining a high-precision, low-noise, and time-consistent multi-angle Hillshade image sequence.

4. The method according to claim 1, characterized in that, Step S3, digital image correlation (DIC) analysis, specifically includes: S301. For Hillshade images at adjacent times with the same projection angle, a phase correlation algorithm is used for sub-pixel level registration. The mean square error test shows that the overall misalignment is ≤0.2 pixels, and the images are cropped to the same image size. S302. Divide the registered image into 150×150 pixel correlation windows with a step size of 50 pixels and 50% overlap. The center of each window forms a measurement node. S303. Using the zero-mean normalized cross-correlation (ZNCC) function as the matching criterion, the maximum value of the correlation coefficient is found within the search area to obtain the integer pixel displacement; S304. For integer displacements, a quadratic polynomial surface fitting is used to fit the relevant peak neighborhood, and the extreme points are used as sub-pixel displacements. S305. Bicubic spline interpolation is used to reconstruct the displacement of each node into a continuous planar displacement field, and median filtering is performed using the local mean ±3σ threshold to remove outlier vectors, thus obtaining (Δx,Δy) from the vertical perspective. S306. Repeat S301-S305 for the Hillshade image with a 45° tilted viewpoint to obtain the composite displacement ΔY′, and invert the vertical displacement Δz according to Δz=ΔY′ / tan 45° to synthesize the three-dimensional displacement vector (Δx,Δy,Δz) of each pixel node; S307. Kalman filtering is used to smooth the temporal three-dimensional displacement field to suppress instantaneous jumps and random noise; S308. Real-time statistics of subset matching success rate. If <95%, automatically expand the window or search range and recalculate until the requirements are met. Finally, output continuous time-series 3D displacement cloud map and key point displacement curve.

5. The method according to claim 1, characterized in that, Step S4, 3D displacement calculation and visualization, specifically includes: S401. The planar displacement (Δx, Δy) and vertical displacement Δz output by DIC are transformed inversely according to the recorded projection matrix and light source vector to complete the unified mapping of pixel-geometric coordinates, so that each pixel displacement corresponds one-to-one with the DEM grid node. S402. Calculate the three-dimensional displacement vector and its resultant displacement of each mesh node. And retain the component sign to determine the direction of upward arching, downward deflection or horizontal offset; S403. Inverse distance weighting and bicubic spline hybrid interpolation are used to reconstruct the spatial continuity of discrete node displacements, forming a continuous three-dimensional displacement field covering the arch ribs, tie rods and support columns; S404. Based on a 3D graphics rendering engine, it overlays the DEM before and after deformation, uses cool-warm color gradients and vector arrows to express the magnitude and direction of displacement, and supports real-time interactive rotation and scaling. S405. Generate a deformation diagram of the difference between adjacent time points and a residual distribution diagram of the DIC-total station. The system verifies that the residual RMSE ≤ 0.8 mm, which meets the accuracy requirements for structural monitoring. S406. Automatically extract the X, Y, Z time-series displacement curves and combined displacement curves of the arch mid-span, arch foot, hanger node and support column bottom, calculate displacement rate and acceleration, and trigger an early warning if the rate increases monotonically three times in a row and exceeds the preset threshold. S407. An uncertainty assessment model is introduced, which integrates DIC matching, point cloud registration and sensor ranging errors, and calculates the overall uncertainty through error propagation analysis. The displacement results of each grid node are accompanied by a confidence interval label, which is convenient for use in subsequent safety assessments.

6. The method according to claim 6, characterized in that, Step S5, Deformation Trend Analysis and Safety Early Warning, specifically includes: S501. The three-dimensional displacement time series data of the same monitoring point are sorted by timestamp, and the displacement curve is smoothed by smoothing filter. The filter window is automatically adjusted according to the sampling frequency to maintain the trend and filter out high-frequency noise. S502. Calculate the rate of change of displacement between adjacent time points using the central difference method to obtain the velocity sequence; then differentiate the velocity sequence to obtain the acceleration curve; by analyzing the variation law of velocity and acceleration, identify the deformation characteristics of the arch rib and support at different stages of construction. S503. Establish a four-level multi-threshold discrimination model: ① Displacement threshold: set according to design specifications; ② Rate threshold: If the displacement rate of a key point exceeds the warning value for three consecutive observations, it is determined to be an abnormal acceleration. ③Acceleration threshold: Potential instability is determined when the acceleration suddenly jumps and continues to exceed twice the standard deviation of the mean; ④ Trend threshold: Using regression analysis and inverse velocity method extrapolation, an early warning is issued when the extrapolated curve approaches the limit state in a certain period of time in the future; The current structural state level is output by combining these parameters using a logical weighted algorithm. S504. Introducing a Bayesian update mechanism to correct the risk probability distribution of each monitoring point in real time and integrate it with historical data to improve the robustness of early warning; when the status of any monitoring point reaches the warning level, the system highlights the area on the 3D interface and pops up the over-limit parameters. When the danger level is reached, an audible and visual alarm is immediately issued, and an electronic notification containing the monitoring point number, coordinates, displacement, rate, acceleration, over-limit ratio and suggested handling measures is pushed to the terminals of the on-site supervisor, general contractor and safety manager through the construction monitoring platform. S505. Set up an automatic verification mechanism: For single-point anomalies, first check the correlation between the displacement of the neighborhood and the point cloud density. If there is no consistent deformation in the neighborhood or the data quality is low, it is judged as data anomaly and removed. Only after verification is it included in the early warning output, reducing the false alarm rate to ≤1%. S506. Establish an alarm event database to automatically record alarm time, monitoring point number, displacement rate and acceleration value, handling measures and results, support historical playback and statistical analysis, and realize a safety assessment with full traceability.

7. The method according to claim 1, characterized in that, Step S6, construction control and feedback, specifically includes: S601. A construction control interface module is set up at the main control terminal of the monitoring system, which is interconnected with the field automation control system and monitoring center through the standard industrial communication protocol to transmit three-dimensional displacement field, velocity curve and early warning level in real time; the communication link adopts bidirectional data verification and redundancy design. S602. The monitoring area is divided into several control units, each unit corresponding to a unique construction parameter, namely concrete pouring rate, layer thickness, support unloading sequence or tension force; when the displacement or rate of a certain unit exceeds the threshold set in S503, the system automatically retrieves the corresponding parameter and generates a digital adjustment command, which is then sent to the construction dispatch terminal after confirmation by the supervisor or on-site personnel. S603. After the control command is executed, the system continuously monitors the change in displacement rate: if the displacement rate decreases and then returns to stability after adjustment, the system automatically determines that the control is effective and records it as "positive feedback"; if the deformation continues to expand or the fluctuation intensifies after adjustment, the system determines that the control is ineffective and prompts for verification of construction measures. S604. Establish an integrated construction monitoring and control database to uniformly store point clouds, DIC results, displacement fields, early warning logs, construction parameters, and control commands. All data are accompanied by timestamps and digital signatures to form a complete time-series archive. The database supports multi-dimensional retrieval and visualization playback by construction stage, monitoring area, and event type, providing a traceable basis for project acceptance, quality assessment, and subsequent maintenance.

Images