Visual deformation monitoring method, device and system and storage medium

Abstract

The invention provides a visual deformation monitoring method, device and system and a storage medium, and belongs to the technical field of image monitoring, and the method comprises the steps: carrying out the image collection of a monitored target through a visual sensor, and obtaining an original target image; performing initialization processing on the original target image to obtain an initialized target image; and importing a temperature control on-off instruction, and performing temperature control analysis according to the temperature control on-off instruction to obtain a temperature control analysis result. According to the invention, high-precision displacement monitoring and deformation analysis can be realized, more comprehensive and visual monitoring data is provided for engineering structure safety assessment, and the intelligent level of structure health monitoring is obviously improved; the problems of insufficient point-like monitoring coverage, high installation and maintenance cost, poor real-time performance and lack of visual data in the existing monitoring technology are solved.

Description

Technical Field

[0001] The present invention mainly relates to the field of image monitoring technology, and in particular to a visual deformation monitoring method, device, system and storage medium. Background Art

[0002] In areas such as railways and bridges, real-time monitoring of structural deformation is crucial for ensuring facility safety. Traditional visual deformation monitoring methods primarily include contact sensors (such as strain gauges, displacement meters, inclinometers, etc.) and optical measurement equipment (such as total stations and laser rangefinders). For example, in railway track monitoring, track inspection vehicles or fixed displacement sensors are typically used to detect track geometric deformation. Bridge monitoring relies on distributed strain sensors or GNSS displacement monitoring systems. However, these traditional visual deformation monitoring methods have certain drawbacks, such as point-based monitoring, limited coverage, complex installation, high maintenance costs, insufficient real-time performance, reliance on manual intervention, and lack of visual data support. In recent years, computer vision-based monitoring technologies have gradually been applied to engineering monitoring, but they still face challenges such as ambient temperature interference, high computational complexity, and reliance on high-precision camera equipment. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to address the deficiencies of the existing technology and provide a visual deformation monitoring method, device, system and storage medium.

[0004] The present invention solves the above technical problems with the following technical solutions: A visual deformation monitoring method comprises the following steps: The image of the monitored target is collected by the visual sensor to obtain multiple original target images; Performing initialization processing on each of the original target images respectively to obtain an initialized target image corresponding to each of the original target images; Importing temperature control on / off instructions, performing temperature control analysis according to the temperature control on / off instructions, and obtaining temperature control analysis results; According to the temperature control analysis result, all the initialized target images are detected and analyzed to obtain a detection and analysis result; and deformation analysis is performed on the detection and analysis result to obtain a deformation monitoring result.

[0005] Another technical solution of the present invention to solve the above technical problem is as follows: a visual deformation monitoring device, comprising: an image acquisition module, configured to acquire images of a monitored target through a visual sensor to obtain a plurality of original target images; an initialization processing module, configured to perform initialization processing on each of the original target images to obtain an initialized target image corresponding to each of the original target images; A temperature control analysis module is used to import temperature control start and close instructions, perform temperature control analysis according to the temperature control start and close instructions, and obtain temperature control analysis results; A detection and analysis module, configured to perform detection and analysis on all the initialized target images according to the temperature control analysis result to obtain a detection and analysis result; The monitoring result obtaining module is used to perform deformation analysis on the detection and analysis results to obtain deformation monitoring results.

[0006] Based on the above-mentioned visual deformation monitoring method, the present invention also provides a visual deformation monitoring system.

[0007] Another technical solution of the present invention to solve the above technical problems is as follows: a visual deformation monitoring system includes a memory, a processor, and a computer program stored in the memory and runnable on the processor. When the processor executes the computer program, the visual deformation monitoring method described above is implemented.

[0008] Based on the above-mentioned visual deformation monitoring method, the present invention also provides a computer-readable storage medium.

[0009] Another technical solution of the present invention to solve the above technical problem is as follows: a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the visual deformation monitoring method as described above is implemented.

[0010] The beneficial effects of the present invention are as follows: an original target image is obtained by capturing an image of the monitored target through a visual sensor, an initialized target image is obtained by initializing the original target image, a temperature control analysis result is obtained according to a temperature control opening and closing instruction, a detection analysis result is obtained by detecting and analyzing the initialized target image according to the temperature control analysis result, and a deformation monitoring result is obtained by performing a deformation analysis on the detection and analysis result. This enables high-precision displacement monitoring and deformation analysis, provides more comprehensive and intuitive monitoring data for engineering structure safety assessment, significantly improves the intelligent level of structural health monitoring, and solves the problems of insufficient point monitoring coverage, high installation and maintenance costs, poor real-time performance, and lack of visual data in existing monitoring technologies. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 A schematic diagram of a flow chart of a visual deformation monitoring method provided by an embodiment of the present invention; Figure 2 This is a module block diagram of the visual deformation monitoring device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0012] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are only used to explain the present invention and are not used to limit the scope of the present invention.

[0013] Figure 1A flowchart of a visual deformation monitoring method provided by an embodiment of the present invention.

[0014] like Figure 1 As shown, a visual deformation monitoring method includes the following steps: The image of the monitored target is collected by the visual sensor to obtain multiple original target images; Performing initialization processing on each of the original target images respectively to obtain an initialized target image corresponding to each of the original target images; Importing temperature control on / off instructions, performing temperature control analysis according to the temperature control on / off instructions, and obtaining temperature control analysis results; According to the temperature control analysis result, all the initialized target images are detected and analyzed to obtain a detection and analysis result; and deformation analysis is performed on the detection and analysis result to obtain a deformation monitoring result.

[0015] It should be understood that the device is started to search for all targets (ie, monitored objects) within the field of view and perform target initialization deployment.

[0016] In the above embodiment, the image of the monitored target is captured by the visual sensor to obtain the original target image, the original target image is initialized to obtain the initialized target image, the temperature control analysis result is obtained according to the temperature control opening and closing instruction, the detection and analysis result is obtained by detecting and analyzing the initialized target image according to the temperature control analysis result, and the deformation analysis of the detection and analysis result is obtained to obtain the deformation monitoring result. This can achieve high-precision displacement monitoring and deformation analysis, provide more comprehensive and intuitive monitoring data for engineering structure safety assessment, significantly improve the intelligence level of structural health monitoring, and solve the problems of insufficient point monitoring coverage, high installation and maintenance costs, poor real-time performance, and lack of visual data in existing monitoring technologies.

