Highly robust 3d digital image correlation method for large deformation measurement

Abstract

The application discloses a high-stable and robust three-dimensional digital image correlation method for large deformation measurement, which comprises the following steps: 1. calibrating a two-camera shooting system and shooting a deformation image of a tensile specimen by the two-camera shooting system; 2. determining grid points on a reference image and searching for corresponding matching points on a right image at the same time; 3. searching for corresponding matching points in time sequence; 4. respectively checking and correcting the matching points on the left image and the right image at the same time, correcting the error matching points on the right image by the left image, correcting the error matching points on the left image by the right image, and forming a closed checking and correcting; 5. correcting the error matching points by updating the reference image when the closed checking cannot be corrected, and then performing a closed checking again to ensure the robustness of the final matching points; and 6. using the checked and corrected matching points and camera calibration data to perform three-dimensional point reconstruction and strain calculation at different times. The application can realize high-precision and high-robust displacement and deformation measurement under large deformation.

Description

Technical Field

[0001] This invention relates to the fields of experimental mechanics and non-contact full-field displacement and strain measurement, and specifically to a high-precision matching method. Background Technology

[0002] Digital image correlation (DIR) is a widely used non-contact 3D measurement technique. Initially applied to materials mechanical property testing, it has now been applied in various fields such as industrial measurement, civil engineering, bridge structures, aerospace, and biomedicine. This is mainly due to its unique advantages, such as simple operation, immunity to ambient light, and the ability to perform full-field measurements. Furthermore, it can be used in conjunction with microscopes for microscopic measurements as well as for large-scale outdoor macroscopic measurements. Since its introduction in the early 1980s, this method has been extensively studied by experts and scholars both domestically and internationally. From the initial two-dimensional DIR method, it has evolved into a three-dimensional DIR method. Compared to two-dimensional DIR, the three-dimensional DIR method can measure out-of-plane displacement, three-dimensional topography, and three-dimensional deformation.

[0003] The main steps of 3D digital image correlation methods are: camera calibration, image acquisition, stereo image pair matching, temporal image pair matching, and 3D coordinate calculation. Currently, camera calibration and image matching under large deformation conditions are the focus of research. Image matching generally includes two steps: coarse search and fine search, which uses the coarse search results as the initial points for the inverse combination Gaussian-Newton algorithm. The accuracy and speed of matching algorithms have always been of great concern. Currently, coarse search has commonly used algorithms such as path propagation methods, intelligent algorithms, point-by-point methods, and feature point methods, with the point-by-point method having the highest accuracy. Since the introduction of the inverse combination algorithm, fine search has become a de facto standard optimization algorithm due to its advantages of high accuracy and high speed. There are three implementation methods for 3D digital image correlation algorithms. Usually, people directly use one of the calculations to obtain the final result, but in 3D large deformation conditions, incorrect matching points can easily occur, affecting the final calculation result. Summary of the Invention

[0004] To overcome the shortcomings of existing technologies, this invention provides a highly robust three-dimensional digital image correlation method for large deformation measurement. This method aims to solve the calculation accuracy problem of existing algorithms when applied under large deformation and large curvature conditions, thereby achieving high-precision and robust displacement and deformation measurement under large deformation conditions and improving the overall three-dimensional measurement accuracy.

[0005] To achieve the above-mentioned objectives, the present invention adopts the following technical solution:

[0006] The present invention provides a highly robust three-dimensional digital image correlation method for large deformation measurement, characterized by comprising the following steps:

[0007] Step 1: Obtain the internal and external parameters of the left and right cameras using the Zhang Zhengyou calibration method. The internal parameters include: focal length, principal point coordinates, and lens distortion coefficient; the external parameters include: rotation matrix R and translation vector T between the left and right cameras.

[0008] Step 2: The left and right cameras synchronously acquire left and right stretched and deformed images at N time points at fixed time intervals, and use the lens distortion coefficient to correct the left and right stretched and deformed images at each time point to obtain the corrected left and right images at each time point.

