A displacement detection method and system for structural health monitoring

Abstract

This invention discloses a displacement detection method and system for structural health monitoring. The displacement detection method includes the following steps: Step 1: Building a displacement detection system; Step 2: Image acquisition and preprocessing using the displacement detection system; Step 3: Spot recognition and center coordinate extraction of the preprocessed image; Step 4: Error compensation of the output of Step 3 to obtain compensated data; Step 5: Transmitting the compensated data to a cloud server for intelligent structural health analysis. This application supports high dynamic exposure control and adaptive image enhancement; the system possesses comprehensive anti-interference capabilities against strong light interference, shadow occlusion, thermal drift, vibration disturbance, and spot distortion; the filtering algorithm introduces a two-level error modeling mechanism to effectively filter out high-frequency jitter and low-frequency drift.

Description

Technical Field

[0001] This invention relates to the field of structural health monitoring and intelligent image processing for optical measurement, and particularly to a displacement detection method and system for structural health monitoring, applicable to online monitoring and safety early warning of important structures such as bridges, dams, wind turbine towers, and high-rise buildings. Background Technology

[0002] Currently, structural health monitoring (SHM) has gradually become a standard feature in infrastructure such as transportation, water conservancy, and energy. However, existing mainstream methods, such as fiber optic grating sensing, GPS measurement, laser ranging, and visual measurement, suffer from the following problems in practical applications:

[0003] • Precision is difficult to achieve at the sub-millimeter level;

[0004] • Sensitive to environmental factors such as strong light, vibration, and temperature drift, exhibiting poor robustness;

[0005] Especially in visual measurement, insufficient spot recognition accuracy and poor environmental adaptability are key bottlenecks restricting its practical application. Summary of the Invention

[0006] In view of the technical problems mentioned in the background section, the purpose of this invention is to provide a displacement detection method and system for structural health monitoring.

[0007] To achieve the objectives of this invention, the technical solution provided by this invention is as follows:

[0008] First aspect

[0009] This application provides a displacement detection method for structural health monitoring, comprising the following steps:

[0010] Step 1: Build a displacement detection system. The displacement detection system includes multiple displacement detection units. Each displacement detection unit includes a laser emission module and a spot processing module. The spot processing module includes a whiteboard, a high-resolution camera, a data acquisition terminal, and a mounting base.

[0011] Step 2: Image acquisition and preprocessing are performed using the displacement detection system. The system employs a high dynamic response image acquisition method that can adapt to rapid changes in ambient brightness in real time, an adaptive Gaussian filtering noise reduction algorithm, and a CLAHE local contrast enhancement algorithm to improve the visibility of light spots in non-ideal backgrounds.

[0012] Step 3: Perform spot recognition and center coordinate extraction on the preprocessed image;

[0013] Step 4: Perform error compensation on the output of Step 3 to obtain the compensated data;

[0014] Step 5: Transmit the compensated data to the cloud server and perform structural health intelligent analysis.

[0015] In step 2, the high dynamic response image acquisition method specifically involves the following: the high-resolution camera uses a built-in ambient light sensor to continuously sample and obtain the light intensity value of the current acquisition point, and introduces a parameter feedback adjustment mechanism based on a PID controller to dynamically control the key parameters of the camera.

[0016] In step 2, the adaptive Gaussian filtering noise reduction algorithm evaluates the local complexity of the image using structural tensor analysis.

[0017] In step 2, the CLAHE local contrast enhancement algorithm is as follows:

[0018] Step 2.1: Divide the image into small pieces;

[0019] Step 2.2: Perform histogram equalization independently for each block;

[0020] Step 2.3: Set the contrast limiting factor to prevent small blocks from excessively amplifying noise;

[0021] Step 2.4: Finally, bilinear interpolation is performed on the equalized image to form a globally smooth result.