[0017] Optionally, as an embodiment of the present invention, the process of performing temperature control analysis according to the temperature control on / off instruction to obtain the temperature control analysis result includes: Determine whether the temperature control opening and closing instruction is for temperature control processing. If so, perform temperature control processing through the temperature control chamber set on the visual sensor, and use the preset first temperature control result as the temperature control analysis result; if not, use the preset second temperature control result as the temperature control analysis result.

[0018] It should be understood that it is determined whether to perform temperature control. If temperature control is required, the temperature of the device (i.e., the temperature control chamber) is adjusted so that the device operates at a stable target temperature.

[0019] In the above embodiment, temperature control analysis is performed according to the temperature control opening and closing instructions to obtain temperature control analysis results, which can achieve high-precision displacement monitoring and deformation analysis, and provide more comprehensive and intuitive monitoring data for engineering structure safety assessment.

[0020] Optionally, as an embodiment of the present invention, the process of performing detection and analysis on all the initialized target images according to the temperature control analysis result to obtain the detection and analysis result includes: Performing a first detection on each of the initialized target images to obtain a first detection result corresponding to each of the original target images; Performing a second detection on each of the initialized target images to obtain a second detection result corresponding to each of the original target images; Determine whether both the first detection result and the second detection result are target loss. If so, generate a target loss signal and use the target loss signal as the detection analysis result; if not, use the initialized target image corresponding to the first detection result as the target image to be processed; Performing coordinate analysis on each of the target images to be processed to obtain target sub-pixel coordinates corresponding to each of the target images to be processed; According to the temperature control analysis results, displacement analysis is performed on each of the target images to be processed and the target sub-pixel coordinates corresponding to each of the target images to be processed to obtain target displacement data corresponding to each of the target images to be processed, and all of the target displacement data are used as detection and analysis results.

[0021] It should be understood that when both the first detection result and the second detection result are target loss, an alarm message needs to be reported.

[0022] In the above embodiment, the detection and analysis results are obtained by performing detection and analysis on the initialized target image based on the temperature control analysis results, thereby improving the reliability and accuracy of monitoring, while taking into account the monitoring efficiency and precision requirements, significantly improving the intelligent level of structural health monitoring, and solving the problems of insufficient point monitoring coverage, high installation and maintenance costs, poor real-time performance, and lack of visual data in existing monitoring technologies.

[0023] Optionally, as an embodiment of the present invention, the process of performing a first detection on each of the initialized target images to obtain a first detection result corresponding to each of the original target images includes: Importing a first original target training image corresponding to each of the original target images, and performing grayscale processing on each of the first original target training images to obtain a grayscaled target training image corresponding to each of the original target images; Performing denoising processing on each of the grayscaled target training images to obtain a denoised target training image corresponding to each of the original target images, wherein the denoised target training image includes a plurality of denoised target training image pixel values; Performing grayscale conversion processing on each of the initialized target images to obtain a converted target image corresponding to each of the original target images; performing histogram equalization processing on each of the converted target images to obtain an equalized target image corresponding to each of the original target images, wherein the equalized target image includes a plurality of equalized target image coordinates and equalized target image pixel values ​​corresponding to each of the equalized target image coordinates; Dividing each of the equalized target images using a pre-constructed sliding window to obtain a plurality of target regions corresponding to each of the original target images; Calculating the average value of the pixel values ​​of the multiple denoised target training images corresponding to each of the original target images respectively to obtain the pixel mean value of the denoised target training images corresponding to each of the original target images; The average value of all equalized target image pixel values ​​in each target area corresponding to each original target image is calculated respectively to obtain the average value of the equalized target image pixels corresponding to each target area in each original target image; the average value of the equalized target image pixels in each target area corresponding to each original target image, the average value of the denoised target training image pixels corresponding to each original target image, the average value of the equalized target image pixels corresponding to each target area in each original target image, and the average value of the denoised target training image pixels corresponding to each original target image are calculated respectively by the first formula to obtain multiple initial similarities corresponding to each original target image, and the first formula is: in, is the initial similarity at (x, y) in the bth target region corresponding to the ath original target image, T a (x′, y′) is the pixel value of the denoised target training image at (x′, y′) in the denoised target training image corresponding to the a-th original target image, is the pixel mean of the denoised target training image corresponding to the a-th original target image, is the equalized target image pixel value at (x+x′, y+y′) in the b-th target area corresponding to the a-th original target image, is the pixel mean of the equalized target image corresponding to the bth target region in the ath original target image; Filtering out the maximum value from all initial similarities corresponding to each of the original target images, and using the equalized target image coordinates corresponding to the maximum similarity as the filtered target image coordinates after filtering, thereby obtaining the filtered target image coordinates corresponding to each of the original target images; Optimizing the coordinates of each of the filtered target images using a sub-pixel interpolation algorithm to obtain optimized target image coordinates corresponding to each of the original target images; Detecting all the optimized target image coordinates using a non-maximum suppression algorithm to obtain a plurality of detected target image coordinates; Determine whether the initial similarity corresponding to the detected target image coordinates is less than a preset similarity threshold. If so, take the target loss as the first detection result; if not, take the detected target image coordinates as the first detection result.

[0024] It should be understood that the template preprocessing stage: collects a standard target image (i.e., the first original target training image) as a template library; performs grayscale and Gaussian filtering denoising on the template image (i.e., the first original target training image); calculates and stores the feature matrix and statistics of the template image.

[0025] Specifically, the real-time matching detection stage: obtain the real-time image frame to be detected (i.e., initialize the target image); preprocess the image (i.e., initialize the target image) (grayscale conversion, histogram equalization); use the sliding window mechanism to traverse the entire image (i.e., equalized target image); calculate the similarity index with the template at each window position (i.e., initial similarity).

[0026] Specifically, similarity calculation: similarity measurement is performed using the formula, and its mathematical expression is:

[0027] It should be understood that the target positioning stage: records the similarity scores of all window positions (i.e., initial similarity); finds the maximum similarity and its corresponding coordinates (i.e., the target image coordinates after screening); applies sub-pixel interpolation technology to improve positioning accuracy; and outputs the precise pixel coordinates of the target (i.e., the optimized target image coordinates).