[0009] Step 3: Using the corrected left image at time 0 as the reference image for stereo matching at time 0, and the corrected right image at time 0 as the target image for stereo matching at time 0, perform stereo matching, verification, and reconstruction of the three-dimensional coordinate points at time 0 on the left and right images at time 0.

[0010] Step 3.1: Select the region of interest on the stereo matching reference image at time 0, and divide the region of interest into a grid. Each intersection point on the grid is used as a point to be calculated.

[0011] Step 3.2: Use a point P to be calculated on the stereo matching reference image at time 0. l0 Using the center point as the reference sub-region at time 0, and selecting an M×M region around the center point as the reference sub-region, the point P to be calculated is obtained using equation (1-1). l0 The epipolar line l on the stereo-matched target image at time 0 pr :

[0012] l pr =Fp l (1-1)

[0013] In equation (1-1), p l For the point P to be calculated on the stereo matching reference image at time 0. l0 The homogeneous coordinates of F are given, where F is the fundamental matrix, and A l and A r Let R be the internal parameters of the left and right cameras, respectively; R is the rotation matrix; and S is the antisymmetric matrix related to the translation vector T of the left and right cameras.

[0014] Select polar lines l in sequence pr Using each coordinate point on the image as the center, an area of ​​size M×M is selected around the center as the target sub-region for stereo matching at time 0. The zero-mean normalized cross-correlation function ZNCC is used to calculate the correlation between the reference sub-region at time 0 and the target sub-region for stereo matching at time 0. The center of the target sub-region with the highest correlation is taken as the point P to be calculated on the stereo matching reference image at time 0. l0 matching points;

[0015] Step 3.3: Use the second-order inverse combination Gauss-Newton algorithm to calculate point P. l0 The matching points are optimized for accuracy to obtain the point P to be calculated. l0 Subpixel matching point P on the stereo matching target image at time 0 r0 and its surrounding sub-pixel grayscale matrix G of size M×M r0 ;

[0016] Step 3.4: Match the sub-pixel matching points P on the stereo matching target image at time 0. r0 As a new point to be calculated, the accuracy is optimized using epipolar constraint coarse search and second-order inverse combination Gauss-Newton algorithm to obtain the new point P. r0 Matching point P′ on the stereo matching reference image at time 0 l0 ;

[0017] Verify P l0 and P′ l0 The distance d between l0 If the distance d l0 If the value is less than the threshold, it indicates a correct match, and step 3.5 is executed; otherwise, the matching point P is discarded. l0 Then return to step 3.2 to continue processing the next point to be calculated;

[0018] Step 3.5: Based on the principle of stereoscopic vision, use the point P to be calculated. l0 Sub-pixel matching point P r0 The coordinates of the points and the camera's intrinsic and extrinsic parameters are used to reconstruct a 3D coordinate point W0 at time 0; thus, the 3D coordinates of all correctly matched points at time 0 are obtained.

[0019] Step 4: Calculate the three-dimensional coordinates of the points at each deformation moment;

[0020] Step 4.0: Define the current time as n and initialize n = 1; use the corrected left image at time 0 as the left reference image;

[0021] Step 4.1: Using the corrected left image at time n as the target image for time-series matching of the left sequence at time n, and performing time-series matching on the left reference image:

[0022] Select the point P to be calculated on the left reference image. l0 Using a reference subregion of size M×M centered at the center, the point P to be calculated is obtained by a limited-range, interval-based search and a first-order inverse combination Gauss-Newton algorithm. l0 At time n, the matching point P on the target image of the left sequence temporal matching. ln , and P l0 and P ln The correlation between them corrl0_ln and P ln A sub-pixel grayscale matrix G with a surrounding size of M×M ln ;

[0023] Step 4.2: Using the right image corrected at time 0 as the right reference image, and the right image corrected at time n as the target image for time-series temporal matching of the right sequence at time n, perform temporal matching:

[0024] The sub-pixel grayscale matrix G on the right reference image r0 As a reference subregion, P is calculated using a limited-range, interval-based search and a first-order inverse combination Gauss-Newton algorithm. r0 At time n, the matching point P on the target image of the right sequence temporal matching. rn , and P r0 and P rn The correlation between them corr r0_rn and P rn A subpixel grayscale matrix G with a surrounding size of M×M rn ;

