A structural modal identification method and apparatus that integrates 3D point cloud reconstruction and vibration displacement identification
By integrating multi-view vision 3D point cloud reconstruction and monocular vision vibration displacement recognition technologies, the problems of high cost and discrete modal information in traditional structural health monitoring are solved, and low-cost, convenient and efficient acquisition of structural modal information is achieved.
Patent Information
- Application Number
- CN202410741387.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-11
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2044-06-11
AI Technical Summary
Traditional structural health monitoring methods require the deployment of numerous sensors on the structure, which is costly and results in high dispersion of modal information.
By combining multi-view vision-based 3D point cloud reconstruction technology with monocular vision-based vibration displacement recognition technology, non-contact acquisition of structural modal information is achieved through computer vision, and modal stitching technology is used to obtain the modal information of the entire structure.
It achieves low-cost, convenient and efficient acquisition of structural modal information, eliminates the time-consuming and labor-intensive sensor deployment, and can improve the modal vibration modes from sparse points to dense points, with high identification quality.
Smart Images

Figure CN118762056B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of engineering structure health monitoring technology, specifically to a structural modal identification method and device based on three-dimensional point cloud reconstruction technology and monocular vision vibration displacement identification. Background Technology
[0002] Structural health monitoring plays a crucial role in ensuring the safety of engineering structures such as bridges. Accurate acquisition of modal parameters is essential for understanding the vibration characteristics of a structure and assessing its performance. These modal parameters, including the structure's modal frequencies and mode shapes, directly reflect the structure's dynamic response under external excitation. Therefore, monitoring bridge structures and identifying their modal parameters is vital for ensuring their normal operation, and efficient methods for acquiring modal parameters have been a focus of research in the field of structural health monitoring.
[0003] Currently, there are various modal parameter extraction methods in the field of structural health monitoring, which can be simply classified according to two modules: data acquisition and data processing. In engineering, there are several data processing methods commonly used for modal parameter identification. In 1985, Juang and Pappa proposed a feature system implementation algorithm to identify modal parameters using impulse response signals; Overschee and Moor proposed a random subspace method in 2012 to identify modal parameters using white noise excitation response; Qu et al. proposed in 2019 to reduce environmental noise in the frequency domain using the concept of transfer function and convert the transfer function into an impulse response function for modal identification; and Yao et al. proposed in 2018 to identify modal parameters using a blind source separation framework. Regarding data acquisition methods, most existing research involves deploying a limited number of sensors on the target structure to acquire the structural response. For example, Ku et al. used accelerometer-acquired acceleration data to identify the modal parameters of a structure in the presence of harmonic and white noise excitation; Kaynardag et al. used displacement data obtained by a laser Doppler vibrometer to identify the vibration bending mode of a railway track; He et al. used monitoring data provided by the Global Navigation Satellite System (GNSS) to identify the multimodal natural frequencies of long-span bridges; and Pieraccini et al. used the microwave interferometric radar IBIS-S to measure the modal frequency parameters of a railway bridge under train loads. However, traditional methods often require the manual deployment of a large number of sensor devices on the structure, which is time-consuming, labor-intensive, and costly. Moreover, when acquiring modal information, the constructed modal information suffers from discreteness due to the data source being a limited number of sensor deployment points. The introduction of computer vision technology provides a new solution to this problem. By integrating existing modal extraction techniques with visual measurement techniques, dynamic response information of bridge structures can be acquired more comprehensively and continuously. Summary of the Invention
[0004] This invention aims to provide a structural modal identification method and device that integrates three-dimensional point cloud reconstruction and vibration displacement identification, solving the problems of high cost, time and labor consumption, and high dispersion of modal information in traditional multi-point monitoring.
[0005] The technical solution of this invention:
[0006] This paper integrates multi-view vision-based 3D point cloud reconstruction technology with monocular vision-based vibration displacement tracking technology to achieve non-contact acquisition of structural modal information through computer vision. First, multi-view vision-based 3D point cloud reconstruction technology is used to obtain a 3D point cloud model of the entire structure, yielding the spatial coordinates of visible points. Then, monocular vision-based vibration displacement recognition technology is integrated to measure the structure region by region, acquiring the displacement time history of all target points within the camera's field of view in each region. Finally, based on the previously measured spatial coordinates and time history responses of the target points, modal stitching technology is used to stitch the entire bridge modally through the overlapping points of adjacent regions, obtaining the modal information of the entire structure, including frequencies and mode shapes.