[0022] Step 3 includes the following:

[0023] Step 3.1: Use adaptive thresholding to divide the input grayscale image into blocks, and calculate the dynamic threshold for each local region based on its average brightness;

[0024] Step 3.2: After the initial extraction of the contour of the spot region, multi-round contour feature filtering and geometric constraint enhancement are used;

[0025] Step 3.3: After initially identifying qualified light spots, perform sub-pixel level light spot center positioning;

[0026] Step 3.4: Use the lightweight convolutional neural network coordinate fine-tuning module to perform sub-pixel level coordinate fine-tuning on the preliminary recognition results.

[0027] In step 4, a two-level error compensation mechanism is adopted, with two complementary strategies designed to address short-term high-frequency noise and long-term slow-varying error drift, respectively.

[0028] Step 5 specifically includes the following:

[0029] Step 5.1: After error compensation is completed, the acquisition terminal packages the timestamp and displacement coordinate data and transmits them to the cloud server;

[0030] Step 5.2: Deploy a high-concurrency API interface service on the cloud server to receive the coordinate data stream uploaded in real time from each node; after parsing and caching, the received data is stored in a time-series database for long-term storage.

[0031] Meanwhile, the front-end visualization platform uses the WebSocket protocol to achieve second-level refresh and dynamically display the following content:

[0032] Real-time two-dimensional coordinate displacement graph: displays the trajectory of the light spot on the monitoring target surface;

[0033] Trend charts and fluctuation charts: plot the trends of xt and yt over a period of time to observe whether there are signs of periodic oscillations or gradual displacement;

[0034] Structural heat map: When multiple monitoring nodes are deployed simultaneously, the relative displacement of each region is displayed through color gradient, realizing a visualized structural deformation map;

[0035] Step 5.3: An alarm will be triggered when one of the following alarm conditions is met:

[0036] (1) The absolute value of the displacement exceeds the safety limit;

[0037] (2) The displacement slope increases abnormally within a certain time window;

[0038] When the above conditions are triggered, the following actions will be performed automatically:

[0039] (1) Push alarm notifications: Real-time alarms are sent via SMS, email or WeChat enterprise account;

[0040] (2) Start the video recording module: transmit the image and raw data at that time for remote expert judgment;

[0041] (3) Linkage control module: sends linkage signals to the field control system.

[0042] Second aspect

[0043] This application provides a displacement detection system for structural health monitoring. The displacement detection system includes multiple displacement detection units. Each displacement detection unit includes a laser emission module and a spot processing module. The spot processing module includes a whiteboard, a high-resolution camera, a data acquisition terminal, and a mounting base.

[0044] The displacement detection system performs the following displacement detection method, which specifically includes the following steps:

[0045] Step S1: Image acquisition and preprocessing are performed using the displacement detection system. The system employs a high dynamic response image acquisition method that can adapt to rapid changes in ambient brightness in real time, an adaptive Gaussian filtering noise reduction algorithm, and a CLAHE local contrast enhancement algorithm to enhance the discernibility of light spots in non-ideal backgrounds.

[0046] Step S2: Perform spot recognition and center coordinate extraction on the preprocessed image;

[0047] Step S3: Perform error compensation on the output of step S2 to obtain the compensated data;

[0048] Step S4: Transmit the compensated data to the cloud server and perform structural health intelligent analysis.

[0049] In step S1, the high dynamic response image acquisition method specifically involves the following: the high-resolution camera uses a built-in ambient light sensor to continuously sample and obtain the light intensity value of the current acquisition point, and introduces a parameter feedback adjustment mechanism based on a PID controller to dynamically control the key parameters of the camera.

[0050] In step S1, the adaptive Gaussian filtering noise reduction algorithm evaluates the local complexity of the image using the structural tensor analysis method.

[0051] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0052] 1. High-precision spot positioning capability

[0053] A composite center extraction algorithm (centroid method + Gaussian fitting + CNN fine-tuning) is used to achieve sub-pixel-level spot center localization; the CNN model introduces an offset correction learning mechanism to deeply compensate for errors caused by boundary distortion and uneven illumination; the measured coordinate positioning error is as low as 0.2 to 0.4 pixels, which is more than 60% better than traditional methods, and supports sub-millimeter displacement detection requirements.