[0025] Step 4.3: Closure check and correction between the left image after correction at time n and the right image after correction at time n;

[0026] Step 4.3.1: Stereo matching between the right image corrected at time n and the left image corrected at time n:

[0027] Using the right image corrected at time n as the right reference image for stereo matching at time n, and the left image corrected at time n as the left target image for stereo matching at time n, the sub-pixel grayscale matrix G is used. rn Using the epipolar constraint and a second-order inverse combination Gauss-Newton algorithm as the reference region, P is calculated. rn Matching point P′ on the left target image of stereo matching at time n. ln , and P rn and P′ ln The correlation between them corr rn_ln And calculate P′ ln With P ln The distance d between ln ;

[0028] Step 4.3.2: Stereo matching between the left image after correction at time n and the right image after correction at time n:

[0029] The left image after correction at time n is used as the left reference image for stereo matching at time n, and the right image after correction at time n is used as the right target image for stereo matching at time n. The sub-pixel grayscale matrix G ln Using the epipolar constraint and a second-order inverse combination Gauss-Newton algorithm as the reference region, P is obtained. lnThe matching point P′ on the right target image of the stereo matching at time n. rn , and P ln and P′ rn The correlation between them is corr ln_rn And calculate P′ rn With P rn The distance d between rn ;

[0030] Step 4.3.3: Define two all-zero flag matrices signL and signR, and the dimensions of the flag matrices signL and signR are the same as the dimensions of the grid.

[0031] Step 4.3.4: Determine corr l0_ln and corr r0_rn All are greater than the correlation threshold, and d ln and d rn If the distance threshold is less than 1, it means that the corresponding two-dimensional coordinate points on the left and right images are correctly matched, and the element values ​​at the corresponding positions in the flag matrices signL and signR are updated to 1; otherwise, proceed to step 4.3.5.

[0032] Step 4.3.5: Determine corr l0_ln and corr ln_rn If all values ​​are greater than the correlation threshold, then P in the right reference image is considered valid. r0 There is a timing mismatch in the right-hand graph at time n, therefore P is... rn The coordinate values ​​are corrected to P′ rn If the coordinates are corrected, update the corresponding elements of the flag matrices signL and signR to 1, indicating that the correction has been made; otherwise, proceed to step 4.3.6.

[0033] Step 4.3.6: Determine corr r0_rn and corr rn_ln If all values ​​are greater than the correlation threshold, then P in the left reference image is considered valid. l0 At time n, the timing match of the left graph is incorrect, therefore P is... ln The coordinate values ​​are corrected to P′ ln If the coordinates are found, update the corresponding elements of the flag matrices signL and signR to 1, indicating that the correction has been made; otherwise, proceed to step 4.3.7.

[0034] Step 4.3.7: Determine whether the flag matrices signL and signR are all-one matrices. If they are, it means that all matching points obtained at time n have passed the verification and been successfully corrected, and proceed to step 4.3.8; otherwise, it means that the matching points on the corrected left and right images obtained in steps 4.1 and 4.2 at time n have not been successfully verified and corrected. In this case, the corrected left and right images at time n / 2 are used as the reference images for the left temporal matching and the right temporal matching, respectively, and the process returns to step 4.1 to continue calculating the coordinates of the coordinate points where the temporal matching failed.

[0035] Step 4.3.8: Reconstruct all three-dimensional coordinate points at time n based on the matching points in the corrected left and right images and the camera parameters;

[0036] Step 5: After assigning n+1 to n, return to step 4.1 and execute sequentially until n>N, thereby obtaining the three-dimensional coordinate points at different times, which are used to calculate the displacement and strain when the three-dimensional coordinates change. N represents the total number of times.

[0037] The present invention provides an electronic device, including a memory and a processor, wherein the memory is used to store a program that supports the processor in executing the highly robust three-dimensional digital image correlation method, and the processor is configured to execute the program stored in the memory.