[0007] A structural modal identification method that integrates 3D point cloud reconstruction and vibration displacement identification, comprising the following steps:
[0008] Step 1: Obtain a 3D point cloud model of the entire structure using multi-view vision-based 3D reconstruction technology.
[0009] (1) Image data acquisition of the target structure. Take pictures of the target structure from different angles. Shooting requirements: First, avoid using shooting modes that produce large image distortion, such as wide-angle lenses, during the shooting process; second, during multi-angle shooting, the camera's trajectory should be centered on the target structure and move in a circular motion around the target structure, that is, the distance between the position of the picture taken at each angle and the target structure should be kept as consistent as possible, and the change of angle each time should not be too large; finally, ensure the image quality as much as possible, keep the exposure of each picture as consistent as possible, cover almost the entire target structure, and pay attention to the effects of occlusion and strong light.
[0010] (2) Reconstruct the target structure using the acquired image data. The acquired target image is processed using multi-view vision-based 3D reconstruction technology, sequentially performing three calculations: sparse point cloud reconstruction, depth information acquisition, and dense point cloud reconstruction, to obtain a 3D point cloud model of the target structure. A schematic diagram of the overall process is shown below. Figure 1The acquired disordered images are sequentially processed through feature point extraction, feature point matching, triangulation, pose estimation, and iterative optimization. The depth information of the original images is recovered by sequentially processing the matching cost, cost accumulation, depth estimation, and depth map optimization. Based on optical and geometric consistency, the sparse point cloud model of the target building structure is filled, fitted, and optimized.
[0011] (3) Calibration of the actual size of the 3D point cloud model based on the scaling factor method. Two calibration points are determined, which must be clearly visible in the dense point cloud model of the target structure constructed in step (2). Then, the lengths of the calibration line segments formed by the two calibration points are measured on the actual target structure and the point cloud model structure, respectively. The ratio of the two length measurements is the scaling factor of the dense point cloud model of the target structure. Subsequent measurement results can be converted from the point cloud model coordinate system to the actual structure coordinate system through this scaling factor.
[0012] Step 2: Divide the target structure into regions based on the measurement conditions.
[0013] (4) Divide the target structure into regions based on the measurement conditions. The region division is essentially determined by the required measurement accuracy and hardware parameters:
[0014] The theoretical accuracy of the displacement recognition algorithm can be expressed by the following formula:
[0015]
[0016] Where s is the sub-pixel accuracy of the displacement recognition algorithm; p is the pixel size; f is the lens focal length; and D is the measurement distance. When both the displacement recognition algorithm and hardware parameters are determined, the displacement measurement accuracy A will depend solely on the measurement distance D, and the two are directly proportional. According to the imaging principle, the camera's field of view can be expressed by the following formula:
[0017]
[0018] Where SS represents the photosensitive area of the camera sensor. When the hardware parameters are fixed, the camera's field of view (FOV) depends solely on the measurement distance D, and the two are directly proportional. Therefore, using the above formula, the region division should be based on the required accuracy, determining the measurement distance and thus the camera's FOV. Each divided region should ensure it does not exceed the camera's FOV. When dividing the target structure into regions, each adjacent region must contain an overlapping portion, and this overlapping portion should contain at least one usable common target point.
[0019] Step 3: Use monocular vision-based displacement recognition technology to obtain the displacement response of the target point in each region.
[0020] (5) Set up the instrument and record video for each area of the target under the specified excitation. Fix the camera shooting position, adjust the camera's field of view according to the previously set target division area to ensure that the entire measurement area is included in the camera frame, and focus the lens to ensure that the target is clearly visible in the camera image. Under the specified excitation, record video for a preset duration for that area, and save the video data for subsequent displacement recognition processing, until the video data of the entire area, i.e., the entire target structure, is acquired.
[0021] (6) Use a target tracking algorithm based on monocular vision to process the video data recorded in each region and obtain the displacement time history of all target points in each region.
[0022] (6.1) Selecting the Region of Interest (ROI). The essence of selecting the ROI is to reduce the image area for the algorithm to search for the target. A portion of the original image is selected as the ROI by manual selection. The ROI should at least contain the entire motion trajectory of the target during the entire time process. That is, it is to outline an approximate area that can cover the target displacement. The algorithm will only search for the target in this area.
