Dic slope deformation measurement method and device

Abstract

The application discloses a DIC (Digital Image Correlation) slope deformation measurement method and device, and belongs to the technical field of slope measurement. The application aims to improve the precision of slope displacement measurement. The application comprises a camera array, a reference marker, a wind deflector and a field processing unit. The camera array is a multi-camera assembly composed of multiple high-resolution cameras fixed on a mounting bracket. A single high-resolution camera covers a specific sector, and the fields of view of all high-resolution cameras are combined to cover the entire road slope. The reference marker is a near-infrared spectrum emitting marker fixed on the monitoring point of the road slope, and the marker has periodic texture through a mask. The wind deflector is arranged beside the mounting bracket to damp the wind-induced oscillation of the camera array. The field processing unit is a control and data processing device deployed on the monitoring site. The cameras in the camera array are connected to the field processing unit, and the camera array collects the road slope image with the reference marker. The application improves the monitoring efficiency.

Description

Technical Field

[0001] This invention belongs to the field of slope measurement technology, specifically relating to a DIC slope deformation measurement method and device. Background Technology

[0002] Currently, slope surface displacement measurement mainly involves two methods: manual measurement, which involves using optical measuring instruments such as total stations and levels, but the monitoring frequency is too low; and automated measurement, which uses GNSS automated monitoring equipment to measure slope displacement. Compared with manual measurement, it has better all-weather adaptability and higher frequency, but its accuracy is slightly inferior to optical measurement, generally at the millimeter level or above. Summary of the Invention

[0003] The problem to be solved by this invention is to provide a wider field of view and higher resolution to improve the accuracy of slope displacement measurement, and to propose a DIC slope deformation measurement method and device.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] A DIC slope deformation measurement device includes a camera array, reference markers, a windbreak, and an on-site processing unit;

[0006] A camera array is a multi-camera assembly consisting of multiple high-resolution cameras fixed to a mounting bracket; a single high-resolution camera covers a specific sector, and the combined field of view of all the high-resolution cameras covers the entire roadside slope.

[0007] The reference marker is a marker that emits near-infrared spectrum and is fixed at the monitoring point of the road slope. The marker has a periodic texture through a masking method.

[0008] A windbreak is installed next to the mounting bracket to dampen wind-induced oscillations of the camera array; the field processing unit is a control and data processing device deployed at the monitoring site, and the cameras in the camera array are connected to the field processing unit, and the camera array acquires roadside slope images with reference markers;

[0009] Determine the basic parameters of a DIC slope deformation measurement device, including obtaining the size of the CMOS photosensitive element of a single camera in the array and the effective focal length of the matching lens; based on the optical projection relationship, determine the horizontal and vertical field-of-view coverage of a single camera on the target surface; according to the preset overlap rate parameter, calculate the horizontal and vertical center-to-center distance between adjacent cameras in the array, such that the distance and the coverage area satisfy the overlap coverage relationship; based on the distance and the target surface size, calculate the number of cameras required in the horizontal and vertical directions of the array, and generate array layout parameters.

[0010] Furthermore, the periodic texture of the reference marker includes one of a checkerboard, a cross grating, and a sine grating, and the periodic texture presents a spatial gradient in each direction, with the reference marker occupying 200 pixels on the camera.

[0011] A method for measuring DIC slope deformation includes the following steps:

[0012] S1. Measure the field of view coverage of the slope and set up the DIC slope deformation measuring device described above;

[0013] S2. Using cameras in a camera array, capture consecutive images of specific areas of the roadside slope, the images containing multiple installed reference markers;

[0014] S3. Perform masking on each frame of the continuous image obtained in step S2 to obtain multiple reference marker regions of interest;

[0015] S4. For the reference marker region of interest obtained in step S3, calculate the displacement of each reference marker between adjacent frames. First, perform displacement estimation based on frequency domain analysis to obtain an integer rough estimate of the reference marker displacement. Then, use the integer rough estimate of the reference marker displacement as the initial iteration value to construct the local region of interest. Feed it into the inverse combined Lucaskin algorithm for optimization to obtain the reference marker displacement.