[0038] The present invention discloses a computer-readable storage medium on which a computer program is stored, wherein the computer program, when executed by a processor, performs the steps of the highly robust three-dimensional digital image correlation method.

[0039] Compared with existing technologies, the beneficial effects of this invention are reflected in:

[0040] 1. This invention proposes to use the left and right images at the same time for verification and correction, forming a closed system that can accurately detect and correct errors, ensuring the reliability and robustness of the final calculation results;

[0041] 2. This invention proposes that during time-series search, matching points on the reference image can be used to narrow the search range on the target image, avoiding the drawback of traditional full-field search being too time-consuming.

[0042] 3. This invention proposes two marker matrices, which are used to identify points that are definitely matched correctly and points that are not matched correctly in the left and right images, respectively. According to the records of the marker matrices, only points that are mismatched are corrected by replacing the reference image, which reduces the amount of computation and some error accumulation.

[0043] 4. The three-dimensional digital image correlation algorithm proposed in this invention can be used for measurement under conditions of large deformation and large curvature, meeting the needs of most application scenarios. Attached Figure Description

[0044] Figure 1 This is a flowchart illustrating the three-dimensional digital image correlation algorithm proposed in this invention;

[0045] Figure 2 This is a grid division diagram of the region of interest on the reference image at time 0 of this invention. Detailed Implementation

[0046] In this embodiment, a highly robust three-dimensional digital image correlation method for large deformation measurement includes the following steps:

[0047] Step 1: Obtain the internal and external parameters of the left and right cameras using the Zhang Zhengyou calibration method. The internal parameters include: focal length, principal point coordinates, and lens distortion coefficients; the external parameters include: the rotation matrix R and translation vector T between the left and right cameras.

[0048] Step 2: The left and right cameras synchronously acquire left and right stretched and deformed images at N time points at fixed time intervals, and use the lens distortion coefficient to correct the left and right stretched and deformed images at each time point to obtain the corrected left and right images at each time point.

[0049] Step 3: Using the corrected left image at time 0 as the reference image for stereo matching at time 0, and the corrected right image at time 0 as the target image for stereo matching at time 0, perform stereo matching, verification, and reconstruction of the three-dimensional coordinate points at time 0 on the left and right images. Figure 1 A diagram illustrating the matching of ① and ②;

[0050] Step 3.1: Select the region of interest on the stereo matching reference image at time 0, and divide the region of interest into a grid. Each intersection point on the grid is used as a point to be calculated, such as... Figure 2 As shown, a rectangular region of interest is selected and divided into a grid, resulting in 17×9 points to be calculated.

[0051] Step 3.2: Use a point P to be calculated on the stereo matching reference image at time 0. l0 Using the center point as the reference point, and selecting an M×M region around the center point as the reference sub-region at time 0, the reference sub-region is as follows: Figure 1 As shown, the point P to be calculated is obtained using equation (1-1). l0 The epipolar line l on the stereo-matched target image at time 0 pr :

[0052] l pr =Fp l (1-1)

[0053] In equation (1-1), p lFor the point P to be calculated on the stereo matching reference image at time 0. l0 The homogeneous coordinates of F are given, where F is the fundamental matrix, and A l and A r Let R be the internal parameters of the left and right cameras, respectively; S is the rotation matrix; S is the antisymmetric matrix related to the translation vector T of the left and right cameras, and S has the specific form shown in equation (1-2).

[0054]

[0055] Select polar lines l in sequence pr Using each coordinate point on the image as the center, and selecting an area of ​​size M×M around the center as the target sub-region for stereo matching at time 0, the correlation between the reference sub-region at time 0 and the target sub-region for stereo matching at time 0 is calculated using Equation (1-3), i.e., the zero-mean normalized cross-correlation function ZNCC. The center of the target sub-region with the highest correlation is then taken as the point P to be calculated on the stereo matching reference image at time 0. l0 matching points;

[0056]

[0057] Step 3.3: Use the second-order inverse combination Gauss-Newton algorithm to calculate point P. l0 The matching points are optimized for accuracy to obtain the point P to be calculated. l0 Subpixel matching point P on the stereo matching target image at time 0 r0 and its surrounding sub-pixel grayscale matrix G of size M×M r0 At this time, the grayscale matrix G r0 Since it is optimized by the second-order inverse combination Gauss-Newton algorithm, it is actually an irregular sub-pixel position interpolation matrix, but its size is still M×M.