[0023] (6.2) Selecting a tracking target template. The target template is selected from artificial target points or natural target points on the target structure that are pre-set on the target structure. The selected target template should have the same motion as the target structure being measured, that is, the target motion selected based on the visual algorithm can represent the motion of the specified target point on the structure, and the two should not have relative displacement during the visual monitoring process. The selected target template should be clearly visible in the image, have features that can be distinguished from the content of other images, have no excessive deformation or external occlusion during the motion, and have no excessively similar regions within the ROI to which the target belongs.
[0024] (6.3) Select a known distance segment for size calibration. By manually selecting a known distance segment in the image and inputting its actual size, the algorithm will automatically calculate the pixel size of the selected distance and then calculate the scaling factor of the image, realizing the conversion from pixel displacement to actual displacement. The acquisition of a known size segment in the image can be achieved based on the 3D point cloud model of the target structure obtained in step one.
[0025] (6.4) Target tracking based on template matching algorithm and sub-pixel registration based on image centroid method, outputting the displacement time history of the selected target point. The template matching algorithm searches for the region with the highest similarity to the target template within the ROI. The similarity is evaluated by calculating a specified objective function. A zero-mean normalized cross-correlation function is used as the objective function for similarity search. It calculates the correlation between template intensity and the overlapping part of the image; a higher correlation value indicates higher similarity, and it is unaffected by changes in scale or illumination shift. The objective function is shown below:
[0026]
[0027] Where x and y represent the pixel coordinates of the original image, x' and y' represent the pixel coordinates of the template image, R represents the image correlation, T represents the template image function, and I represents the original image function. After performing a similarity search on the ROI region, the pixel coordinates corresponding to the location with the maximum image correlation (R value) are the target integer pixel coordinates obtained by the template matching algorithm.
[0028] By using a template matching algorithm to search for the target within the ROI region, the current integer-pixel coordinates of the target in each frame of the image can be calculated. At this point, the accuracy limit of the displacement is the actual size represented by one pixel. Combined with the image centroid method, the centroid of the target image in each frame is calculated. The pixel neighborhood of the target's integer-pixel coordinates locked by the template matching algorithm is fitted. The neighborhood expansion width is preferably 1 / 10 of the size of the target template in the direction selected in step (6.2). Based on the image grayscale, the centroid mean of the target and its neighboring pixels is calculated. The finally obtained centroid coordinates are the sub-pixel coordinates of the target in each frame of the image.
[0029] Using the above method, all target points selected from all previously acquired regional videos are processed sequentially, and then the displacement time history of all measured target points of the entire target structure with sub-pixel accuracy is output.
[0030] Step 4: Obtain modal information of the entire structure using modal stitching technology.
[0031] (7) Based on the previously measured spatial coordinates and displacement response of all target points, modal splicing technology is used, that is, the common target points in the overlapping part of each adjacent region are used as the bridge for splicing the two regions. After obtaining the mode shape of each measurable modal frequency of each region of the target structure, the amplitude and direction of the displacement of the common target point between the two adjacent regions are kept consistent, and each region is assigned its own consistency coefficient to ensure that the vibration ratio coefficient of each region of the whole structure is consistent. Then, all regions of the whole structure are spliced modally to obtain the modal information of the whole structure, including frequency, mode shape, etc.
[0032] 2. A dynamic displacement monitoring device based on a monocular vision template matching algorithm, comprising:
[0033] The video acquisition module is used to acquire raw video data for dynamic displacement monitoring of the target structure. It includes a camera module, a fixed-focus lens, and a camera tripod for fixing the equipment, and is connected to the processor via a ribbon cable.
[0034] The processor, using a microcontroller as a microcomputer, executes computer programs. When the computer program is executed, the processor is used to: read the vibration video of the target structure acquired by the video acquisition module; provide an ROI and target selection page to facilitate users in specifying the range and target for displacement recognition; provide a calibration size selection page and size input line, automatically calculate the scaling factor and store it for subsequent displacement output scaling conversion; run a target tracking program based on template matching algorithm and a sub-pixel registration program based on image centroid method, output the displacement time history curve of the selected target point through the display screen of the output display module, and automatically store the displacement data.
