Three-dimensional measurement method and system for chemical milling adhesive film profiling by coupling gray-scale and phase information

Abstract

The present invention discloses a three-dimensional measurement method for chemical milling adhesive film profiling that couples gray-scale and phase information, belonging to the field of three-dimensional measurement. First, obtain the uniformly illuminated gray-scale images and grating images of the left and right cameras as initial data, and obtain the corrected left-camera gray-scale image, the phase images of the left and right cameras, and the mapped gray-scale image of the right-camera gray-scale image onto the left-camera gray-scale image; then, based on the left-camera gray-scale image, perform preliminary image segmentation based on gray-scale gradient to extract the initial contour; next, construct an external energy field and a gradient vector field, and combine with the internal energy field to construct an overall energy function to obtain a sub-pixel accuracy contour; finally, realize stereo matching and three-dimensional reconstruction of sub-pixel points of the contour based on phase information. Compared with traditional passive binocular measurement methods, it can not only stably extract the center line of the profiling pattern under interference such as oil stains and scratches, but also overcome the problem of inaccurate center line extraction caused by factors such as uneven gray-scale changes and inconsistent viewing angles of the left and right cameras.

Description

Technical Field

[0001] The present invention belongs to the field of three-dimensional measurement, and more specifically, relates to a three-dimensional measurement method and system for chemical milling film profiling by coupling gray-scale and phase information. Background Technique

[0002] The chemical milling (chemical machining) film profiling directly affects the machining accuracy of structures such as bosses and stiffeners, and is one of the key bottleneck technologies determining the manufacturing quality of thin-walled metal components in aerospace and other fields. Measuring the machining pattern obtained from chemical milling film profiling can not only meet the machining accuracy detection requirements during the production process, but also provide basic data for subsequent trajectory planning, profiling process parameter optimization, etc., and is an important method to ensure the chemical milling machining accuracy.

[0003] The existing profiling pattern detection methods are mainly divided into two categories: contact type and non-contact type. Contact measurement mainly refers to template comparison, where the profiled part is manually brought into contact and aligned with the machining template for comparison. It has low efficiency and can only visually judge whether the part is qualified, and cannot accurately obtain error data. Non-contact measurement mainly includes passive binocular reconstruction and three-dimensional point cloud analysis. Three-dimensional point cloud analysis means first reconstructing the complete three-dimensional morphology of the profiled part, and then comparing and analyzing it with the standard template. It can only detect point clouds with larger depth and width, and it is difficult to accurately extract the three-dimensional centerline point cloud data, unable to meet the measurement requirements of the profiling pattern. Passive binocular reconstruction means using the binocular vision principle to directly match and reconstruct by extracting the centerline contour on the left and right camera images, and three-dimensional centerline point clouds can be obtained for accuracy judgment, which is more intuitive and specific compared to three-dimensional point cloud analysis. However, at present, affected by factors such as the accuracy and stability of the centerline extraction of the profiling pattern contour and corresponding point matching, the reconstruction accuracy and integrity of this method still need to be improved.