[0028] Specifically, post-processing optimization: non-maximum suppression is used to eliminate duplicate detections; a similarity threshold (i.e., a preset similarity threshold) is set to filter out false matches; and target tracking is performed in combination with timing information.

[0029] In the above embodiment, each initialized target image is respectively detected for the first time to obtain a first detection result, thereby achieving millisecond-level rapid detection and ensuring continuous tracking of the target.

[0030] Optionally, as an embodiment of the present invention, the process of performing a second detection on each of the initialized target images to obtain a second detection result corresponding to each of the original target images includes: Importing second original target training images corresponding to each of the original target images, and training a pre-built YOLOv5 model using all of the second original target training images to obtain a trained YOLOv5 model; Update the parameters of the trained YOLOv5 model using the SGD optimizer to obtain an updated YOLOv5 model; evaluate the updated YOLOv5 model according to preset rules to obtain an evaluation result; Determine whether the evaluation result meets the evaluation criteria; if not, re-import the second original target training image corresponding to each of the original target images; if so, quantize the updated YOLOv5 model to obtain a quantized YOLOv5 model; Performing pruning optimization processing on the quantized YOLOv5 model to obtain a detection model; Each of the initialized target images is detected using the detection model to obtain a second detection result corresponding to each of the original target images.

[0031] Specifically, the processing flow for the second detection is as follows: (1) Target detection based on deep learning is implemented using the YOLOv5 model. First, a target image dataset containing different scenes is constructed, and after standardized annotation, it is divided into a training set, a validation set, and a test set.

[0032] (2) A lightweight YOLOv5 model was selected, using a RepVGG-style backbone network and an efficient decoupled detection head structure, optimized for a single-category detection task, and training parameters such as the initial learning rate 1e-3 and the SGD optimizer were set for model training.

[0033] (3) Evaluate the model performance by monitoring indicators such as GIoU loss and mAP, and verify the detection accuracy on the test set. If it does not meet the requirements, adjust the data enhancement strategy or hyperparameters and retrain until the accuracy requirements are met.

[0034] The trained model (i.e., the quantized YOLOv5 model) is quantized and pruned, and then deployed to the embedded main control board to implement a complete detection process that collects image input models in real time and outputs target pixel coordinates.

[0035] In the above embodiment, each initialized target image is subjected to a second detection to obtain a second detection result, thereby improving the reliability and accuracy of monitoring, while taking into account both monitoring efficiency and precision requirements, and significantly improving the intelligent level of structural health monitoring.

[0036] Optionally, as an embodiment of the present invention, the process of performing coordinate analysis on each of the target images to be processed to obtain target sub-pixel coordinates corresponding to each of the target images to be processed includes: Each of the target images to be processed is corrected using pre-calibrated camera parameters to obtain a corrected target image corresponding to each of the target images to be processed, wherein the corrected target image includes a plurality of corrected target image pixels; a Sobel operator algorithm is used to perform gradient calculation on each of the corrected target image pixels to obtain an X-axis gradient corresponding to each of the corrected target image pixels and a Y-axis gradient corresponding to each of the corrected target image pixels; and a second formula is used to calculate the X-axis gradient corresponding to each of the corrected target image pixels and the Y-axis gradient corresponding to each of the corrected target image pixels to obtain a gradient direction angle corresponding to each of the corrected target image pixels, wherein the second formula is: Among them, θ d is the gradient direction angle corresponding to the dth corrected target image pixel, is the Y-axis gradient corresponding to the d-th corrected target image pixel, is the X-axis gradient corresponding to the d-th corrected target image pixel; Performing a pixel search along the gradient direction angle direction with the corrected target image pixel as the center to obtain a plurality of searched target image pixels corresponding to each of the corrected target image pixels; Using a bilinear interpolation algorithm to perform gradient amplitude calculation on each of the searched target image pixels, to obtain a plurality of gradient amplitudes corresponding to each of the corrected target image pixels; Fitting the multiple gradient amplitudes corresponding to the pixels of the corrected target image respectively to obtain a gradient amplitude sequence polynomial corresponding to the pixels of the corrected target image; Solving each of the gradient amplitude sequence polynomials using a least squares algorithm to obtain the vertex abscissa corresponding to each pixel of the corrected target image; The third and fourth equations are used to calculate each pixel of the corrected target image, the gradient direction angle corresponding to each pixel of the corrected target image, and the vertex abscissa corresponding to each pixel of the corrected target image, respectively, to obtain the sub-pixel edge point coordinates corresponding to each pixel of the corrected target image. The sub-pixel edge point coordinates corresponding to each pixel of the corrected target image are then used to obtain multiple sub-pixel edge point coordinates corresponding to each target image to be processed. The third equation is: The fourth formula is: Among them, x d * is the X-axis coordinate of the sub-pixel edge point corresponding to the d-th corrected target image pixel, is the X-axis coordinate of the rectified target image pixel corresponding to the d-th rectified target image pixel, i d * is the horizontal coordinate of the vertex corresponding to the dth corrected target image pixel, θ d is the gradient direction angle corresponding to the dth corrected target image pixel, y d * is the Y-axis coordinate of the sub-pixel edge point corresponding to the d-th corrected target image pixel, is the Y-axis coordinate of the rectified target image pixel corresponding to the d-th rectified target image pixel; Using the least squares algorithm to fit the coordinates of multiple sub-pixel edge points corresponding to each of the target images to be processed, to obtain a fitting circle equation corresponding to each of the target images to be processed; Solving each of the fitted circle equations respectively to obtain a major axis ratio parameter corresponding to each of the target images to be processed, a minor axis ratio parameter corresponding to each of the target images to be processed, a rotation angle parameter corresponding to each of the target images to be processed, a first center coordinate parameter corresponding to each of the target images to be processed, and a second center coordinate parameter corresponding to each of the target images to be processed; The target sub-pixel coordinates corresponding to each target image to be processed are obtained by respectively calculating the major axis ratio control parameter, the minor axis ratio control parameter corresponding to each target image to be processed, the rotation angle control parameter corresponding to each target image to be processed, the first center coordinate control parameter corresponding to each target image to be processed, and the second center coordinate control parameter corresponding to each target image to be processed through the fifth and sixth equations. The fifth equation is: The sixth formula is: Among them, x k 0 is the target sub-pixel X coordinate corresponding to the kth target image to be processed, y k 0 is the target sub-pixel Y coordinate corresponding to the kth target image to be processed, A k is the long axis ratio control parameter corresponding to the kth target image to be processed, B k is the rotation angle control parameter corresponding to the kth target image to be processed, C k is the short axis ratio control parameter corresponding to the kth target image to be processed, Dk is the first center coordinate control parameter corresponding to the kth target image to be processed, E k is the second center coordinate control parameter corresponding to the kth target image to be processed.