[0035] The output display module is used for human-computer interaction to select ROI, target and calibration dimensions, and to output visualized images of the monitoring site. It includes a touch screen and a wireless mouse and keyboard, and connects to the core processor via HDMI and USB interfaces.
[0036] A schematic diagram of the dynamic displacement recognition device based on monocular vision template matching algorithm is shown below. Figure 2 As shown.
[0037] The beneficial effects of this invention are:
[0038] The technology and device proposed in this invention can conveniently acquire structural modal information. The entire process, from acquiring the spatial coordinates of the target point to measuring the displacement of the target point and constructing the coordinates of the mode shape, is based on non-contact measurement using computer vision technology, eliminating the time-consuming and labor-intensive process of deploying sensors on the structure. Only one camera is needed to complete the task, which is much cheaper than a large number of sensor instruments and equipment. Furthermore, the monocular vision displacement measurement method can, in principle, track an infinite number of target points within the field of view, no longer limited to the number of target points deployed by a finite number of sensors. It can improve the mode shape from sparse points to dense points or even distributed mode shapes, resulting in high-efficiency modal recognition quality. Attached Figure Description
[0039] Figure 1 This is a schematic diagram of the overall process of 3D reconstruction;
[0040] Figure 2 This is a schematic diagram of a visual displacement recognition device;
[0041] Figure 3 This is an external view of the cantilever beam model;
[0042] Figure 4 This is a 3D point cloud model of a cantilever beam.
[0043] Figure 5 The experimental setup includes a diagram showing the working conditions and area division.
[0044] Figure 6 This diagram illustrates the process of selecting ROI and target template.
[0045] Figure 7 This is a schematic diagram illustrating the ROI and target template selection results within a single region.
[0046] Figure 8 Output the time history curve of the target point displacement;
[0047] Figure 9 This is a schematic diagram of the overall process of modal recognition experiment. Detailed Implementation
[0048] The following describes the embodiments of the present invention in conjunction with the technical solution.
[0049] A cantilever beam model in a laboratory was used as the target structure for experimental verification of this invention. The cantilever beam model is a steel experimental model, with the fixed end connected integrally to the model's base. The base is bolted to a steel tabletop in the laboratory. The cantilever beam model is 60cm long, approximately 6mm thick, and approximately 3cm wide. Sixty target points were manually marked on the side of the cantilever beam model, with adjacent target points spaced 1cm apart. (See attached image for cantilever beam model appearance.) Figure 3 Six accelerometers were installed at six equal divisions along the beam length, and two laser displacement gauges were installed at the mid-span and free end for experimental comparison. The specific implementation method is as follows:
[0050] Step 1: Obtain a 3D point cloud model of the entire structure using multi-view vision-based 3D reconstruction technology.
[0051] (1) First, image data acquisition was carried out. The target structure was moved in a circular motion, and one image of the target was taken after moving a certain distance. A total of 38 target images were taken for subsequent 3D reconstruction.
[0052] (2) 3D point cloud reconstruction of the target structure was performed using the acquired image data. The 38 target images were processed using a 3D reconstruction algorithm, sequentially performing three calculations: sparse point cloud reconstruction, depth information acquisition, and dense point cloud reconstruction. Finally, a 3D dense point cloud model of the target structure was obtained. The reconstruction results are shown in [Figure number missing]. Figure 4 .
[0053] (3) Calibration of the actual size of the point cloud model based on the scaling factor method. By pre-marking target points at known distances on the side of the cantilever beam model, the scaling factor of the three-dimensional point cloud model was calculated to be 6.9636. The virtual space coordinates of the three-dimensional point cloud model were transformed to real space coordinates that can represent the actual size using the obtained scaling factor.
[0054] Step 2: Divide the structure into zones based on the measurement conditions.
[0055] (4) The structure was divided into regions based on the measurement conditions. Based on the experimental data of the cantilever beam model obtained from the preliminary experiment, the camera mounting distance for visual displacement recognition was set at 3.9m. The cantilever beam model was divided into 6 regions, each containing 13 target points. The overlapping parts of adjacent regions contained 3-4 shared target points for subsequent modal stitching. The test setup and region division are as follows: Figure 5 As shown.
[0056] Step 3: Obtain the displacement response of the target point region by region using monocular vision-based displacement recognition technology.