[0004] In response to the above problems, many scholars at home and abroad have conducted relevant research. In terms of laser engraving pattern contour extraction, Yang et al. (Yang et al., 2020) proposed a method to remove false target edges based on prior information, and used the normal direction gradient weighted contour of the initially extracted contour pixel points as the final contour, which improved the accuracy and stability of the engraving pattern contour extraction to a certain extent. Zhang et al. (Zhang et al., 2022) constructed an optical path system based on the side light source for the reflective characteristics of aircraft skins, and used the multiple exposure fusion algorithm to improve the image quality to reduce the difficulty of contour extraction. Tong et al. (Tong et al., 2019) aimed at the actual measurement requirements of aircraft skin engraving patterns. First, they extracted the two edges of the engraving pattern contour, and then took the average value in the normal direction as the center line contour, which was more accurate than directly using the Canny algorithm to extract the two side contours for reconstruction. He et al. (He et al., 2019) proposed a fast contour extraction algorithm based on phase guidance. According to the principle that the phase changes due to the different depths before and after 3D printing, the contour was quickly located, and then the low-rank matrix decomposition was used to reduce noise, effectively improving the speed of on-line measurement in the 3D printing scenario, but the accuracy still needed to be improved. In terms of contour corresponding point matching, Yang et al. (Yang et al., 2020) calculated the matching point pairs according to the left-right consistency and neighborhood distance minimization criteria, which improved the accuracy of contour corresponding point matching while ensuring the continuity of the reconstructed contour point cloud. Zhang et al. (Zhang et al., 2022) calculated the disparity map based on ZNCC (zero-normalized cross-correlation), which improved the robustness of contour stereo matching. Xie et al. (Xie et al., 2019) divided the contour into left and right parts for separate matching to remove the problem that the upper and lower edge points could not be accurately matched, which could meet the measurement requirements of parts with significant features such as round holes, but it was difficult to meet the measurement requirements of various irregular contours in laser engraving patterns. Ju et al. (Ju et al., 2015) first extracted the contour of the left camera image, and then directly searched for the corresponding contour on the right camera image according to constraints such as contour continuity and local consistency, which could ensure the continuity and integrity of contour matching and was applicable to scenarios where the surrounding changes were large, but it was difficult to be applicable to scenarios where the surrounding areas were basically the same, such as laser engraving patterns. Hu et al. (Xiaotong Hu and Shaoying Guo, 2016) first searched for the contour of the right camera image according to the contour of the left camera image, and then verified whether the matching points were accurate based on the inverse essential matrix, effectively removing some false matching points.

[0005] Generally speaking, the above - mentioned method can improve the accuracy and stability of the three - dimensional reconstruction of the chemical - milling adhesive film engraving pattern to a certain extent, but there are still problems in the following aspects: 1) The accuracy of extracting the center line of the engraving pattern is poor. The current method still uses grayscale to extract the center line of the pattern, which is easily affected by factors such as viewing angle, illumination, and inconsistent grayscale in the surrounding area, resulting in deviation in the extraction of the center line. At the same time, the grayscale center line only represents the center line of the pattern visually, and there may be a deviation between it and the center line of the pattern in the actual physical sense, affecting the measurement accuracy. 2) The extraction of the left - and - right camera contours is inconsistent. Affected by factors such as different viewing angles and illuminations of the left - and - right cameras, there are differences in the corresponding contours in the left - and - right camera images. Extracting the left - and - right image contours separately relying on grayscale information will surely lead to inconsistent contour extraction, affecting the matching accuracy of subsequent corresponding contour points. 3) The reconstruction accuracy at the edges of the upper - and - lower contours is poor. When performing stereo matching at the edges of the upper - and - lower contours, there are multiple similar matching point pairs. It is difficult to achieve accurate matching relying only on constraints such as epipolar lines and features, resulting in severe warping and deformation of the reconstructed point cloud.

[0006] Therefore, there is an urgent need for a high - precision three - dimensional measurement method for chemical - milling adhesive film engraving to improve the detection efficiency and accuracy and provide basic data for the process optimization of chemical - milling engraving. Summary of the Invention

[0007] In view of the above - mentioned defects or improvement requirements of the prior art, the present invention provides a three - dimensional measurement method for chemical - milling adhesive film engraving that couples grayscale - phase information to ensure the accuracy and stability of the measurement of chemical - milling engraving patterns.

[0008] To achieve the above object, a three - dimensional measurement method for chemical - milling adhesive film engraving that couples grayscale - phase information is provided, including:

[0009] S1. First, obtain the uniform - illumination grayscale images and grating grayscale images of the left - and - right cameras as initial data. Then, based on distortion correction and epipolar correction, obtain the corrected left - and - right camera grayscale images and left - and - right camera phase images. Then, map the right - camera grayscale image to the left - camera based on phase matching.