[0016] S5. Using the displacement of the reference marker obtained in step S4, perform Kalman filter time series analysis based on adaptive displacement confidence to obtain the accurate displacement of the reference marker.

[0017] Furthermore, the specific implementation method of step S3 includes the following steps:

[0018] S3.1. Input each frame of the continuous image obtained in step S2 into the pre-trained YOLOv8 convolutional neural network, and output the initial position (x0, y0, w) of each reference marker, where x0 and y0 are the coordinates of the center point of the reference marker, and w is the size of the reference marker;

[0019] S3.2. Add a defined region a to w obtained in step S3.1. The center point coordinates of a are x0 and y0, and the size is w*(1+e), where e is the increase in size and e is less than 0.3.

[0020] S3.3. Use the a obtained in step S3.2 to create a mask M. The value of the central region of M is 1, and the values ​​of other regions are smoothly reduced until they become 0 near the edge. Then multiply the value of each pixel of the image by each value of M to obtain the reference marker region after masking.

[0021] S3.4. Perform a Fourier transform on the image of the reference marker region after masking to obtain the spectral distribution, then filter the spectrum, setting the bandwidth of each filter to [value missing]. ,in is a filter constant, which can be 2 or 3 to obtain multiple reference marker regions of interest.

[0022] Furthermore, the specific implementation method of step S4 includes the following steps:

[0023] S4.1. Define the image between two adjacent frames in multiple reference marker regions of interest as... and Then, the results obtained by performing a Fourier transform are as follows: , ;

[0024] S4.2. Calculate the cross spectrum , = ,in yes Conjugate;

[0025] S4.3. Calculate the phase correlation equation , ,in It is a very small constant, with a value less than ;

[0026] S4.4. Perform an inverse Fourier transform to obtain the phase correlation surface. , , Given the x-axis and y-axis coordinates of the phase correlation surface, a rough estimate of the x-axis and y-axis displacements of the reference marker is obtained as follows:

[0027]

[0028] in, It is a rough estimate of the x-axis displacement of the reference marker. It is a rough estimate of the y-axis displacement of the reference marker. It is the phase correlation surface, and argmax represents the coordinates for finding the maximum value of the phase correlation surface;

[0029] S4.5. Then, the integer rough estimate of the displacement of the benchmark is used as the initial iteration value to construct the local region of interest, which is then fed into the inverse combined Lucaskin algorithm for optimization to obtain the displacement of the benchmark.

[0030] Furthermore, the specific implementation method of step S5 includes the following steps:

[0031] S5.1. Establish a constant velocity Kalman filter to analyze the four elements of the state vector: cumulative total displacement in the x-direction, inter-frame displacement rate of change in the x-direction, cumulative total displacement in the y-direction, and inter-frame displacement rate of change in the y-direction; the observations are: inter-frame displacement rate of change in the x-direction and inter-frame displacement rate of change in the y-direction.

[0032] S5.2. Phase correlation surface obtained in step S4 The ratio of peak value to global mean in the calculation of the main energy peak. Main peak to secondary peak ratio and neighborhood mean ;

[0033] S5.3. Constructing ternary coupled confidence values ;

[0034] S5.4. Define the baseline confidence value for a normal, clear image as: Then the confidence bias is obtained. ;

[0035] S5.5. Observation noise covariance based on confidence bias-adjusted Kalman filter , L is a defined constant. The observation noise covariance serves as the baseline. The dynamic change of the observation noise covariance R will adaptively adjust the Kalman gain K. The Kalman gain K is related to the observation noise covariance. The relationship is represented as:

[0036]

[0037] Where P is the prediction error covariance, which is a stable value;

[0038] Then implement according to Adjustment ;

[0039] Then based on the Kalman gain The adjustment allows the Kalman filter to adaptively adjust to the image quality and complete the temporal smoothing and denoising of inter-frame displacement data and deformation trend fitting, outputting the accurate displacement of the reference marker.