[0057] (5) Set up the instrument and record video of each area of the target under the specified excitation. As explained in step (4), based on the known frequency and displacement amplitude of the cantilever beam model, the appropriate position of the camera from the target structure is estimated to be 3.9m using the optical imaging principle. The instrument is then set up here to record video of each area. This experiment uses a hammer excitation and retains the hammer impact excitation data, which will be used for modal stitching.
[0058] (6) Use a target tracking algorithm based on monocular vision to process the video data recorded in each region and obtain the displacement time history of all target points in each region.
[0059] (6.1) Select the region of interest. Based on the previously recorded vibration video, obtain the approximate motion trajectory of the target vibration. When selecting the ROI, the selection area should be slightly larger than the vibration amplitude of the target.
[0060] (6.2) Select the tracking target template. Within each ROI region, select the target points that were manually marked on the side of the cantilever beam model before the experiment. An example of the ROI and target template selection process is shown below. Figure 6 As shown. The selection results within a single area are as follows. Figure 7 As shown in the figure, the 13 strip rectangles are ROIs, and the 13 rectangles inside them are the selected targets.
[0061] (6.3) Select known distance segments for size calibration. Since the target structure has been manually marked, the actual distance between each adjacent target point is known to be 1cm. To ensure the accuracy of size calibration, select 10 target points with an interval of 10cm as the calibration size and obtain the scale factor for testing each area.
[0062] (6.4) Target tracking based on template matching algorithm and sub-pixel registration based on image centroid method, outputting the displacement time history of the selected target points. By processing the video data frame by frame through the target tracking algorithm and sub-pixel registration algorithm in the computer program, the displacement time history curves of all 13 artificially marked target points in each region can be obtained efficiently. The output of the displacement time history curves of the target points is shown in [link to output]. Figure 8 As shown in the figure, the upper part represents the measured displacement time history curve in the horizontal direction, and the lower part represents the measured displacement time history curve in the vertical direction. By performing the aforementioned processing on each region sequentially, sub-pixel-level precision displacement data for all 60 target points of the cantilever beam model can be obtained.
[0063] Step 4: Obtain modal information of the entire structure using modal stitching technology.
[0064] (7) Modal information of the entire structure is obtained using modal stitching technology. Based on the previously measured spatial coordinates and time history responses of all 60 target points, modal stitching technology is used to stitch together the modal information of the entire structure into six regions, including modal frequencies and mode shapes. A schematic diagram of the overall modal identification test process is shown below. Figure 9 As shown, by combining 3D reconstruction technology with visual displacement recognition algorithms, structural modal information can be identified in a non-contact, low-cost, convenient and efficient manner.
Claims
1. A structural modal identification method that integrates three-dimensional point cloud reconstruction and vibration displacement identification, characterized in that, The specific steps are as follows: Step 1: Obtain a 3D point cloud model of the entire structure using multi-view vision-based 3D reconstruction technology. (1) Acquire image data of the target structure; take pictures of the target structure from different angles; (2) Use the acquired image data to reconstruct the three-dimensional point cloud of the target structure; use the three-dimensional reconstruction technology based on multi-view vision to process the acquired target image, and perform three calculations in sequence: sparse point cloud reconstruction, depth information acquisition and dense point cloud reconstruction to obtain the three-dimensional point cloud model of the target structure. (3) Calibration of the actual size of the 3D point cloud model based on the scaling factor method; Two calibration points are determined, and then the lengths of the calibration line segments formed by the two calibration points on the actual target structure and the point cloud model structure are measured respectively. The ratio of the two length measurement results is the scaling factor of the dense point cloud model of the target structure. Step 2: Divide the target structure into regions based on the measurement conditions. (4) Divide the target structure into regions based on the measurement conditions; the region division is essentially determined by the required measurement accuracy and hardware parameters: The theoretical accuracy of the displacement recognition algorithm can be expressed by the following formula: wherein s is the sub-pixel accuracy of the displacement recognition algorithm; p is the pixel size; f is the lens focal length; D is the measured distance; when both the displacement recognition algorithm and the hardware parameters are determined, the displacement measurement accuracy A will only depend on the measured distance D and is directly proportional to it; according to the imaging principle, the camera field of view range can be expressed by the following formula: wherein SS A is the camera sensor photosite area; When hardware parameters are fixed, the camera's field of view FOV It will depend only on the measurement distance. D Furthermore, the two are directly proportional; therefore, using the above formula, the area division should be determined based on the required accuracy to determine the measurement distance and thus the camera's field of view. Each divided area should ensure that it