[0058] Step 3.4: Match the sub-pixel matching point P on the stereo matching target image at time 0. r0 As a new point to be calculated, the accuracy is optimized using epipolar constraint coarse search and second-order inverse combination Gauss-Newton algorithm to obtain the new point P. r0 Matching point P′ on the stereo matching reference image at time 0 l0 ;

[0059] Verify P l0 and P′ l0 The distance d between l0 If the distance d l0 If the value is less than the threshold, it indicates a correct match, and step 3.5 is executed; otherwise, the matching point P is discarded. l0 Then return to step 3.2 to continue processing the next point to be calculated;

[0060] Step 3.5: Based on the principle of stereoscopic vision, use the point P to be calculated. l0 Sub-pixel matching point P r0 The coordinates of the points and the camera's internal and external parameters are used to reconstruct a three-dimensional coordinate point W0 at time 0 according to equation (1-4); thus obtaining the three-dimensional coordinate points corresponding to all correctly matched points at time 0.

[0061]

[0062] Step 4: Calculate the three-dimensional coordinates of the points at each deformation moment;

[0063] Step 4.0: Define the current time as n and initialize n = 1; use the corrected left image at time 0 as the left reference image;

[0064] Step 4.1: Using the corrected left image at time n as the target image for temporal matching of the left sequence at time n, and performing temporal matching on the left reference image, corresponding to... Figure 1 The matching process on the ③ side:

[0065] Select the point P to be calculated on the left reference image. l0 Using a reference subregion of size M×M centered at the center, the point P to be calculated is obtained by a limited-range, interval-based search and a first-order inverse combination Gauss-Newton algorithm. l0 At time n, the matching point P on the target image of the left sequence temporal matching. ln , and P l0 and P ln The correlation between them corr l0_ln and P ln A sub-pixel grayscale matrix G with a surrounding size of M×M ln ;

[0066] Step 4.2: Using the right image corrected at time 0 as the right reference image, and the right image corrected at time n as the target image for time-series temporal matching at time n, temporal matching is performed, corresponding to... Figure 1 The matching process above (④):

[0067] The sub-pixel grayscale matrix G on the right reference image r0 As a reference subregion, P is calculated using a limited-range, interval-based search and a first-order inverse combination Gauss-Newton algorithm. r0 At time n, the matching point P on the target image of the right sequence temporal matching. rn , and P r0 and P rn The correlation between them corr r0_rn and P rn A subpixel grayscale matrix G with a surrounding size of M×M rn ;

[0068] Step 4.3: Closure check and correction between the left image after correction at time n and the right image after correction at time n;

[0069] Step 4.3.1: Stereo matching between the right image corrected at time n and the left image corrected at time n:

[0070] The right image after correction at time n is taken as the right reference image for stereo matching at time n, and the left image after correction at time n is taken as the left target image for stereo matching at time n. (Corresponding to...) Figure 1 The matching process on step ⑤ uses the sub-pixel grayscale matrix G. rn Using the epipolar constraint and a second-order inverse combination Gauss-Newton algorithm as the reference region, P is calculated. rn Matching point P′ on the left target image of stereo matching at time n. ln , and P rn and P′ ln The correlation between them corr rn_ln And calculate P′ ln With P ln The distance d between ln ;

[0071] Step 4.3.2: Stereo matching between the left image after correction at time n and the right image after correction at time n:

[0072] The left image after correction at time n is taken as the left reference image for stereo matching at time n, and the right image after correction at time n is taken as the right target image for stereo matching at time n, corresponding to... Figure 1 The matching process on step ⑥ involves the sub-pixel grayscale matrix G. ln Using the epipolar constraint and a second-order inverse combination Gauss-Newton algorithm as the reference region, P is obtained. ln The matching point P′ on the right target image of the stereo matching at time n. rn , and P ln and P′ rn The correlation between them is corr ln_rn And calculate P′ rn With P rn The distance d between rn ;

[0073] Step 4.3.3: Define two zero-based sign matrices, signL and signR, with the same dimensions as the grid.