[0037] It should be understood that target positioning uses a sub-pixel target positioning algorithm, which can break through the limitations of sensor pixel resolution and significantly improve the displacement measurement accuracy in the horizontal and vertical directions.

[0038] Specifically, image acquisition and preprocessing: Acquire real-time image data (i.e., the target image to be processed); then perform distortion correction on the image (i.e., the target image to be processed) based on pre-calibrated camera parameters (including intrinsic parameters and distortion coefficients). Note: Camera calibration is an offline processing step and must be completed before the system is used to obtain accurate distortion correction parameters.

[0039] It should be understood that when a target is detected, the Sobel operator is first used to perform pixel-level edge extraction; then, with the edge pixel (i.e., the pixel of the corrected target image) as the center, sub-pixel interpolation calculation is performed in the local neighborhood to accurately fit the edge position, thereby obtaining higher-precision sub-pixel coordinates.

[0040] Specifically, the specific method of sub-pixel edge extraction is as follows: Use the Sobel operator to calculate the gradient G of the image I(x, y) (i.e., the corrected target image pixel) in the x and y directions x (i.e., X-axis gradient) and G y (i.e., Y-axis gradient), the gradient direction θ (i.e., gradient direction angle) is calculated by the following formula: Take the current pixel (x c ,y c ) as the center, expand n points along the gradient direction θ (i.e., gradient direction angle), and use bilinear interpolation to calculate the gradient amplitude G of these points (i.e., the target image pixels after the search) i . For the gradient amplitude sequence {G i} Perform quadratic polynomial fitting as follows: G(i)=ai 2 +bi+c, Solve the parabola vertex abscissa (i.e. vertex abscissa) by least squares method The coordinates of the sub-pixel edge point (x * ,y * ) (i.e., sub-pixel edge point coordinates) are: x * =x c +i*cosθ,y * =yc +i*sinθ.

[0041] It should be understood that after extracting all sub-pixel edge coordinates of the target (ie, sub-pixel edge point coordinates), circle fitting is performed on the edges (ie, sub-pixel edge point coordinates) to further obtain the sub-pixel position of the target.

[0042] Specifically, all sub-pixel edge points (i.e., sub-pixel edge point coordinates) are extracted Then, the least square method is used to fit the circle equation, as follows: Ax 2 +Bxy+Cy 2 +Dx+Ey+F=0 The constraint condition is B 2 -4AC<0. A and C determine the ratio of the major and minor axes of the circle.

[0043] B determines the angle of rotation of the circle.

[0044] D, E: Center coordinates of the control circle.

[0045] F: A constant term related to the overall scaling and translation of the circle.

[0046] The parameters are obtained by solving the following linear system: Mv=0,v=[A,B,C,D,E,F] T Where M is a matrix consisting of edge points. The sub-pixel center of the circle (x0, y0) (i.e., the target sub-pixel coordinate) is calculated by the following formula:

[0047] In the above embodiment, coordinate analysis is performed on each target image to be processed to obtain the target sub-pixel coordinates, which can break through the limitation of the sensor pixel resolution and significantly improve the displacement measurement accuracy in the horizontal and vertical directions.

[0048] Optionally, as an embodiment of the present invention, the process of performing displacement analysis on each of the target images to be processed and the target sub-pixel coordinates corresponding to each of the target images to be processed according to the temperature control analysis result to obtain target displacement data corresponding to each of the target images to be processed includes: Determine whether the temperature control analysis result indicates that temperature control processing is to be performed. If so, perform a difference calculation between the target sub-pixel coordinates at the current moment and the target sub-pixel coordinates at the previous moment, and use the calculation result as the target displacement data; if not, import the temperature data corresponding to each of the target images to be processed; Predicting each of the target images to be processed and the temperature data corresponding to each of the target images to be processed by a pre-built regression model, thereby obtaining the target three-dimensional coordinates corresponding to each of the target images to be processed, the heading angle corresponding to each of the target images to be processed, the pitch angle corresponding to each of the target images to be processed, the roll angle corresponding to each of the target images to be processed, the X-axis initial focal length corresponding to each of the target images to be processed, the Y-axis initial focal length corresponding to each of the target images to be processed, the X-axis focal length change corresponding to each of the target images to be processed, the Y-axis focal length change corresponding to each of the target images to be processed, the first principal point offset parameter corresponding to each of the target images to be processed, and the second principal point offset parameter corresponding to each of the target images to be processed; The target three-dimensional coordinates corresponding to each target image to be processed, the heading angle corresponding to each target image to be processed, the pitch angle corresponding to each target image to be processed, the roll angle corresponding to each target image to be processed, the initial X-axis focal length corresponding to each target image to be processed, the initial Y-axis focal length corresponding to each target image to be processed, the X-axis focal length change corresponding to each target image to be processed, the Y-axis focal length change corresponding to each target image to be processed, the first principal point offset parameter corresponding to each target image to be processed, and the second principal point offset parameter corresponding to each target image to be processed are calculated respectively by the seventh formula to obtain the first image point drift corresponding to each target image to be processed and the second image point drift corresponding to each target image to be processed. The fixed drift corresponding to each target image to be processed is obtained by the first image point drift corresponding to each target image to be processed and the second image point drift corresponding to each target image to be processed. The seventh formula is: Among them, du k is the drift of the first image point corresponding to the kth target image to be processed, dv k is the drift of the second image point corresponding to the kth target image to be processed, is the first principal point offset parameter corresponding to the kth target image to be processed, is the second principal point offset parameter corresponding to the kth target image to be processed, is the initial focal length of the X axis corresponding to the kth target image to be processed, is the initial focal length of the Y axis corresponding to the kth target image to be processed, is the target X-axis coordinate corresponding to the kth target image to be processed, is the Y-axis coordinate of the target corresponding to the k-th target image to be processed, is the target Z-axis coordinate corresponding to the kth target image to be processed, is the X-axis focal length change corresponding to the k-th target image to be processed, is the Y-axis focal length change corresponding to the k-th target image to be processed, α k is the heading angle corresponding to the kth target image to be processed, β k is the pitch angle corresponding to the kth target image to be processed, γ k is the roll angle corresponding to the kth target image to be processed; The sub-pixel coordinates of each target are calculated separately from the fixed drift amount corresponding to each target image to be processed to obtain the compensated pixel coordinates corresponding to each target image to be processed, and the compensated pixel coordinates are used as the target displacement data.