[0054] 2. Extremely strong environmental adaptability

[0055] It supports high dynamic exposure control and adaptive image enhancement (structure tensor + CLAHE); the system has comprehensive anti-interference capabilities against strong light interference, shadow occlusion, thermal drift, vibration disturbance, and spot distortion; the filtering algorithm introduces a two-level error modeling mechanism (multi-window filtering + adaptive Kalman), which effectively filters out high-frequency jitter and low-frequency drift.

[0056] 3. Lightweight, modular, and highly integrated design

[0057] The computing module can be deployed on low-power edge devices such as Raspberry Pi / Jetson, with real-time inference latency of less than 20ms; each functional module is independent (laser, acquisition, filtering, algorithm, transmission), which is conducive to maintenance and functional expansion; it supports flexible networking and data transmission via USB, WiFi, 5G, and fiber optics.

[0058] 4. Cloud-based intelligent analysis and automated early warning mechanism

[0059] It combines a cloud platform with local terminals to achieve dual redundancy monitoring; the data supports trend modeling, structural heat map generation, and vibration spectrum analysis; the safety early warning supports rule setting, autonomous triggering, remote notification, and linkage control. Attached Figure Description

[0060] Figure 1 This is a schematic diagram of the displacement detection unit provided in this application;

[0061] In the diagram, 1 is the laser emission module, 2 is the light spot processing module, 3 is the whiteboard, 4 is the high-resolution camera, and 5 is the data acquisition terminal. Detailed Implementation

[0062] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0063] Existing laser spot center identification algorithms mostly employ simple centroid methods or image intensity weighting algorithms for center extraction. These methods are effective under ideal conditions, but in situations with direct sunlight, noise interference, or complex backgrounds, spot center identification is prone to shifting or failure, leading to a significant increase in displacement monitoring errors and making it difficult to simultaneously meet the requirements of high accuracy and high robustness.

[0064] To address the above problems, this application provides a displacement detection method for structural health monitoring, comprising the following steps:

[0065] Step 1: Build a displacement detection system. The displacement detection system includes multiple displacement detection units. Each displacement detection unit includes a laser emission module 1 and a spot processing module 2. The spot processing module 2 includes a whiteboard 3, a high-resolution camera 4, a data acquisition terminal 5, and a mounting base.

[0066] It should be noted that the detection unit mainly consists of two parts: a laser projection end and a receiving end.

[0067] • Laser emission module: Mounted on a stable bracket, it emits a stable and collimated laser beam;

[0068] The light spot processing module includes: a whiteboard (diffuse reflection target surface), a high-resolution camera, and a light filter.

[0069] The components include chip modules, shock absorption modules, USB data transmission modules, and data acquisition terminals.

[0070] Both modules are equipped with levels at both ends to ensure that the projection axis is aligned with the imaging axis and improve measurement stability.

[0071] Once the system is started, the laser emission module generates a stable, collimated monochromatic laser beam and precisely projects this beam along the optical axis onto the target area—the surface of a high-reflectivity whiteboard. The whiteboard, acting as a diffuse reflective medium, has a diffuse reflection coating that allows it to form a light spot pattern with clear boundaries and high intensity contrast under laser illumination, effectively improving the signal-to-noise ratio (SNR) for subsequent image recognition. After the light spot forms on the whiteboard surface, it is captured by a high-resolution CMOS image acquisition device installed within a specific angle and focal length range. This camera uses a bandpass filter to allow only light within the laser wavelength range (e.g., 650±10nm) to enter the photosensitive element, thereby suppressing interference from ambient background light and enhancing the laser feature information in the image. Throughout the acquisition process, the image acquisition system scans the light spot on the whiteboard in real time, outputting a high-contrast image sequence, providing a clear and stable input data source for subsequent spot positioning, displacement calculation, and error analysis. During this process, the system can also dynamically adjust the exposure time and gain to adapt to changes in external lighting, ensuring stable acquisition of high-quality images even under natural light or strong outdoor light conditions.

[0072] Step 2: Image acquisition and preprocessing are performed using the displacement detection system. The system employs a high dynamic response image acquisition method that can adapt to rapid changes in ambient brightness in real time, an adaptive Gaussian filtering noise reduction algorithm, and a CLAHE local contrast enhancement algorithm to improve the visibility of light spots in non-ideal backgrounds.