[0074] Step 4.3.4: Determine corr l0_ln and corr r0_rn All are greater than the correlation threshold, and d ln and d rnIf the distance threshold is less than 1, it means that the corresponding two-dimensional coordinate points on the left and right images are correctly matched, and the element values ​​at the corresponding positions in the flag matrices signL and signR are updated to 1; otherwise, proceed to step 4.3.5.

[0075] Step 4.3.5: Determine corr l0_ln and corr ln_rn If all values ​​are greater than the correlation threshold, then P in the right reference image is considered valid. r0 At time n, a timing mismatch occurs on the right-hand graph because condition 4.3.4 is not met, indicating that at least one of the timing mismatches, either the left or right sequence, is incorrect. At this point, the distance d... ln and d rn They are all necessarily wrong, so use Figure 1 The correlation coefficients (corr) obtained from the matching in steps ③ and ⑥ above l0_ln and corr ln_rn If both are greater than the correlation threshold, it means that the temporal matching between the left images and the stereo matching from the left image to the right image are correct. Therefore, the temporal matching of the right image must be incorrect. Thus, P is set to... rn The coordinate values ​​are corrected to P′ rn If the coordinates are corrected, update the corresponding elements of the flag matrices signL and signR to 1, indicating that the correction has been made; otherwise, proceed to step 4.3.6.

[0076] Step 4.3.6: Determine corr r0_rn and corr rn_ln If all values ​​are greater than the correlation threshold, then P in the left reference image is considered valid. l0 At time n, the timing match of the left graph is incorrect, therefore P is... ln The coordinate values ​​are corrected to P′ ln If the coordinates are found, update the corresponding elements of the flag matrices signL and signR to 1, indicating that the correction has been made; otherwise, proceed to step 4.3.7.

[0077] Step 4.3.7: Determine whether the flag matrices signL and signR are all-one matrices. If they are, it means that all matching points obtained at time n have passed the verification and been successfully corrected, and proceed to step 4.3.8; otherwise, it means that the points obtained by temporal matching on the corrected left and right images at time n obtained in steps 4.1 and 4.2 have not been successfully verified and corrected. In this case, the corrected left and right images at time n / 2 are used as the reference images for the left temporal matching and the right temporal matching, respectively, and the process returns to step 4.1 to continue calculating the coordinates of the coordinate points that failed to match the temporal matching.

[0078] Step 4.3.8: Based on the corrected left and right image matching points and camera parameters at time n, reconstruct all three-dimensional coordinate points at time n using equation (1-4);

[0079] Step 5: After assigning n+1 to n, return to step 4.1 and execute sequentially until n>N, thereby obtaining the three-dimensional coordinate points at different times, which are used to calculate the displacement and strain when the three-dimensional coordinates change. N represents the total number of times.

[0080] In this embodiment, an electronic device includes a memory and a processor. The memory stores a program that supports the processor in executing the above-described method, and the processor is configured to execute the program stored in the memory.

[0081] In this embodiment, a computer-readable storage medium stores a computer program, which is executed by a processor to perform the steps of the above method.