[0049] It is understood that in the field of visual deformation monitoring, such as rail displacement detection and long-term bridge health monitoring, video measurement technology is widely used due to its advantages such as non-contact, high precision, and real-time performance. However, changes in ambient temperature can significantly affect measurement accuracy, leading to image point drift. When the temperature fluctuates, internal camera components such as the lens and sensor target plate will experience thermal deformation. These changes cause the image position to drift. In long-distance monitoring scenarios, this drift is amplified by the optical leverage effect, resulting in measurement errors on the millimeter or even centimeter scale. To model the temperature drift effect, multiple sets of indoor temperature control experiments were designed. Using a temperature-controlled chamber to simulate different temperature ranges (-15°C to 45°C) and temperature change rates (20°C / h to 60°C), the system collected data on the camera's internal temperature, ambient temperature, and image point drift. Data analysis revealed that the focal length change is linearly related to temperature, while the principal point shift exhibits nonlinear characteristics. Based on these findings, a variety of regression algorithms (including linear regression, neural networks, and Gaussian process regression) were used to establish a mapping relationship between temperature and camera parameter changes.

[0050] Specifically, in the actual compensation process, the real-time temperature (i.e., temperature data) is first input into the regression model (i.e., the pre-built regression model) to predict the camera parameter changes at the current temperature (i.e., the target three-dimensional coordinates, heading angle, pitch angle, roll angle, X-axis initial focal length, Y-axis initial focal length, X-axis focal length change, Y-axis focal length change, first principal point offset parameter, and second principal point offset parameter). These parameters are then substituted into the image point drift model to calculate the theoretical drift value (i.e., the fixed drift value). Finally, the predicted drift value (i.e., the fixed drift value) is subtracted from the measured image point coordinates (i.e., the target sub-pixel coordinates) to achieve real-time compensation. The image point drift model is as follows: The compensation implementation formula is as follows: p compensated =P measured -dp fixed .

[0051] In the above embodiment, displacement analysis is performed on the target image to be processed and the target sub-pixel coordinates based on the temperature control analysis results to obtain target displacement data, which can achieve high-precision displacement monitoring and deformation analysis, provide more comprehensive and intuitive monitoring data for engineering structure safety assessment, significantly improve the intelligence level of structural health monitoring, and solve the problems of insufficient point monitoring coverage, high installation and maintenance costs, poor real-time performance, and lack of visual data in existing monitoring technologies.

[0052] Optionally, as an embodiment of the present invention, the process of performing deformation analysis on the detection and analysis results to obtain deformation monitoring results includes: Deformation judgment is performed on all the target displacement data according to the preset deformation judgment rule, and the judgment result is used as the deformation monitoring result.

[0053] It should be understood that the structural deformation of the target is comprehensively judged based on the displacement conditions of all targets (ie, target displacement data), and then it is judged whether the target displacement will cause risks.

[0054] In the above embodiment, deformation analysis is performed on the detection and analysis results to obtain deformation monitoring results, which solves the problems of insufficient point monitoring coverage, high installation and maintenance costs, poor real-time performance and lack of visual data in existing monitoring technologies.

[0055] Optionally, as another embodiment of the present invention, the present invention can combine real-time image acquisition and AI deformation analysis technology to achieve all-weather automated monitoring and provide visual early warning to improve the safety management level of key infrastructure such as railways and bridges.

[0056] Optionally, as another embodiment of the present invention, in response to the problems existing in existing structural deformation monitoring technologies, such as insufficient point monitoring coverage, high installation and maintenance costs, poor real-time performance, and lack of visual data, the present invention uses visual measurement technology to achieve large-scale, high-density surface monitoring, which can be widely used in engineering monitoring scenarios such as bridge deformation and rail displacement. The use of non-contact measurement methods can effectively reduce installation and maintenance costs, while improving the reliability and service life of the monitoring system. Combined with real-time image processing technology and artificial intelligence algorithms, the present invention can achieve high-precision displacement monitoring and deformation analysis, and integrate visual data acquisition functions to provide more comprehensive and intuitive monitoring data for engineering structure safety assessments, significantly improving the intelligence level of structural health monitoring. It solves the problems of existing monitoring equipment being difficult to deploy, being greatly affected by ambient temperature, and having high subsequent maintenance costs, and solves the problem that existing equipment is unable to report images of on-site conditions after problems occur on-site.

[0057] Optionally, as another embodiment of the present invention, the present invention monitors the structural changes of the target by monitoring a target installed and deployed on the target to be monitored. The main process of monitoring the target is as follows: 1. Start the device, search for all targets in the field of view, and perform target initialization deployment.

[0058] 2. Determine whether to perform temperature control. If temperature control is required, adjust the device temperature so that the device operates at a stable target temperature.

[0059] 3. Perform target detection to determine whether the target exists. If not, an alarm message is reported; if the target exists, proceed to the subsequent positioning steps.

[0060] 4. Position the target at the pixel level, and then at the sub-pixel level. If temperature control is not performed, temperature drift compensation calculation is required, and then the displacement of the target is calculated based on the historical position of the target.

[0061] 5. Based on the displacement of all targets, comprehensively judge the structural deformation of the target and then determine whether the target displacement will cause risks.

[0062] 6. Repeat steps 3 to 6 above.