[0010] S2. Taking the left - camera grayscale image as the reference, first perform preliminary image segmentation based on grayscale gradient to obtain the initial area of the chemical - milling engraving pattern. Then, perform image dilation to remove holes. Then, perform screening based on connected components to remove discontinuous and noisy areas. Finally, extract the skeleton of the connected component as the initial contour.

[0011] S3. Construct an external energy field based on the left - camera grayscale image, the left - camera phase image, and the mapped image of the right - camera grayscale image on the left - camera. Then, construct a gradient vector field based on the external energy field, and combine it with the internal energy field to construct an overall energy function. Iteratively solve for the minimum energy to obtain a sub - pixel - accuracy contour.

[0012] S4. Realize sub - pixel point stereo matching and three - dimensional reconstruction of the contour based on phase information.

[0013] Preferably, the right camera grayscale image is mapped to the left camera based on phase matching, specifically:

[0014] S11. Traverse each pixel of the left camera grayscale image in the order from top to bottom and from left to right, and find the corresponding matching points on the right camera grayscale image according to the phase information;

[0015] S12. After finding the corresponding points, perform bilinear interpolation according to the sub-pixel coordinates to obtain the corresponding grayscale values on the right camera grayscale image;

[0016] S13. After the traversal is completed, obtain the mapped image of the right camera grayscale image on the left camera.

[0017] Preferably, an external energy field is constructed based on the left camera grayscale image, the left camera phase image, and the mapped image of the right camera grayscale image on the left camera, specifically:

[0018] E GP ex = ω left E GP exl + ω right E GP exr

[0019] where ω left and ω right are the weights of E GP exl and E GP exr respectively; E GP exl represents the external energy of the contour curve on the left camera grayscale image; E GP exr represents the external energy of the right camera grayscale image corresponding to the contour of the left camera grayscale image.

[0020] Preferably, the external energy of the contour curve on the left camera grayscale image is specifically:

[0021] E GP exl = ∫ε left (X(s))ds == ∫ -[ω l ε l + ω e ε e + ω t ε t + ω p ε p ds

[0022] where X(s) represents the two-dimensional coordinate points of the contour, and ω l and ω e, ω t , ω p are respectively the weight coefficients corresponding to ε l , ε e , ε t , ε p .

[0023] ε l represents the image gray - level information, attracting the contour to the position with the maximum gray - level. Specifically:

[0024] ε l =I

[0025] ε t represents the end - point information of the gray - level image, attracting the contour to the line - end point or corner point. Specifically, it is expressed as:

[0026]

[0027] Is = G σ *I

[0028] where the subscripts x and y respectively represent the derivatives in the x and y directions;

[0029] ε e represents the second - order gray - level gradient amplitude information of the image, attracting the contour to the position with the maximum second - order gray - level gradient amplitude. Specifically:

[0030] ε e =-||▽||G' σ *I||2||2

[0031] ε p represents the second - order phase - gradient information, attracting the contour to the position with the maximum second - order phase - gradient. Specifically:

[0032] ε p =-||▽||G' σ *P||2||2

[0033] where I represents the gray - level image, Is represents the gray - level image after Gaussian filtering, G σ represents the Gaussian kernel, σ represents the variance, and P represents the phase map.

[0034] Preferably, the external energy of the contour curve on the gray - level image of the right camera is specifically:

[0035] E GP exr =∫ε right (X(s))ds = ∫-[ω l ε l +ω e ε e +ω t ε tds

[0036] The parameters in the formula are consistent with the external energy of the contour curve on the left camera grayscale image. Since the external energy of the contour curve on the right camera grayscale image is mainly calculated based on the mapped image (the result calculated based on phase) of the right camera grayscale image on the left camera, there is one less phase term compared to the external energy of the left camera image.