[0040] The beneficial effects of this invention are:

[0041] The present invention discloses a DIC slope deformation measurement method, which uses an array-type DIC camera assembly to conduct real-time measurement of slope surface deformation, fully acquires relevant image sequences of the observed object, and preprocesses the images using various noise reduction and image enhancement techniques; then, it combines traditional and deep learning-based methods to perform multi-benchmark tracking, matching, and masking to provide conditions for time-domain analysis; finally, it optimizes the measured displacement through time-series analysis and provides backoff and protection mechanisms.

[0042] This invention discloses a DIC slope deformation measurement method. Compared to commonly used single-camera methods, multi-camera arrays can provide a wider field of view and higher resolution to improve the accuracy of displacement measurement. It directly, linearly, and symmetrically maps visual observation quality to Kalman observation noise R, achieving real-time adaptive linkage between image quality, sensor uncertainty, and filter gain. The invention matches the physical period of the reference marker with the frequency domain filter. The reference marker provides a high-quality signal, and the frequency domain filter purifies and efficiently utilizes the signal, achieving collaborative optimization between the reference marker and the software. This invention overcomes the problems of low frequency and insufficient accuracy in general optical flow monitoring methods, and the difficulty of reducing slope monitoring costs and improving monitoring efficiency through frequency domain analysis. Attached Figure Description

[0043] Figure 1 This is a flowchart of a DIC slope deformation measurement method according to the present invention. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only for explaining the invention and are not intended to limit the invention; that is, the described specific embodiments are merely a part of the embodiments of the invention, and not all of them. The components of the specific embodiments of the invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations, and the invention may also have other embodiments.

[0045] Therefore, the following detailed description of specific embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected specific embodiments of the invention. All other specific embodiments obtained by those skilled in the art based on these specific embodiments without inventive effort are within the scope of protection of this invention.

[0046] To further understand the invention's content, features, and effects, the following specific embodiments are provided, along with accompanying drawings. Figure 1 Detailed explanation is as follows:

[0047] Example 1:

[0048] A DIC slope deformation measurement device includes a camera array, reference markers, a windbreak, and an on-site processing unit;

[0049] A camera array is a multi-camera assembly consisting of multiple high-resolution cameras fixed to a mounting bracket; a single high-resolution camera covers a specific sector, and the combined field of view of all the high-resolution cameras covers the entire roadside slope.

[0050] The reference marker is a marker that emits near-infrared spectrum and is fixed at the monitoring point of the road slope. The marker has a periodic texture through a masking method.

[0051] A windbreak is installed next to the mounting bracket to dampen wind-induced oscillations of the camera array; the field processing unit is a control and data processing device deployed at the monitoring site, and the cameras in the camera array are connected to the field processing unit, and the camera array acquires roadside slope images with reference markers;

[0052] Determine the basic parameters of a DIC slope deformation measurement device, including obtaining the size of the CMOS photosensitive element of a single camera in the array and the effective focal length of the matching lens; based on the optical projection relationship, determine the horizontal and vertical field-of-view coverage of a single camera on the target surface; according to the preset overlap rate parameter, calculate the horizontal and vertical center-to-center distance between adjacent cameras in the array, such that the distance and the coverage area satisfy the overlap coverage relationship; based on the distance and the target surface size, calculate the number of cameras required in the horizontal and vertical directions of the array, and generate array layout parameters.

[0053] Furthermore, the periodic texture of the reference marker includes one of a checkerboard, a cross grating, and a sine grating, and the periodic texture presents a spatial gradient in each direction, with the reference marker occupying 200 pixels on the camera.

[0054] Furthermore, reference markers emitting near-infrared (NIR) spectra (center wavelength 850 nm) are installed at designated monitoring points on the slope. The monitoring points require that the reference markers and their textures be completely covered by the field of view of at least one camera. Slope displacement is inferred by tracking the movement of these markers. Near-infrared illumination is chosen because it offers better visibility and noise resistance compared to visible light under different lighting conditions. The markers have a masking mechanism that gives them periodic textures to facilitate frequency domain analysis; such textures include, but are not limited to, checkerboard patterns, crosshairs, and sinusoidal gratings. They also exhibit high spatial gradients in all directions to aid in accurate distance measurement. These characteristics enable the markers to be reliably detected by traditional computer vision algorithms and deep learning models, even under harsh environmental conditions. The reference markers occupy more than 200 pixels on the camera.