Claims

1. A highly robust three-dimensional digital image correlation method for large deformation measurement, characterized in that, Includes the following steps: Step 1: Obtain the internal and external parameters of the left and right cameras using the Zhang Zhengyou calibration method. The internal parameters include: focal length, principal point coordinates, and lens distortion coefficient; the external parameters include: rotation matrix R and translation vector T between the left and right cameras. Step 2: The left and right cameras synchronously acquire left and right stretched and deformed images at N time points at fixed time intervals, and use the lens distortion coefficient to correct the left and right stretched and deformed images at each time point to obtain the corrected left and right images at each time point. Step 3: Using the corrected left image at time 0 as the reference image for stereo matching at time 0, and the corrected right image at time 0 as the target image for stereo matching at time 0, perform stereo matching, verification, and reconstruction of the three-dimensional coordinate points at time 0 on the left and right images at time 0. Step 3.1: Select the region of interest on the stereo matching reference image at time 0, and divide the region of interest into a grid. Each intersection point on the grid is used as a point to be calculated. Step 3.2: Use a point P to be calculated on the stereo matching reference image at time 0. l0 Using the center point as the reference sub-region at time 0, and selecting an M×M region around the center point as the reference sub-region, the point P to be calculated is obtained using equation (1-1). l0 The epipolar line l on the stereo-matched target image at time 0 pr : l pr =Fp l (1-1) In equation (1-1), p l For the point P to be calculated on the stereo matching reference image at time 0. l0 The homogeneous coordinates of F are given, where F is the fundamental matrix, and A l and A r Let R be the internal parameters of the left and right cameras, respectively; R is the rotation matrix; and S is the antisymmetric matrix related to the translation vector T of the left and right cameras. Select polar lines l in sequence pr Using each coordinate point on the image as the center, an area of ​​size M×M is selected around the center as the target sub-region for stereo matching at time 0. The zero-mean normalized cross-correlation function ZNCC is used to calculate the correlation between the reference sub-region at time 0 and the target sub-region for stereo matching at time 0. The center of the target sub-region with the highest correlation is taken as the point P to be calculated on the stereo matching reference image at time 0. l0 matching points; Step 3.3: Use the second-order inverse combination Gauss-Newton algorithm to calculate point P. l0 The matching points are optimized for accuracy to obtain the point P to be calculated. l0 Subpixel matching point P on the stereo matching target image at time 0 r0 and its surrounding sub-pixel grayscale matrix G of size M×M r0 ; Step 3.4: Match the sub-pixel matching points P on the stereo matching target image at time 0. r0 As a new point to be calculated, the accuracy is optimized using epipolar constraint coarse search and second-order inverse combination Gauss-Newton algorithm to obtain the new point P. r0 Matching point P′ on the stereo matching reference image at time 0 l0 ; Verify P l0 and P′ l0 The distance d between l0 If the distance d l0 If the value is less than the threshold, it indicates a correct match, and step 3.5 is executed; otherwise, the matching point P is discarded. l0 Then return to step 3.2 to continue processing the next point to be calculated; Step 3.5: Based on the principle of stereoscopic vision, use the point P to be calculated. l0 Sub-pixel matching point P r0 The coordinates of the points and the camera's intrinsic and extrinsic parameters are used to reconstruct a 3D coordinate point W0 at time 0; thus, the 3D coordinates of all correctly matched points at time 0 are obtained. Step 4: Calculate the three-dimensional coordinates of the points at each deformation moment; Step 4.0: Define the current time as n and initialize n = 1; use the corrected left image at time 0 as the left reference image; Step 4.1: Using the corrected left image at time n as the target image for time-series matching of the left sequence at time n, and performing time-series matching on the left reference image: Select the point P to be calculated on the left reference image. l0 Using a reference subregion of size M×M centered at the center, the point P to be calculated is obtained by a limited-range, interval-based search and a first-order inverse combination Gauss-Newton algorithm. l0 At time n, the matching point P on the target image of the left sequence temporal matching. ln , and P l0 and P ln The correlation between them corr l0_ln and P ln A sub-pixel grayscale matrix G with a surrounding size of M×M ln ; Step 4.2: Using the right image corrected at time 0 as the right reference image, and the right image corrected at time n as the target image for time-series temporal matching of the right sequence at time n, perform temporal matching: The sub-pixel grayscale matrix G on the right reference image r0 As a reference subregion, P is calculated using a limited-range, interval-based search and a first-order inverse combination Gauss-Newton algorithm. r0 At time n, the matching point P on the target image of the right sequence temporal matching. rn , and P r0 and P rn The correlation between them corr r0_rn and P rn A subpixel grayscale matrix G with a surrounding size of M×M rn ; Step 4.3: Closure check and correction between the left image after correction at time n and the right image after correction at time n; Step 5: After assigning n+1 to n, return to step 4.1 and execute sequentially until n>N, thereby obtaining the three-dimensional coordinate points at different times, which are used to calculate the displacement and strain when the three-dimensional coordinates change. N represents the total number of times.