[0063] Optionally, as another embodiment of the present invention, several steps of the processing flow of the present invention mainly use the following key technologies: (1) Dual-target detection method; (2) High-precision target positioning and displacement detection algorithm; (3) Methods for suppressing equipment temperature control and temperature drift; (4) Temperature drift compensation algorithm.

[0064] Optionally, as another embodiment of the present invention, the present invention adopts a hybrid detection strategy that combines machine learning and deep learning to achieve real-time monitoring of targets and abnormal alarms. Under normal monitoring conditions, the system runs a lightweight machine learning algorithm based on template matching, and achieves millisecond-level rapid detection through a sliding window mechanism to ensure continuous tracking of the target. When the system detects that the target may be lost, it automatically triggers the deep learning enhanced detection mechanism and calls the pre-trained YOLOv5 model for high-precision secondary confirmation. Only when both detection algorithms confirm that the target is lost will the system trigger an alarm signal. This dual verification mechanism significantly improves the reliability and accuracy of the system, while taking into account the detection efficiency and accuracy requirements.

[0065] Optionally, as another embodiment of the present invention, the present invention further includes a method for suppressing device temperature control and temperature drift, which is specifically as follows: Self-heating or ambient temperature changes during camera operation can cause thermal deformation and refractive index drift in the optical system. This thermal effect can cause changes in lens focal length, sensor target surface displacement, and other parameters, ultimately causing image position shifts, seriously affecting measurement accuracy.

[0066] In order to suppress the impact of temperature fluctuations on the measurement system, the present invention adopts active temperature control technology to accurately manage the temperature of key optical components. The system designs independent temperature control chambers for image sensors and lenses, and adjusts the temperature in real time through PID control algorithms to stabilize it near the set value. The temperature control system uses PWM modulation technology to drive semiconductor refrigeration chips (TEC) and heating films to achieve rapid heating and precise cooling. In terms of control strategy, the system adopts an improved PID algorithm with temperature lag compensation. Its control quantity output comprehensively considers the current temperature deviation, historical cumulative error and temperature change trend to ensure that the temperature fluctuation of the temperature control chamber is controlled within ±0.5℃. The PID control differential equation is as follows: The calculated pidout result is converted into the PWM value of the thermostat to achieve temperature control of the device.

[0067] Alternatively, as another embodiment of the present invention, addressing the issues of existing structural deformation monitoring technologies, such as limited monitoring range, significant impact from ambient temperature, and high maintenance costs, the present invention utilizes innovative visual measurement technology to enable high-precision displacement monitoring of multiple points on structures such as bridges and railroad tracks with a single system, achieving large-scale, planar coverage monitoring. Temperature compensation for target displacement without equipment temperature control can significantly reduce equipment deployment and monitoring costs. Furthermore, when an anomaly is detected, the system automatically captures the scene and transmits it back in real time, supporting remote visual verification.

[0068] In addition, the device integrates edge computing capabilities and can be expanded to connect to various environmental sensors (such as temperature, humidity, vibration, etc.). Through multi-source data fusion and analysis, it can realize intelligent and continuous monitoring and early warning of structural health status such as bridge deformation and rail displacement, significantly improving the efficiency and reliability of engineering structure safety monitoring.

[0069] Figure 2 This is a module block diagram of a visual deformation monitoring device provided by an embodiment of the present invention.

[0070] Alternatively, as another embodiment of the present invention, Figure 2 As shown, a visual deformation monitoring device includes: an image acquisition module for acquiring images of a monitored target through a visual sensor to obtain a plurality of original target images; an initialization processing module for performing initialization processing on each of the original target images to obtain an initialized target image corresponding to each of the original target images; A temperature control analysis module is used to import temperature control start and close instructions, perform temperature control analysis according to the temperature control start and close instructions, and obtain temperature control analysis results; A detection and analysis module, configured to perform detection and analysis on all the initialized target images according to the temperature control analysis result to obtain a detection and analysis result; The monitoring result obtaining module is used to perform deformation analysis on the detection and analysis results to obtain deformation monitoring results.

[0071] Alternatively, another embodiment of the present invention provides a visual deformation monitoring system comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the visual deformation monitoring method described above is implemented. The system may be a computer or other system.

[0072] Optionally, another embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the visual deformation monitoring method as described above is implemented.

[0073] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0074] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0075] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of units is merely a logical functional division. In actual implementation, other division methods may be used, such as combining or integrating multiple units or components into another system, or ignoring or not implementing certain features.

[0076] Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected based on actual needs to achieve the objectives of the embodiments of the present invention.

[0077] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0078] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, and other media that can store program code.

[0079] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A visual deformation monitoring method, characterized in that: The steps include: The image of the monitored target is collected by the visual sensor to obtain multiple original target images; Performing initialization processing on each of the original target images respectively to obtain an initialized target image corresponding to each of the original target images; Importing temperature control on / off instructions, performing temperature control analysis according to the temperature control on / off instructions, and obtaining temperature control analysis results; Performing detection and analysis on all the initialized target images according to the temperature control analysis result to obtain a detection and analysis result; Perform deformation analysis on the detection and analysis results to obtain deformation monitoring results.

2. The visual deformation monitoring method according to claim 1, characterized in that: Determine whether the temperature control opening and closing instruction is for temperature control processing. If so, perform temperature control processing through the temperature control chamber set on the visual sensor, and use the preset first temperature control result as the temperature control analysis result; if not, use the preset second temperature control result as the temperature control analysis result.

3. The visual deformation monitoring method according to claim 2, characterized in that: The process of performing detection and analysis on all the initialized target images according to the temperature control analysis result to obtain the detection and analysis result includes: Performing a first detection on each of the initialized target images to obtain a first detection result corresponding to each of the original target images; Performing a second detection on each of the initialized target images to obtain a second detection result corresponding to each of the original target images; Determine whether both the first detection result and the second detection result are target loss. If so, generate a target loss signal and use the target loss signal as the detection analysis result; if not, use the initialized target image corresponding to the first detection result as the target image to be processed; Performing coordinate analysis on each of the target images to be processed to obtain target sub-pixel coordinates corresponding to each of the target images to be processed; According to the temperature control analysis results, displacement analysis is performed on each of the target images to be processed and the target sub-pixel coordinates corresponding to each of the target images to be processed to obtain target displacement data corresponding to each of the target images to be processed, and all of the target displacement data are used as detection and analysis results.