does not exceed the camera's field of view. When dividing the target structure into areas, each adjacent area should contain an overlapping part, and the overlapping part should contain at least one or more usable common target points. Step 3: Use monocular vision-based displacement recognition technology to obtain the displacement response of the target point in each region. (5) Set up the instrument and record video of each area of the target under the specified excitation; The camera's field of view is adjusted according to the previously set target area division to ensure that the entire measurement area is included in the camera frame. Under the specified excitation, the area is recorded for a preset duration, and the video data is saved for subsequent displacement recognition processing until the video data of the entire area is acquired in sequence. (6) Use a target tracking algorithm based on monocular vision to process the video data recorded in each region and obtain the displacement time history of all target points in each region; (6.1) Select the Region of Interest (ROI); Select a portion of the original image as the ROI by manually selecting a bounding box. This ROI should cover the entire area of the target's motion trajectory. (6.2) Select the target template for tracking; The selected target template should have the same motion as the target structure being measured. That is, the target motion selected based on the visual algorithm can represent the motion of the specified target point on the structure, and the two should not have relative displacement during the visual monitoring process. The selected target template should be clearly visible in the image, have features that can distinguish it from the content of other images, have no excessive deformation or external occlusion during the motion, and have no excessively similar regions within the ROI to which the target belongs. (6.3) Select a known distance segment for size calibration; by manually selecting a known distance segment in the image and inputting its actual size, the algorithm will automatically calculate the pixel size of the selected distance and then calculate the scaling factor of the image, realizing the conversion from pixel displacement to actual displacement; as for obtaining a known size segment in the image, it can be based on the target structure three-dimensional point cloud model obtained in step one; (6.4) Target tracking based on template matching algorithm and sub-pixel registration based on image centroid method, output the displacement time history of the selected target point; use template matching algorithm to search for the region with the highest similarity to the target template in ROI, and evaluate the similarity by calculating the specified objective function; use the zero-mean normalized cross-correlation function as the objective function for similarity search, the objective function is as follows: in x , y These represent the pixel coordinates of the original image. x’ , y’ These represent the pixel coordinates of the template image, R Indicates image correlation. T Represents the template image function. I Function representing the original image; After performing a similarity search on the ROI, the pixel coordinates corresponding to the location with the highest image relevance are the target integer pixel coordinates obtained by the template matching algorithm. Using the image centroid method, the centroid of the target image in each frame is calculated. The pixel neighborhood of the target's integer pixel-level coordinates is fitted by the template matching algorithm. The centroid mean of the target and its neighboring pixels is calculated based on the image grayscale. The centroid coordinates obtained are the sub-pixel-level coordinates of the target in each frame. The displacement time history of the target with sub-pixel precision is further obtained by combining each frame. Step 4: Obtain modal information of the entire structure using modal stitching technology. (7) Based on the spatial coordinates and displacement response of all target points previously measured, modal splicing technology is used to obtain the mode shape of each measurable mode frequency of each region of the target structure. The amplitude and direction of the displacement of the common target point between two adjacent regions are kept consistent, and each region is assigned its own consistency coefficient to ensure that the vibration ratio coefficient of each region of the whole structure remains consistent. Then, all regions of the whole structure are spliced modally.
2. A dynamic displacement monitoring device based on a monocular vision template matching algorithm, used to implement the structural modal identification method that integrates three-dimensional point cloud reconstruction and vibration displacement identification as described in claim 1, characterized in that, It includes: The video acquisition module is used to acquire raw video data for monitoring the dynamic displacement of the target structure, including a camera module, a fixed-focus lens, and a camera tripod for fixing the equipment. The processor uses a microcontroller as a microcomputer to execute computer programs. When the computer program is executed, the processor is used to: read the vibration video of the target structure acquired by the video acquisition module; and provide ROI and target selection pages to facilitate users in specifying the range and target for displacement recognition. It provides a calibration dimension selection page and dimension input line, automatically calculates the scaling factor and stores it for subsequent displacement output scaling conversion; Run the target tracking program based on template matching algorithm and the subpixel registration program based on image centroid method, output the displacement time history curve of the selected target point through the display screen of the output display module, and automatically store the displacement data; The output display module is used for human-computer interaction to select ROI, target and calibration dimensions, and to output visualized images of the monitoring site.