[0055] Furthermore, the cameras in the camera array use a global shutter, have a pixel count of over 20 million, and a lens focal length of over 135mm.

[0056] Furthermore, the horizontal field of view coverage C_w and the vertical field of view coverage C_h respectively satisfy: C_w=D×S_w / f, C_h=D×S_h / f, where D is the vertical installation distance, S_w and S_h are the effective photosensitive target width and height respectively, f is the effective focal length, and all physical quantities adopt a unified unit system.

[0057] The overlap rate parameter O ranges from 10% to 50%, and the horizontal center spacing d_x and the vertical center spacing d_y satisfy: d_x=C_w·(1-O), d_y=C_h·(1-O).

[0058] The calculation of the number of cameras includes: N_x = L / d_x, N_y = W / d_y, where L and W are the length and width of the target surface, N_x and N_y are positive integers after rounding up, and satisfy (N_x-1)·d_x + C_w ≥ L, (N_y-1)·d_y + C_h ≥ W.

[0059] Example 2:

[0060] A method for measuring DIC slope deformation includes the following steps:

[0061] S1. Measure the field of view coverage of the slope and set up a DIC slope deformation measuring device as described in Example 1;

[0062] S2. Using cameras in a camera array, capture consecutive images of specific areas of the roadside slope, the images containing multiple installed reference markers;

[0063] S3. Perform masking on each frame of the continuous image obtained in step S2 to obtain multiple reference marker regions of interest;

[0064] Furthermore, the specific implementation method of step S3 includes the following steps:

[0065] S3.1. Input each frame of the continuous image obtained in step S2 into the pre-trained YOLOv8 convolutional neural network, and output the initial position (x0, y0, w) of each reference marker, where x0 and y0 are the coordinates of the center point of the reference marker, and w is the size of the reference marker;

[0066] Furthermore, the convolutional neural network used was YOLO (YouOnlyLookOnce) v8 because of its strong object recognition capabilities and significantly faster computation speed compared to other convolutional neural networks. First, reference markers were photographed under different lighting and slope conditions, and then manually labeled. A YOLOv8 model was then pre-trained using these photos. During system installation, the model was retrained and fine-tuned specifically for the slopes to be detected. The fine-tuning process was the same as the initial training, but the size of the AncherBox in the YOLOv8 model was adjusted based on the size of the reference markers in the images to increase the accuracy of reference marker recognition.

[0067] S3.2. Add a defined region a to w obtained in step S3.1. The center point coordinates of a are x0 and y0, and the size is w*(1+e), where e is the increase in size and e is less than 0.3.

[0068] Furthermore, w corresponds to the width of a square and the diameter of a circle;

[0069] S3.3. Use the a obtained in step S3.2 to create a mask M. The value of the central region of M is 1, and the values ​​of other regions are smoothly reduced until they become 0 near the edge. Then multiply the value of each pixel of the image by each value of M to obtain the reference marker region after masking.

[0070] S3.4. Perform a Fourier transform on the image of the reference marker region after masking to obtain the spectral distribution, then filter the spectrum, setting the bandwidth of each filter to [value missing]. ,in is a filter constant, which can be 2 or 3 to obtain multiple reference marker regions of interest.

[0071] Furthermore, a Fourier transform is first performed on the masked reference marker region image to obtain its spectral distribution. Since the periodic texture exhibits several discrete and energy-concentrated characteristic peaks in the frequency domain, while noise and background are uniformly diffused in the spectrum, the spectrum is filtered. This eliminates high-frequency noise and static background interference, preserving the pure reference marker texture signal and providing high-quality input for subsequent phase-correlation displacement calculations. The filter center point of the frequency domain filter is determined by analyzing the reference marker texture. The reference marker texture has several sinusoidal frequency components in the frequency domain; each frequency component is extracted using a set of symmetrical narrow-bandpass filters.