[0037] Preferably, a gradient vector field is constructed based on the external energy field, specifically:

[0038]

[0039] where V(x, y) = [u(x, y), v(x, y)]; represents the smoothing term, making the vector field change smoothly; μ is the weight; |▽ε GP | 2 (V - ▽ε GP ) 2 represents the data term of the vector field, used to approach ▽ε GP . It can be seen that when ▽ε GP is small, the vector flow field is determined by the smoothing term, so the change is relatively smooth; when ▽ε GP is large, the energy is minimized when V = ▽ε GP .

[0040] Preferably, the internal energy field of the contour curve, specifically:

[0041]

[0042] where α(s) and β(s) are weighting coefficients; f s , f ss are the first derivative and the second derivative of X(s) respectively, representing the contour curve length change rate and the curvature vector. α(s)|f s | 2 is mainly used to control the contraction of the contour curve. The larger α is, the faster the contour curve contracts. β(s)|f ss | 2 is mainly used to control the moving speed of the contour curve along the normal direction. When β is large, the contour curve is not easily bent and has a greater intensity. Therefore, when the internal energy of the contour curve is made smaller, the contour curve will contract and become smoother.

[0043] Preferably, the overall energy function, specifically:

[0044] E GP (C) = E GP in + E GP ex

[0045] Among them, E GP (C) represents the gray-phase energy of the contour curve; E GP in represents the internal energy of the contour curve; E GP ex represents the external gray-phase energy of the contour curve; C is the contour curve, defined as the set of X(s) = (x(s), y(s)), where X(s) represents the two-dimensional coordinate points of the contour and s is the normalized arc length.

[0046] On the other hand, the present invention provides a three-dimensional measurement system for chemical milling adhesive film profiling that couples gray-phase information, including: a computer-readable storage medium and a processor;

[0047] The computer-readable storage medium is used to store executable instructions;

[0048] The processor is used to read the executable instructions stored in the computer-readable storage medium and execute the above-mentioned three-dimensional measurement method for chemical milling adhesive film profiling that couples gray-phase information.

[0049] Generally speaking, compared with the prior art through the above technical solutions conceived by the present invention, the following beneficial effects can be achieved:

[0050] 1. Based on the analysis of gray-phase changes, the present invention comprehensively considers the gray change and phase change of the camera to construct an external energy function. Combining with the internal energy function of the contour curve, it can realize the stable extraction of the contour under the interference of oil stains, scratches, etc., while ensuring the smoothness of the contour;

[0051] 2. The present invention constructs a gray mapping diagram of the left and right cameras based on phase matching to comprehensively consider the gray information of the left and right cameras, construct a more accurate external energy field, and overcome the problem of inaccurate centerline extraction caused by factors such as uneven gray change and inconsistent viewing angles of the left and right cameras.

[0052] 3. The present invention builds a three-dimensional reconstruction framework for the chemical milling adhesive film profiling pattern contour, including data preprocessing, preliminary contour positioning, sub-pixel positioning, and three-dimensional reconstruction, and realizes the automatic three-dimensional reconstruction of the chemical milling adhesive film profiling pattern contour. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0054] Figure 1Flow chart of three-dimensional contour reconstruction by coupling gray-scale and phase information of the present invention;

[0055] Figure 2 In (a) and (b) of are respectively schematic diagrams of the composition of the processing areas of test specimens 1 and 2;

[0056] Figure 3 are a grating image and a phase image. (a) is the first grating image and (b) is the absolute phase image;

[0057] Figure 4 In (a) and (b) of are respectively the left and right camera gray-scale images after distortion correction and epipolar correction; (c) is the gray-scale image of the right camera mapped to the corrected left camera; (d) is the local enlarged view in (a) and (b);

[0058] Figure 5 In (a) of is the preliminary segmentation map based on gray-scale gradient; (b) is the dilated map of the preliminary segmentation map; (c) is the contour screening map in (b); (d) is the preliminary contour map;