[0073] In this system, the image acquisition and preprocessing stage is responsible for converting the optical information of the light spot into high-quality digital image data, and is a key front-end module in the entire light spot recognition and displacement extraction chain. Compared with traditional methods, we have made systematic improvements to the image acquisition hardware configuration, dynamic parameter control mechanism, and image enhancement process to adapt to changing environmental conditions and improve the system's robustness and accuracy.

[0074] The high dynamic response image acquisition methods include the following:

[0075] Traditional image acquisition systems operate with fixed parameter settings and cannot adaptively adjust in real time to rapid changes in ambient brightness, leading to severe overexposure or underexposure and consequently affecting the accuracy of spot recognition. This system employs a built-in ambient light sensor (covering a wavelength range of 400–700 nm) to continuously sample and acquire the light intensity value L(t) at the current acquisition point. A parameter feedback adjustment mechanism based on a PID (proportional-integral-derivative) controller is introduced to dynamically control the camera's key parameters.

[0076] Exposure time Te: Automatically decreases in strong light to avoid image saturation;

[0077] Gain coefficient Ga: Increases in low-light environments to improve image brightness;

[0078] White balance parameter: Real-time fine-tuning of the red and blue channel ratio to enhance the prominence of the red laser spot.

[0079] The PID controller formula is as follows:

[0080]

[0081] Where e(t) = Ltarget - L(t), the target brightness is estimated from the image entropy or mean gray level.

[0082] To address the issue of blurred edges in complex texture regions caused by traditional fixed-kernel Gaussian filtering, we introduce a structure tensor analysis method to evaluate the local complexity of the image:

[0083] Calculate the local gradient tensor for image I(x,y):

[0084]

[0085] The tensor eigenvalues ​​are used to determine whether a region is an edge / texture region, and the standard deviation σ of the Gaussian kernel is dynamically adjusted.

[0086]

[0087] Where λmax(J) represents the gradient intensity of the principal direction in the local image.

[0088] This approach achieves an adaptive noise reduction effect that retains less blur in edge areas and enhances smoothness in texture areas, suppressing high-frequency noise while preserving the clear outline of the light spot edges.

[0089] To further enhance the visibility of the light spot against a non-ideal background, the CLAHE (Contrast Limited Adaptive Histogram Equalization) technique is introduced to perform local contrast enhancement on the image. The algorithm flow is as follows:

[0090] • Divide the image into small blocks (e.g., 8×8 or 16×16);

[0091] • Perform histogram equalization independently on each block;

[0092] • Set the contrast limiting factor Climit to prevent small blocks from being over-amplified with noise;

[0093] Finally, bilinear interpolation is performed on the equalized image to form a globally smooth result.

[0094] The CLAHE algorithm can significantly improve the brightness difference between the spot area and the background, enhancing the "detectability" of the laser point under background conditions such as strong light, dust, and dark walls.

[0095] Step 3: Perform spot recognition and center coordinate extraction on the preprocessed image;

[0096] In laser spot detection systems, accurately extracting the geometric center of the spot is a crucial step in displacement calculation. Traditional methods typically employ "image grayscale thresholding + contour extraction + geometric centroid calculation" to identify and locate the spot. While this approach is simple to implement and computationally efficient, it suffers from significant drawbacks in complex environments: the fixed threshold method is sensitive to changes in illumination, easily leading to the loss of spots in overexposed / underexposed areas; when background noise (such as reflections or dust) is strong in the image, non-spot areas may be misidentified; and when the spot boundary is non-circular or distorted, the calculated geometric centroid deviates from the true center, increasing measurement errors.

[0097] To address the aforementioned issues, this system constructs a hierarchical, modular, and integrated spot recognition scheme, which combines classic image algorithms with a lightweight deep learning model to ensure sub-pixel-level center localization even in complex environments.