[0072] S4. For the reference marker region of interest obtained in step S3, calculate the displacement of each reference marker between adjacent frames. First, perform displacement estimation based on frequency domain analysis to obtain an integer rough estimate of the reference marker displacement. Then, use the integer rough estimate of the reference marker displacement as the initial iteration value to construct the local region of interest. Feed it into the inverse combined Lucaskin algorithm for optimization to obtain the reference marker displacement.

[0073] Furthermore, the specific implementation method of step S4 includes the following steps:

[0074] S4.1. Define the image between two adjacent frames in multiple reference marker regions of interest as... and Then, the results obtained by performing a Fourier transform are as follows: , ;

[0075] S4.2. Calculate the cross spectrum , = ,in yes Conjugate;

[0076] S4.3. Calculate the phase correlation equation , ,in It is a very small constant, with a value less than ;

[0077] S4.4. Perform an inverse Fourier transform to obtain the phase correlation surface. , , Given the x-axis and y-axis coordinates of the phase correlation surface, a rough estimate of the x-axis and y-axis displacements of the reference marker is obtained as follows:

[0078]

[0079] in, It is a rough estimate of the x-axis displacement of the reference marker. It is a rough estimate of the y-axis displacement of the reference marker. It is the phase correlation surface, and argmax represents the coordinates for finding the maximum value of the phase correlation surface;

[0080] S4.5. Then, the integer rough estimate of the displacement of the benchmark is used as the initial iteration value to construct the local region of interest, which is then fed into the inverse combined Lucaskin algorithm for optimization to obtain the displacement of the benchmark.

[0081] Furthermore, the integer-level coarse displacement obtained by the phase correlation method is used as the initial value for the initial iteration, and the periodic texture region of the reference marker is used as the template to be matched to construct the gray-level conservation objective function of the inverse combination Lucaskin algorithm; by iteratively solving the deformation parameters, the sub-pixel level precise inter-frame displacement of the reference marker is obtained after convergence.

[0082] S5. Using the displacement of the reference marker obtained in step S4, perform Kalman filter time series analysis based on adaptive displacement confidence to obtain the accurate displacement of the reference marker.

[0083] Furthermore, the specific implementation method of step S5 includes the following steps:

[0084] S5.1. Establish a constant velocity Kalman filter to analyze the four elements of the state vector: cumulative total displacement in the x-direction, inter-frame displacement rate of change in the x-direction, cumulative total displacement in the y-direction, and inter-frame displacement rate of change in the y-direction; the observations are: inter-frame displacement rate of change in the x-direction and inter-frame displacement rate of change in the y-direction.

[0085] S5.2. Phase correlation surface obtained in step S4 The ratio of peak value to global mean in the calculation of the main energy peak. Main peak to secondary peak ratio and neighborhood mean ;

[0086] S5.3. Constructing ternary coupled confidence values ;

[0087] S5.4. Define the baseline confidence value for a normal, clear image as: Then the confidence bias is obtained. ;

[0088] S5.5. Observation noise covariance based on confidence bias-adjusted Kalman filter , L is a defined constant. The observation noise covariance serves as the baseline. The dynamic change of the observation noise covariance R will adaptively adjust the Kalman gain K. The Kalman gain K is related to the observation noise covariance. The relationship is represented as:

[0089]

[0090] Where P is the prediction error covariance, which is a stable value;

[0091] Then implement according to Adjustment ;

[0092] Then based on the Kalman gain The adjustment allows the Kalman filter to adaptively adjust to the image quality and complete the temporal smoothing and denoising of inter-frame displacement data and deformation trend fitting, outputting the accurate displacement of the reference marker.