[0059] Figure 6 In (a) and (b) of are respectively schematic diagrams of simulated gray-scale changes in two cases where the affected area is less than and greater than the unchanged area, and their corresponding first derivative schematic diagrams and second derivative schematic diagrams;

[0060] Figure 7 In (a), (b), and (c) of are respectively the actual gray-scale image of specimen 2, its first-order gray-scale gradient amplitude map, and second-order gray-scale gradient amplitude map;

[0061] Figure 8 In (a), (b), and (c) of are respectively the simulated phase map and its first derivative schematic diagram and second derivative schematic diagram;

[0062] Figure 9 In (a), (b), and (c) of are respectively the actual phase map of specimen 2 and its first-order gradient amplitude map and second-order gradient amplitude map;

[0063] Figure 10 Schematic diagram of the gradient vector field;

[0064] Figure 11 Schematic diagram of iteratively obtaining the sub-pixel accuracy contour;

[0065] Figure 12 Schematic diagram of the three-dimensional contour point cloud from different perspectives;

[0066] Figure 13 is the contour extraction result map of various types of engraved pattern. (a)-(f) respectively represent the contour extraction results of different styles of engraved patterns;

[0067] Figure 14 Schematic diagram of 3D reconstruction results of various types of contours. (a)-(f) are respectively Figure 13 the 3D reconstruction results of the corresponding contours in (a)-(f) in

[0068] Figure 15 Verification diagram of 3D contour measurement accuracy. (a) is the true value schematic diagram of the fillet rectangle workpiece; (b) is the schematic diagram of measuring point cloud analysis; (c) is the statistical chart of measurement results. Specific implementation manners

[0069] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0070] An embodiment of the present invention provides a three-dimensional measurement method for chemical milling adhesive film profiling that couples gray-scale and phase information. The specific process is as Figure 1 shown and includes:

[0071] S1. Project a uniform image and a grating image onto the surface of the part to be measured (the part to be measured is as Figure 2 shown), obtain the uniform illumination images and grating images of the left and right cameras as initial data, then obtain the corrected gray-scale image of the left camera and the corrected phase images of the left and right cameras based on distortion correction and epipolar correction, and then map the gray-scale image of the right camera to the left camera based on phase matching.

[0072] Specifically, the three-frequency four-step method is used to calculate the phase. Therefore, 26 images (a total of 2*12 gratings + 2 uniform illuminations are obtained by the left and right cameras) are first obtained as initial data, as Figure 3 shown; then relative phase calculation and phase unwrapping are performed on this basis to obtain the absolute phase map; then, based on the pre-calibrated parameters of the left and right cameras, distortion correction and epipolar correction are performed on the phase map and the gray-scale image to obtain the corrected phase maps and gray-scale corrected maps of the left and right cameras; finally, based on the principle of phase matching, the gray-scale image of the right camera is mapped onto the corrected left camera, as Figure 4 shown.

[0073] Specifically, mapping the gray-scale image of the right camera to the left camera based on phase matching, the specific process is:

[0074] S11. Traverse each pixel of the left camera image in the order from top to bottom and from left to right, and find the corresponding matching points on the right camera image according to the phase information;

[0075] S12. After finding the corresponding points, perform bilinear interpolation according to the sub-pixel coordinates to obtain the corresponding gray-scale value on the right camera image;

[0076] S13. After the traversal ends, obtain the mapping image of the right camera grayscale image on the left camera.

[0077] S2. Based on the left camera image as a reference, first perform preliminary image segmentation based on the grayscale gradient to obtain the initial area of the chemical milling pattern, then perform image dilation to remove holes, then perform screening based on connected components to remove discontinuous and noise areas, and finally extract the connected component skeleton as the initial contour.

[0078] Specifically, first perform threshold segmentation based on the grayscale gradient to obtain the initial positions of the processing area and the affected area, as shown in (a) of Figure 5 . On this basis, use the image dilation algorithm to fill areas such as the central hole and the edge hole, as shown in (b) of 5. Then calculate the connected regions in the image, and remove non-contour regions based on constraints such as prior size and ratio, as shown in Figure 5 . Then use the bwskel function to extract the center skeleton of the binary image and sort it in the counterclockwise direction as the initial contour, as shown in Figure 5 .