[0098] 1. Adaptive Gaussian Thresholding

[0099] The input grayscale image is segmented using OpenCV's adaptiveThreshold method. A dynamic threshold is calculated for each local region based on its average brightness, effectively eliminating segmentation failures caused by uneven overall illumination. The basic principle is as follows:

[0100] T(x,y)=μ N ( x,y) -C

[0101] Where, μ N(x，y) Let N represent the mean value of the pixel's neighborhood, and C be an adjustment constant. This method ensures that the light spot can be effectively extracted under both bright and shadow conditions.

[0102] 2. Multi-round contour feature filtering and geometric constraint enhancement

[0103] To further eliminate false spots and high-brightness noise, this system designs a "three-dimensional comprehensive filtering algorithm" after the initial extraction of the spot region contour, which is based on the following indicators for identification:

[0104] Area A: The spot should have a medium area in the image; too small is noise, and too large is spill light.

[0105] Circularity:

[0106]

[0107] Where L is the perimeter of the outline. A roundness close to 1 indicates an approximate perfect circle; if the roundness is less than 0.7, it is excluded.

[0108] Edge gradient strength: The Sobel operator is used to evaluate edge sharpness. If the edge is too blurry or the gradient is discontinuous, it is considered an invalid contour.

[0109] 3. Subpixel-level spot center positioning

[0110] After initially identifying qualified light spots, the center coordinates were extracted using a combination of two methods:

[0111] Image geometric moments centroid method

[0112] Calculate image geometric moments based on pixel intensity of the region image:

[0113]

[0114] Among them, M pq Let be the (p+q)th order moment of the image. This method is fast and efficient, and is suitable for light spots with regular boundaries.

[0115] Gaussian distribution fitting

[0116] When the light spot shape is good and the light intensity distribution is approximately symmetrical, a two-dimensional Gaussian function model is used to fit the light spot to improve positioning accuracy.

[0117]

[0118] The (x0, y0) parameter in the fitting parameters represents the center position of the spot intensity maximum, providing sub-pixel resolution. This algorithm is poorly robust to background interference and is suitable for clean image scenes. It is used in conjunction with the centroid method, and the results are compared to obtain a reliable value.

[0119] 4. Lightweight Convolutional Neural Network (CNN) Coordinate Fine-tuning Module

[0120] To further improve the accuracy of spot center coordinate recognition, especially in complex environments such as strong sunlight, dust, and reflections, this application designs and deploys a lightweight convolutional neural network (CNN) model based on the traditional centroid method and Gaussian fitting. This model is used for sub-pixel-level coordinate fine-tuning of the initial recognition results. This module can be embedded into low-power devices (such as Raspberry Pi 4B and Jetson Nano) for real-time operation, balancing accuracy and speed.

[0121] The overall implementation of this module consists of the following five logical steps:

[0122] (1) Model goal setting and function definition

[0123] The core task of this CNN model is to predict the offset (Δx, Δy) of a patch region extracted from an image relative to the traditional recognition result, thereby accurately correcting the results of the centroid method or Gaussian fitting. The final center coordinates are represented as:

[0124] (x final ,y final )=(x raw +△x, y raw +△y)

[0125] This design effectively overcomes the problem of large recognition deviations in traditional methods under conditions of irregular light spot edges and non-uniform illumination.

[0126] (2) Input data construction and preprocessing strategies

[0127] The network input is a 64×64 grayscale image patch, which is cropped from the original image and centered at the initial center coordinates. To improve the model's adaptability to various interference conditions, the following image augmentation operations are added to the training set:

[0128] • Add Gaussian noise (σ = 5~20);

[0129] • Perform brightness perturbation (±15%);

[0130] • Perform affine transformations (rotation, scaling, translation);

[0131] • All images are uniformly normalized to the [0,1] interval.

[0132] These enhancements help the model learn a "robust compensation capability" against factors such as spot deformation, offset, and background noise.

[0133] (3) Model structure design and parameter configuration

[0134] This network employs a concise and efficient seven-layer structure, primarily consisting of two convolutional-pooling combined layers, one flattening layer, and two fully connected layers. The structure of each layer is as follows:

[0135]

[0136]

[0137] The entire model has approximately 29,762 parameters, and the trained model file is less than 100KB, making it easy to deploy on embedded devices.