2. The highly robust three-dimensional digital image correlation method for large deformation measurement according to claim 1, characterized in that, Step 4.3 includes the following steps: Step 4.3.1: Stereo matching between the right image corrected at time n and the left image corrected at time n: Using the right image corrected at time n as the right reference image for stereo matching at time n, and the left image corrected at time n as the left target image for stereo matching at time n, the sub-pixel grayscale matrix G is used. rn Using the epipolar constraint and a second-order inverse combination Gauss-Newton algorithm as the reference region, P is calculated. rn Matching point P′ on the left target image of stereo matching at time n. ln , and P rn and P′ ln The correlation between them corr rn_ln And calculate P′ ln With P ln The distance d between ln ; Step 4.3.2: Stereo matching between the left image after correction at time n and the right image after correction at time n: The left image after correction at time n is used as the left reference image for stereo matching at time n, and the right image after correction at time n is used as the right target image for stereo matching at time n. The sub-pixel grayscale matrix G ln Using the epipolar constraint and a second-order inverse combination Gauss-Newton algorithm as the reference region, P is obtained. ln The matching point P′ on the right target image of the stereo matching at time n. rn , and P ln and P′ rn The correlation between them is corr ln_rn And calculate P′ rn With P rn The distance d between rn ; Step 4.3.3: Define two all-zero flag matrices signL and signR, and the dimensions of the flag matrices signL and signR are the same as the dimensions of the grid. Step 4.3.4: Determine corr l0_ln and corr r0_rn All are greater than the correlation threshold, and d ln and d rn If the distance threshold is less than 1, it means that the corresponding two-dimensional coordinate points on the left and right images are correctly matched, and the element values ​​at the corresponding positions in the sign matrices signL and signR are updated to 1. Otherwise, proceed to step 4.3.5; Step 4.3.5: Determine corr l0_ln and corr ln_rn If all values ​​are greater than the correlation threshold, then P in the right reference image is considered valid. r0 There is a timing mismatch in the right-hand graph at time n, therefore P is... rn The coordinate values ​​are corrected to P′ rn The coordinate values ​​are updated, and the corresponding elements in the flag matrices signL and signR are updated to 1, indicating that the correction has been made. Otherwise, proceed to step 4.3.6; Step 4.3.6: Determine corr r0_rn and corr rn_ln If all values ​​are greater than the correlation threshold, then P in the left reference image is considered valid. l0 At time n, the timing match of the left graph is incorrect, therefore P is... ln The coordinate values ​​are corrected to P′ ln The coordinate values ​​are updated, and the corresponding element values ​​in the flag matrices signL and signR are updated to 1, indicating that the correction has been made. Otherwise, proceed to step 4.3.7; Step 4.3.7: Determine whether the flag matrices signL and signR are all-one matrices. If they are, it means that all matching points obtained at time n have passed the verification and been successfully corrected, and proceed to step 4.3.8; otherwise, it means that the matching points on the corrected left and right images obtained in steps 4.1 and 4.2 at time n have not been successfully verified and corrected. In this case, the corrected left and right images at time n / 2 are used as the reference images for the left temporal matching and the right temporal matching, respectively, and the process returns to step 4.1 to continue calculating the coordinates of the coordinate points where the temporal matching failed. Step 4.3.8: Reconstruct all three-dimensional coordinate points at time n based on the matching points of the left and right images after correction and the camera parameters.

3. An electronic device, comprising a memory and a processor, characterized in that, The memory is used to store a program that supports the processor in executing the highly robust three-dimensional digital image correlation method of claim 1 or 2, the processor being configured to execute the program stored in the memory.

4. A computer-readable storage medium storing a computer program thereon, characterized in that, The computer program, when run by a processor, performs the steps of the highly robust three-dimensional digital image correlation method according to claim 1 or 2.

Images