[0079] S3. Construct an external energy field based on the left camera grayscale image, the left camera phase image, and the mapping image of the right camera grayscale image on the left camera. Then construct a gradient vector field based on the external energy field, combine the internal energy field to construct an overall energy function, and iteratively solve for the minimum energy to obtain a sub-pixel accuracy contour.

[0080] Specifically, construct an external energy field based on the left camera grayscale image, the left camera phase image, and the mapping image of the right camera grayscale image on the left camera, specifically:

[0081] E GP ex = ω left E GP exl + ω right E GP exr

[0082] Among them, ω left and ω right are the weights of E GP exl and E GP exr , and are set to 1 and 0.5 respectively; E GP exl represents the external energy of the contour curve on the left camera image; E GP exr The external energy of the right camera image corresponding to the left camera image contour.

[0083] Specifically, the external energy of the contour curve on the left camera image is specifically as follows:

[0084] E GP exl =∫ε left (X(s))ds==∫-[ω l ε l +ω e ε e +ω t ε t +ω p ε p ds

[0085] Among them, ω l 、ω e 、ω t 、ω p are the weight coefficients corresponding to ε l 、ε e 、ε t 、ε p respectively, and are set to 0.1, 0.6, 0.1, 0.4 respectively. The reasons for choosing these parameters to construct the energy function are as Figures 6-9 shown. At the center line of the chemical milling adhesive film, the contour gray value is the largest, the second-order gray gradient is the smallest, and the second-order phase gradient is also the largest.

[0086] ε l represents the image gray information and attracts the contour to the position with the largest gray value. Specifically, it is:

[0087] ε l =I(x,y)

[0088] ε t represents the gray image end point information and attracts the contour to the line end point or corner point. Specifically, it is expressed as:

[0089]

[0090] Is=G σ *I(x,y)

[0091] ε e represents the image second-order gray gradient amplitude information and attracts the contour to the position with the largest second-order gray gradient amplitude. Specifically, it is:

[0092] ε e =-||▽||G' σ *I(x,y)||2||2

[0093] ε p represents the second-order phase gradient information and attracts the contour to the position with the largest second-order phase gradient. Specifically, it is:

[0094] εp = - ||▽||G' σ *P(x, y)||2||2

[0095] Where, I(x, y) represents the grayscale image, P(x, y) represents the phase image, Is represents the grayscale image after Gaussian filtering, G σ represents the Gaussian kernel, σ represents the variance, and is set to 1.

[0096] Specifically, the external energy of the contour curve on the right camera image is specifically:

[0097] E GP exr = ∫ε right (X(s))ds = ∫ - [ω l ε l + ω e ε e + ω t ε t ds

[0098] Where, the parameters are the same as the external energy of the contour curve on the left camera image. Since the external energy of the contour curve on the right camera image is mainly calculated based on the mapped image (the result based on phase calculation) of the right camera grayscale image on the left camera, there is one less phase term compared to the external energy of the left camera image.

[0099] Specifically, a gradient vector field is constructed based on the external energy field, as Figure 10 shown, specifically:

[0100]

[0101] Where, V(x, y) = [u(x, y), v(x, y)]; represents the smoothing term, making the vector field change smoothly; μ is the weight; |▽ε GP | 2 (V - ▽ε GP ) 2 represents the data term of the vector field, used to approach ▽ε GP . It can be seen that when ▽ε GP is small, the vector flow field is determined by the smoothing term, so the change is relatively smooth; when ▽ε GP is large, the energy is minimized when V = ▽ε GP .