[0138] (4) Training methods and optimization strategies

[0139] To improve model accuracy and generalization ability, the training process uses the following configuration:

[0140] • Data source: Constructed from real-world shooting data and simulated images, all with the actual center coordinates manually annotated;

[0141] • Loss function: Use mean squared error (MSE) as the regression objective function:

[0142]

[0143] • Optimizer: Adam, with an adaptive learning rate initially set to 1e-4;

[0144] • Training mechanism: The Early Stopping strategy is used to prevent overfitting. With 3000 images and 20 rounds of iterations, the final residual error on the validation set is less than 0.2px.

[0145] (5) Embedded deployment performance and actual results

[0146] After model compression and acceleration optimization, the performance on edge devices is as follows:

[0147] • Deployment platforms: Raspberry Pi 5, Jetson Nano;

[0148] • Inference time: The processing time for a single frame patch is 13-18ms;

[0149] • Improved accuracy:

[0150] Traditional centroid method residual: 0.6–1.1px

[0151] CNN fine-tuned residuals: 0.2–0.4px

[0152] The average error was reduced by more than 60%.

[0153] Step 4: Perform error compensation on the output of Step 3 to obtain the compensated data;

[0154] In actual operation, the positioning result of the laser spot center is inevitably affected by a variety of external and internal system disturbances, including:

[0155] • Slight vibration of the equipment (such as vibration of the support frame or ground vibration);

[0156] • Minor mechanical deformation (thermal drift) of the camera platform due to temperature fluctuations;

[0157] • Fluctuations in image brightness caused by drastic changes in external ambient light;

[0158] • Sudden noise interference (such as fan airflow, dust obstruction).

[0159] To address these interference sources, this application proposes a two-level error compensation mechanism, designing two complementary strategies to target short-term high-frequency noise and long-term slow-varying error drift:

[0160] 1. Multi-Window Adaptive Moving Average Filter: To improve the balance between real-time performance and smoothness, this system no longer uses a single-window-size averaging filter, but instead employs a multi-window parallel evaluation mechanism. The specific implementation is as follows:

[0161] At each time t, the system uses different window lengths (n = 3, 5, 7) to represent the center coordinate sequence x of the current time series. t Perform a moving average:

[0162]

[0163] Calculate the local variance for the three window outputs:

[0164]

[0165] The window with the smallest variance is selected as the current compensated coordinate value to avoid delayed response due to an excessively long window, while also enhancing adaptability to short-term disturbances.

[0166] 2. Improved Kalman Filtering (Adaptive Kalman Filtering with SensorFusion)

[0167] Traditional Kalman filtering assumes that process noise Q and observation noise R are constant, which makes it difficult to guarantee filtering accuracy in real nonlinear and non-stationary environments. To address this, this application fuses Kalman filtering with multi-source sensor inputs and introduces a noise dynamic estimation module to achieve adaptive adjustment.

[0168] Filtering process:

[0169] Prediction phase:

[0170]

[0171]

[0172] Update phase:

[0173]

[0174]

[0175]

[0176] in:

[0177] · Estimated coordinates for the current frame;

[0178] ·Z k : Current frame measurement coordinates (CNN / centroid output);

[0179] ·P k Covariance matrix;

[0180] ·Q k R k Process / measurement noise covariance, dynamically adjusted.

[0181] Noise dynamic estimation method:

[0182] The rate of change of light intensity (from the ambient light sensor) is used as R k Adjustment factor, increase R during periods of high volatility. k To suppress observational instability caused by brightness fluctuations;

[0183] The accelerometer outputs mean and variance (used to detect whether the system is vibrating) to dynamically update Q. k This allows for more accurate modeling of system state changes.

[0184] Step 5: Transmit the compensated data to the cloud server and perform structural health intelligent analysis.