4. The visual deformation monitoring method according to claim 3, characterized in that: The process of performing a first detection on each of the initialized target images to obtain a first detection result corresponding to each of the original target images includes: importing a first original target training image corresponding to each of the original target images, performing grayscale processing on each of the first original target training images to obtain a grayscaled target training image corresponding to each of the original target images; performing denoising processing on each of the grayscaled target training images to obtain a denoised target training image corresponding to each of the original target images, wherein the denoised target training image includes a plurality of denoised target training image pixel values; Performing grayscale conversion processing on each of the initialized target images to obtain a converted target image corresponding to each of the original target images; performing histogram equalization processing on each of the converted target images to obtain an equalized target image corresponding to each of the original target images, wherein the equalized target image includes a plurality of equalized target image coordinates and equalized target image pixel values ​​corresponding to each of the equalized target image coordinates; Dividing each of the equalized target images using a pre-constructed sliding window to obtain a plurality of target regions corresponding to each of the original target images; Calculating the average value of the pixel values ​​of the multiple denoised target training images corresponding to each of the original target images respectively to obtain the pixel mean value of the denoised target training images corresponding to each of the original target images; Calculating the average value of all equalized target image pixel values ​​in each target area corresponding to each original target image respectively, to obtain the average value of the equalized target image pixel corresponding to each target area in each original target image; The first formula is used to calculate the pixel values ​​of each equalized target image in each target area corresponding to each original target image, the pixel mean of the denoised target training image corresponding to each original target image, the pixel mean of the equalized target image corresponding to each target area in each original target image, and the pixel value of each denoised target training image corresponding to each original target image, to obtain multiple initial similarities corresponding to each original target image. The first formula is: in, is the initial similarity at (x, y) in the bth target region corresponding to the ath original target image, T a (x′, y′) is the pixel value of the denoised target training image at (x′, y′) in the denoised target training image corresponding to the a-th original target image, is the pixel mean of the denoised target training image corresponding to the a-th original target image, is the equalized target image pixel value at (x+x′, y+y′) in the b-th target area corresponding to the a-th original target image, is the pixel mean of the equalized target image corresponding to the bth target region in the ath original target image; Filtering out the maximum value from all initial similarities corresponding to each of the original target images, and using the equalized target image coordinates corresponding to the maximum similarity as the filtered target image coordinates after filtering, thereby obtaining the filtered target image coordinates corresponding to each of the original target images; Optimizing the coordinates of each of the filtered target images using a sub-pixel interpolation algorithm to obtain optimized target image coordinates corresponding to each of the original target images; Detecting all the optimized target image coordinates using a non-maximum suppression algorithm to obtain a plurality of detected target image coordinates; Determine whether the initial similarity corresponding to the detected target image coordinates is less than a preset similarity threshold. If so, take the target loss as the first detection result; if not, take the detected target image coordinates as the first detection result.

5. The visual deformation monitoring method according to claim 3, characterized in that: The process of performing a second detection on each of the initialized target images to obtain a second detection result corresponding to each of the original target images includes: importing a second original target training image corresponding to each of the original target images, training a pre-built YOLOv5 model using all of the second original target training images, and obtaining a trained YOLOv5 model; Performing parameter updates on the trained YOLOv5 model using an SGD optimizer to obtain an updated YOLOv5 model; Evaluate the updated YOLOv5 model according to preset rules to obtain an evaluation result; Determine whether the evaluation result meets the evaluation criteria; if not, re-import the second original target training image corresponding to each of the original target images; if so, quantize the updated YOLOv5 model to obtain a quantized YOLOv5 model; Performing pruning optimization processing on the quantized YOLOv5 model to obtain a detection model; Each of the initialized target images is detected using the detection model to obtain a second detection result corresponding to each of the original target images.

6. The visual deformation monitoring method according to claim 3, characterized in that: The process of performing coordinate analysis on each of the target images to be processed to obtain target sub-pixel coordinates corresponding to each of the target images to be processed includes: performing correction processing on each of the target images to be processed using pre-calibrated camera parameters to obtain a corrected target image corresponding to each of the target images to be processed, wherein the corrected target image includes a plurality of corrected target image pixels; performing gradient calculation on each of the corrected target image pixels using a Sobel operator algorithm to obtain an X-axis gradient corresponding to each of the corrected target image pixels and a Y-axis gradient corresponding to each of the corrected target image pixels; and calculating the X-axis gradient corresponding to each of the corrected target image pixels and the Y-axis gradient corresponding to each of the corrected target image pixels using a second formula to obtain a gradient direction angle corresponding to each of the corrected target image pixels, wherein the second formula is: Among them, θ d is the gradient direction angle corresponding to the dth corrected target image pixel, is the Y-axis gradient corresponding to the d-th corrected target image pixel, is the X-axis gradient corresponding to the d-th corrected target image pixel; Performing a pixel search along the gradient direction angle direction with the corrected target image pixel as the center to obtain a plurality of searched target image pixels corresponding to each of the corrected target image pixels; Using a bilinear interpolation algorithm to perform gradient amplitude calculation on each of the searched target image pixels, to obtain a plurality of gradient amplitudes corresponding to each of the corrected target image pixels; Fitting the multiple gradient amplitudes corresponding to the pixels of the corrected target image respectively to obtain a gradient amplitude sequence polynomial corresponding to the pixels of the corrected target image; Solving each of the gradient amplitude sequence polynomials using a least squares algorithm to obtain the vertex abscissa corresponding to each pixel of the corrected target image; The third and fourth equations are used to calculate each pixel of the corrected target image, the gradient direction angle corresponding to each pixel of the corrected target image, and the vertex abscissa corresponding to each pixel of the corrected target image, respectively, to obtain the sub-pixel edge point coordinates corresponding to each pixel of the corrected target image. The sub-pixel edge point coordinates corresponding to each pixel of the corrected target image are then used to obtain multiple sub-pixel edge point coordinates corresponding to each target image to be processed. The third equation is: The fourth formula is: Among them, x d * is the X-axis coordinate of the sub-pixel edge point corresponding to the d-th corrected target image pixel, is the X-axis coordinate of the rectified target image pixel corresponding to the d-th rectified target image pixel, i d * is the horizontal coordinate of the vertex corresponding to the dth corrected target image pixel, θ d is the gradient direction angle corresponding to the dth corrected target image pixel, y d * is the Y-axis coordinate of the sub-pixel edge point corresponding to the d-th corrected target image pixel, is the Y-axis coordinate of the rectified target image pixel corresponding to the d-th rectified target image pixel; Using the least squares algorithm to fit the coordinates of multiple sub-pixel edge points corresponding to each of the target images to be processed, to obtain a fitting circle equation corresponding to each of the target images to be processed; Solving each of the fitted circle equations respectively to obtain a major axis ratio parameter corresponding to each of the target images to be processed, a minor axis ratio parameter corresponding to each of the target images to be processed, a rotation angle parameter corresponding to each of the target images to be processed, a first center coordinate parameter corresponding to each of the target images to be processed, and a second center coordinate parameter corresponding to each of the target images to be processed; The target sub-pixel coordinates corresponding to each target image to be processed are obtained by respectively calculating the major axis ratio control parameter, the minor axis ratio control parameter corresponding to each target image to be processed, the rotation angle control parameter corresponding to each target image to be processed, the first center coordinate control parameter corresponding to each target image to be processed, and the second center coordinate control parameter corresponding to each target image to be processed through the fifth and sixth equations. The fifth equation is: The sixth formula is: Among them, x k 0 is the target sub-pixel X coordinate corresponding to the kth target image to be processed, y k 0 is the target sub-pixel Y coordinate corresponding to the kth target image to be processed, A k is the long axis ratio control parameter corresponding to the kth target image to be processed, B k is the rotation angle control parameter corresponding to the kth target image to be processed, C k is the short axis ratio control parameter corresponding to the kth target image to be processed, D k is the first center coordinate control parameter corresponding to the kth target image to be processed, E k is the second center coordinate control parameter corresponding to the kth target image to be processed.