[0093] Furthermore, in the Kalman filter, a larger filter gain K emphasizes the use of the measured displacement in the current frame; a smaller filter gain K emphasizes the prediction results based on historical deformation. Since P remains relatively stable, and slopes are subject to slow deformation, generally not sudden changes, the deformation rate and displacement changes of the slope in each frame are very gradual. Therefore, the model prediction error P should be stable. With P remaining stable, a larger R results in a smaller K, and a smaller R results in a larger K. According to the formulas in s5.3-5.4, a higher image confidence level (i.e., higher image quality) indicates higher image quality. The higher the value, The higher the confidence level, the smaller the R value, and the larger the K value, indicating a greater emphasis on the measured displacement of the current frame. Conversely, the lower the image confidence level (i.e., the lower the image quality). The lower the value, The higher the Kalman gain, the larger the K, and the smaller the K, thus relying more on historical deformation prediction results. Based on the adjustment of the Kalman gain K, the Kalman filter adaptively adjusts to the image quality and completes the temporal smoothing and denoising of inter-frame displacement data and deformation trend fitting, outputting stable and high-precision cumulative deformation displacement of the reference marker.

[0094] Furthermore, the lower the image confidence level, the greater the observation noise weight; the higher the confidence level, the less the observation noise weight.

[0095] Furthermore, based on the phase correlation domain IC-LK iterative calculation of inter-frame relative displacement, rather than capturing absolute spatial positions, the global cumulative deformation is obtained by sequentially accumulating the frame-by-frame displacement increments. Therefore, we replace the absolute spatial position variable commonly used in Kalman filters with displacement increments to adapt to our incremental observations and ensure the continuous reliability of long-term displacement monitoring data. This invention designs R to be linearly and symmetrically adaptively adjusted with the ternary confidence level, achieving real-time linkage between observation uncertainty and image quality while maintaining the baseline physical meaning of R0. Moreover, this linkage is linearly symmetrical, meaning that the impact on R is symmetrical when the degree of image quality deterioration and improvement is the same.

[0096] It should be noted that relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0097] Although this application has been described above with reference to specific embodiments, various modifications can be made and components can be replaced with equivalents without departing from the scope of this application. In particular, as long as there is no structural conflict, the features in the specific embodiments disclosed in this application can be combined with each other in any way. The lack of an exhaustive description of these combinations in this specification is merely for the sake of brevity and resource conservation. Therefore, this application is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A method for measuring DIC slope deformation, characterized in that, Includes the following steps: S1. Measure the field of view coverage of the slope and set up the DIC slope deformation measurement device, including camera array and reference markers; A camera array is a multi-camera assembly consisting of multiple high-resolution cameras fixed to a mounting bracket; a single high-resolution camera covers a specific sector, and the combined field of view of all the high-resolution cameras covers the entire roadside slope. The reference marker is a marker that emits near-infrared spectrum and is fixed at the monitoring point of the road slope. The marker has a periodic texture through a masking method. S2. Using cameras in a camera array, capture consecutive images of the roadside slope area, the images containing multiple installed reference markers; S3. Perform masking on each frame of the continuous image obtained in step S2 to obtain multiple reference marker regions of interest; S4. For the reference marker region of interest obtained in step S3, calculate the displacement of each reference marker between adjacent frames. First, perform displacement estimation based on frequency domain analysis to obtain an integer rough estimate of the reference marker displacement. Then, use the integer rough estimate of the reference marker displacement as the initial iteration value to construct the local region of interest. Feed it into the inverse combined Lucaskin algorithm for optimization to obtain the reference marker displacement. The specific implementation method of step S4 includes the following steps: S4.

1. Define the image between two adjacent frames in multiple reference marker regions of interest as... and Then, the results obtained by performing a Fourier transform are as follows: , ; S4.

2. Calculate the cross spectrum , = ,in yes The conjugate; S4.

3. Calculate the phase correlation equation , ,in It is a very small constant, with a value less than ; S4.

4. Perform an inverse Fourier transform to obtain the phase correlation surface. , , Given the x-axis and y-axis coordinates of the phase correlation surface, a rough estimate of the x-axis and y-axis displacements of the reference marker is obtained as follows: in, It is a rough estimate of the x-axis displacement of the reference marker. It is a rough estimate of the y-axis displacement of the reference marker. It is the phase correlation surface, and argmax represents the coordinates for finding the maximum value of the phase correlation surface; S4.