[0185] (1) Edge-to-Cloud Transmission Module

[0186] The data acquisition terminal (such as Raspberry Pi or Jetson Nano) has local caching and network communication capabilities. After error compensation is completed, it packages the timestamp and displacement coordinate data and transmits it to the cloud via any of the following links:

[0187] 5G communication module: suitable for high-speed dynamic monitoring scenarios (such as bridge vibration, mobile platforms);

[0188] Wi-Fi LAN: Suitable for areas with existing network coverage, such as tunnels, dams, and factory areas;

[0189] Fiber optic Ethernet: Suitable for fixed monitoring points, featuring high bandwidth and low latency.

[0190] The data is packaged in standard JSON format, with an example content as follows:

[0191]

[0192] To ensure data integrity and security, a TLS encrypted channel is used and a retransmission mechanism is configured to prevent packet loss.

[0193] (2) Cloud-based data receiving and visualization platform

[0194] The cloud server deploys a high-concurrency API interface service (such as based on Flask / FastAPI + Redis caching) to receive real-time coordinate data streams uploaded by each node. The received data is parsed, cached, and then stored in the system.

[0195] Sequential databases (InfluxDB, TimescaleDB) are used for long-term storage.

[0196] Meanwhile, the front-end visualization platform uses the WebSocket protocol to achieve second-level refresh and dynamically display the following content:

[0197] • Two-dimensional coordinate displacement graph (real-time): Displays the trajectory of the light spot on the monitoring target surface;

[0198] Trend charts and fluctuation charts: plotting x over a period of time. t ,y t Observe the changing trends and whether there are signs such as periodic vibrations or gradual displacement;

[0199] • Structural heat map (multi-point system): When multiple monitoring nodes are deployed simultaneously, the relative displacement of each region is displayed through color gradient, realizing a visualized structural deformation map.

[0200] (3) Safety threshold alarm and linkage mechanism

[0201] The system allows users to customize alarm logic, such as setting threshold conditions like the following:

[0202] • The absolute value of the displacement exceeds the safety limit (e.g., >3mm);

[0203] • The displacement slope increases abnormally within a certain time window;

[0204] Once the above conditions are triggered, the system will automatically perform the following operations:

[0205] • Push alarm notifications: Real-time alerts via SMS, email, or WeChat Enterprise Account;

[0206] • Activate the video recording module: transmit the images and raw data from that time for remote expert evaluation;

[0207] • Linkage control module (optional): Sends linkage signals to the field control system (such as stopping operation, closing valves, issuing audible and visual alarms, etc.).

[0208] This application is applicable to the following scenarios: Bridge deformation measurement: analyzing the stress state of bridges to assist in maintenance decisions. Tunnel structure monitoring: measuring geological deformation to prevent collapse accidents. Dam settlement monitoring: detecting minute displacements on the dam surface and providing early warnings of crack risks. Mine surface deformation monitoring: monitoring surface subsidence caused by mining. Wind turbine vibration monitoring: analyzing the stress on wind turbine blades to improve equipment lifespan. High-rise building health monitoring: long-term monitoring of building structural deformation to improve safety.

[0209] In addition, this application provides a displacement detection system for structural health monitoring. The displacement detection system includes multiple displacement detection units. Each displacement detection unit includes a laser emission module and a spot processing module. The spot processing module includes a whiteboard, a high-resolution camera, a data acquisition terminal, and a mounting base.

[0210] The displacement detection system performs the following displacement detection method, which specifically includes the following steps:

[0211] Step S1: Image acquisition and preprocessing are performed using the displacement detection system. The system employs a high dynamic response image acquisition method that can adapt to rapid changes in ambient brightness in real time, an adaptive Gaussian filtering noise reduction algorithm, and a CLAHE local contrast enhancement algorithm to enhance the discernibility of light spots in non-ideal backgrounds.

[0212] Step S2: Perform spot recognition and center coordinate extraction on the preprocessed image;

[0213] Step S3: Perform error compensation on the output of step S2 to obtain the compensated data;

[0214] Step S4: Transmit the compensated data to the cloud server and perform structural health intelligent analysis.

[0215] Finally, it should be noted that the above embodiments are merely illustrative and explanatory of the present invention, and are not intended to limit the present invention to the scope of the described embodiments. Furthermore, those skilled in the art will understand that the present invention is not limited to the above embodiments, and many more variations and modifications can be made based on the teachings of the present invention, all of which fall within the scope of protection claimed by the present invention.