7. The visual deformation monitoring method according to claim 3, characterized in that: The process of performing displacement analysis on each of the target images to be processed and the target sub-pixel coordinates corresponding to each of the target images to be processed according to the temperature control analysis result to obtain target displacement data corresponding to each of the target images to be processed includes: Determine whether the temperature control analysis result indicates that temperature control processing is to be performed. If so, perform a difference calculation between the target sub-pixel coordinates at the current moment and the target sub-pixel coordinates at the previous moment, and use the calculation result as the target displacement data; if not, import the temperature data corresponding to each of the target images to be processed; Predicting each of the target images to be processed and the temperature data corresponding to each of the target images to be processed by a pre-built regression model, thereby obtaining the target three-dimensional coordinates corresponding to each of the target images to be processed, the heading angle corresponding to each of the target images to be processed, the pitch angle corresponding to each of the target images to be processed, the roll angle corresponding to each of the target images to be processed, the X-axis initial focal length corresponding to each of the target images to be processed, the Y-axis initial focal length corresponding to each of the target images to be processed, the X-axis focal length change corresponding to each of the target images to be processed, the Y-axis focal length change corresponding to each of the target images to be processed, the first principal point offset parameter corresponding to each of the target images to be processed, and the second principal point offset parameter corresponding to each of the target images to be processed; The target three-dimensional coordinates corresponding to each target image to be processed, the heading angle corresponding to each target image to be processed, the pitch angle corresponding to each target image to be processed, the roll angle corresponding to each target image to be processed, the initial X-axis focal length corresponding to each target image to be processed, the initial Y-axis focal length corresponding to each target image to be processed, the X-axis focal length change corresponding to each target image to be processed, the Y-axis focal length change corresponding to each target image to be processed, the first principal point offset parameter corresponding to each target image to be processed, and the second principal point offset parameter corresponding to each target image to be processed are calculated respectively by the seventh formula to obtain the first image point drift corresponding to each target image to be processed and the second image point drift corresponding to each target image to be processed. The fixed drift corresponding to each target image to be processed is obtained by the first image point drift corresponding to each target image to be processed and the second image point drift corresponding to each target image to be processed. The seventh formula is: Among them, du k is the drift of the first image point corresponding to the kth target image to be processed, dv k is the drift of the second image point corresponding to the kth target image to be processed, is the first principal point offset parameter corresponding to the kth target image to be processed, is the second principal point offset parameter corresponding to the kth target image to be processed, is the initial focal length of the X axis corresponding to the kth target image to be processed, is the initial focal length of the Y axis corresponding to the kth target image to be processed, is the target X-axis coordinate corresponding to the kth target image to be processed, is the Y-axis coordinate of the target corresponding to the k-th target image to be processed, is the target Z-axis coordinate corresponding to the kth target image to be processed, is the X-axis focal length change corresponding to the k-th target image to be processed, is the Y-axis focal length change corresponding to the k-th target image to be processed, α k is the heading angle corresponding to the kth target image to be processed, β k is the pitch angle corresponding to the kth target image to be processed, γ k is the roll angle corresponding to the kth target image to be processed; The sub-pixel coordinates of each target are calculated separately from the fixed drift amount corresponding to each target image to be processed to obtain the compensated pixel coordinates corresponding to each target image to be processed, and the compensated pixel coordinates are used as the target displacement data.

8. The visual deformation monitoring method according to claim 3, characterized in that: The process of performing deformation analysis on the detection and analysis results to obtain deformation monitoring results includes: Deformation judgment is performed on all the target displacement data according to the preset deformation judgment rule, and the judgment result is used as the deformation monitoring result.

9. A visual deformation monitoring device, characterized in that: include: An image acquisition module is used to acquire images of the monitored target through a visual sensor to obtain multiple original target images; an initialization processing module, configured to perform initialization processing on each of the original target images to obtain an initialized target image corresponding to each of the original target images; A temperature control analysis module is used to import temperature control start and close instructions, perform temperature control analysis according to the temperature control start and close instructions, and obtain temperature control analysis results; A detection and analysis module, configured to perform detection and analysis on all the initialized target images according to the temperature control analysis result to obtain a detection and analysis result; The monitoring result obtaining module is used to perform deformation analysis on the detection and analysis results to obtain deformation monitoring results.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the visual deformation monitoring method according to any one of claims 1 to 8 is implemented.

Images