5. Then, the integer rough estimate of the displacement of the benchmark is used as the initial iteration value to construct the local region of interest, which is then fed into the inverse combined Lucaskin algorithm for optimization to obtain the displacement of the benchmark. S5. Using the displacement of the reference marker obtained in step S4, perform Kalman filter time series analysis based on adaptive displacement confidence to obtain the accurate displacement of the reference marker.

2. The method for measuring DIC slope deformation according to claim 1, characterized in that, The specific implementation method of step S3 includes the following steps: S3.

1. Input each frame of the continuous image obtained in step S2 into the pre-trained YOLOv8 convolutional neural network, and output the initial position (x0, y0, w) of each reference marker, where x0 and y0 are the coordinates of the center point of the reference marker, and w is the size of the reference marker; S3.

2. Add a defined region a to w obtained in step S3.

1. The center point coordinates of a are x0 and y0, and the size is w*(1+e), where e is the increase in size and e is less than 0.

3. S3.

3. Use the a obtained in step S3.2 to create a mask M. The value of the central region of M is 1, and the values ​​of other regions are smoothly reduced until they become 0 near the edge. Then multiply the value of each pixel of the image by each value of M to obtain the reference marker region after masking. S3.

4. Perform a Fourier transform on the image of the reference marker region after masking to obtain the spectral distribution, then filter the spectrum, setting the bandwidth of each filter to [value missing]. ,in is a filter constant, which can be 2 or 3 to obtain multiple reference marker regions of interest.

3. The method for measuring DIC slope deformation according to claim 2, characterized in that, The specific implementation method of step S5 includes the following steps: S5.

1. Establish a constant velocity Kalman filter to analyze the four elements of the state vector: cumulative total displacement in the x-direction, inter-frame displacement rate of change in the x-direction, cumulative total displacement in the y-direction, and inter-frame displacement rate of change in the y-direction; the observations are: inter-frame displacement rate of change in the x-direction and inter-frame displacement rate of change in the y-direction. S5.

2. Phase correlation surface obtained in step S4 The ratio of peak value to global mean in the calculation of the main energy peak. Main peak to secondary peak ratio and neighborhood mean ; S5.

3. Constructing ternary coupled confidence values ; S5.

4. Define the baseline confidence value for a normal, clear image as: Then the confidence bias is obtained. ; S5.

5. Observation noise covariance based on confidence bias-adjusted Kalman filter , L is a defined constant. The observation noise covariance serves as the baseline. The dynamic change of the observation noise covariance R will adaptively adjust the Kalman gain K. The Kalman gain K is related to the observation noise covariance. The relationship is represented as: Where P is the prediction error covariance, which is a stable value; Then implement according to Adjustment ; Then based on the Kalman gain The adjustment allows the Kalman filter to adaptively adjust to the image quality and complete the temporal smoothing and denoising of inter-frame displacement data and deformation trend fitting, outputting the accurate displacement of the reference marker.

4. The method for measuring DIC slope deformation according to claim 3, characterized in that, The DIC slope deformation measurement device also includes a windbreak and an on-site processing unit; A windbreak is installed next to the mounting bracket to dampen wind-induced oscillations of the camera array; the field processing unit is a control and data processing device deployed at the monitoring site, and the cameras in the camera array are connected to the field processing unit, and the camera array acquires roadside slope images with reference markers; Determine the basic parameters of a DIC slope deformation measurement device, including obtaining the size of the CMOS photosensitive element of a single camera in the array and the effective focal length of the matching lens; Based on the optical projection relationship, the horizontal and vertical field-of-view coverage of a single camera on the target surface are determined; according to the preset overlap rate parameter, the horizontal and vertical center-to-center distances of adjacent cameras in the array are calculated, such that the distances and the coverage ranges satisfy the overlapping coverage relationship. Based on the spacing and the target surface size, the number of cameras required for the array in the horizontal and vertical directions is calculated, and array layout parameters are generated.

5. The method for measuring DIC slope deformation according to claim 4, characterized in that, The periodic texture of the reference marker includes one of a checkerboard, a crosshair, and a sine grating, and the periodic texture presents a spatial gradient in all directions, with the reference marker occupying 200 pixels on the camera.

Images