Claims

1. A displacement detection method for structural health monitoring, characterized in that, Includes the following steps: Step 1: Build a displacement detection system. The displacement detection system includes multiple displacement detection units. Each displacement detection unit includes a laser emission module and a spot processing module. The spot processing module includes a whiteboard, a high-resolution camera, a data acquisition terminal, and a mounting base. Step 2: Image acquisition and preprocessing are performed using the displacement detection system. The system employs a high dynamic response image acquisition method that can adapt to rapid changes in ambient brightness in real time, an adaptive Gaussian filtering noise reduction algorithm, and a CLAHE local contrast enhancement algorithm to improve the visibility of light spots in non-ideal backgrounds. Step 3: Perform spot recognition and center coordinate extraction on the preprocessed image; Step 4: Perform error compensation on the output of Step 3 to obtain the compensated data; Step 5: Transmit the compensated data to the cloud server and perform structural health intelligent analysis; In step 2, the high dynamic response image acquisition method specifically involves: the high-resolution camera employs a built-in ambient light sensor to continuously sample and acquire the light intensity value at the current acquisition point, and introduces a parameter feedback adjustment mechanism based on a PID controller to dynamically control the key parameters of the camera; In step 2, the adaptive Gaussian filtering noise reduction algorithm evaluates the local complexity of the image using the structural tensor analysis method; In step 2, the CLAHE local contrast enhancement algorithm proceeds as follows: Step 2.1: Divide the image into small pieces; Step 2.2: Perform histogram equalization independently for each block; Step 2.3: Set the contrast limiting factor to prevent small blocks from excessively amplifying noise; Step 2.4: Finally, perform bilinear interpolation fusion on the equalized image to form a globally smooth result; Step 3 includes the following steps: Step 3.1: Use adaptive thresholding to divide the input grayscale image into blocks, and calculate the dynamic threshold for each local region based on its average brightness; Step 3.2: After the initial extraction of the contour of the spot region, multiple rounds of contour feature filtering and geometric constraint enhancement are used to eliminate false spots and high-brightness noise; Step 3.3: After initially identifying qualified light spots, perform sub-pixel level light spot center positioning; Step 3.4: Use a lightweight convolutional neural network coordinate fine-tuning module to perform sub-pixel level coordinate fine-tuning on the preliminary recognition results; In step 4, a two-level error compensation mechanism is adopted, with two complementary strategies designed to address short-term high-frequency noise and long-term slow-varying error drift, respectively.

2. The displacement detection method for structural health monitoring according to claim 1, characterized in that, Step 5 specifically includes the following: Step 5.1: After error compensation is completed, the acquisition terminal packages the timestamp and displacement coordinate data and transmits them to the cloud server; Step 5.2: Deploy a high-concurrency API interface service on the cloud server to receive the coordinate data stream uploaded in real time from each node; after parsing and caching, the received data is stored in a time-series database for long-term storage. Meanwhile, the front-end visualization platform uses the WebSocket protocol to achieve second-level refresh and dynamically display the following content: Real-time two-dimensional coordinate displacement graph: displays the trajectory of the light spot on the monitoring target surface; Trend charts and fluctuation charts: plot the trends of xt and yt over a period of time to observe whether a cycle exists. Sexual vibration, signs of gradual displacement; Structural heat map: When multiple monitoring nodes are deployed simultaneously, the relative displacement of each region is displayed through color gradient, realizing a visualized structural deformation map; Step 5.3: An alarm will be triggered when one of the following alarm conditions is met: (1) The absolute value of the displacement exceeds the safety limit; (2) The displacement slope increases abnormally within a certain time window; When the above conditions are triggered, the following actions will be performed automatically: (1) Push alarm notifications: Real-time alarms via SMS, email or WeChat enterprise account; (2) Start the video recording module: transmit the image and raw data at that time for remote expert judgment; (3) Linkage control module: sends linkage signals to the field control system.

Images