[0102] Specifically, the internal energy field of the contour curve is specifically:

[0103]

[0104] Where, α(s), β(s) are weighting coefficients, and are set to 0.1 and 0.2 respectively; fs , f ss are the first derivative and the second derivative of X(s), representing the contour curve length change rate and the curvature vector, respectively. α(s)|f s | 2 is mainly used to control the contraction of the contour curve. The larger α is, the faster the contour curve contracts. β(s)|f ss | 2 is mainly used to control the movement speed of the contour curve along the normal direction. When β is larger, the contour curve is not easily bent and has a greater strength. Therefore, when the internal energy of the contour curve becomes smaller, the contour curve will contract and become smoother.

[0105] Specifically, the overall energy function is:

[0106] E GP (C) = E GP in + E GP ex

[0107] where E GP (C) represents the gray-phase energy of the contour curve; E GP in represents the internal energy of the contour curve; E GP ex represents the external gray-phase energy of the contour curve; C is the contour curve, defined as the set of X(s) = (x(s), y(s)), where X(s) represents the two-dimensional coordinate points of the contour and s is the normalized arc length.

[0108] Specifically, the steps for minimizing the overall energy function E GP are as follows:

[0109] Simplify the overall energy function E GP :

[0110]

[0111] where ε GP (X(s)) = ω left ε left (X(s)) + ω right ε right (X(s)), ω left is set to 1, and ω right is set to 0.5.

[0112] Then, according to the variational principle, minimize the contour energy to obtain the Euler-Lagrange equation, which is specifically as follows:

[0113]

[0114] can be simplified to:

[0115] αX” - βX”' - ∇ε GP = F in + F ex = 0

[0116] The above equation is actually an internal - external force balance equation, where the internal force refers to F in = αX” - βX”', which controls the stretching and bending of the contour curve, and the external force controls the movement of the contour curve towards the ideal center line.

[0117] To represent the iterative solution process, it is written in the form of motion with respect to time t, specifically as:

[0118] γX t (s,t) = αX”(s) - βX”'(s) - ∇ε GP (s)

[0119] where γ represents the step size, and X t represents the derivative of X with respect to t. The specific results are as Figure 11 shown.

[0120] S4. Realize sub - pixel point stereo matching and 3D reconstruction of the contour based on phase information.

[0121] Specifically, based on the sub - pixel point coordinates of the contour, calculate the absolute phase value on the absolute phase map of the left camera; then perform matching on the absolute phase map of the right camera based on this to find the corresponding points; finally, reconstruct the 3D contour point cloud based on the corresponding points according to the principle of triangulation. The specific contour extraction results and 3D reconstruction results are as Figures 12-14 shown. The measurement accuracy of the current method is as Figure 15 shown, where L, W, R, and Std of fitting are the length, width, fillet radius, and fitting deviation of the rectangle respectively.

[0122] Those skilled in the art can easily understand that the above - mentioned are only the preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention should be included within the protection scope of the present invention.

Claims

1. A three-dimensional measurement method for chemical milling adhesive film profiling that couples gray-scale and phase information, characterized in that, Including: S1. First, obtain the uniformly illuminated grayscale images and grating grayscale images of the left and right cameras as initial data. Then, based on distortion correction and epipolar correction, obtain the corrected grayscale images and phase images of the left and right cameras. Next, map the grayscale image of the right camera to the left camera based on phase matching; S2. Using the grayscale image of the left camera as a reference, first perform preliminary image segmentation based on grayscale gradient to obtain the initial area of the chemical milling pattern. Then, perform image dilation to remove holes. Next, perform screening based on connected components to remove discontinuous and noisy areas. Finally, extract the skeleton of the connected components as the initial contour; S3. Construct an external energy field E based on the grayscale image of the left camera, the phase image of the left camera, and the mapped image of the grayscale image of the right camera on the left camera. GP ex , specifically: E GP ex = ω left E GP exl + ω right E GP exr Among them, ω left and ω right are the weights of E GP exl and E GP exr respectively. E GP exl represents the external energy of the contour curve on the left camera grayscale image, and E GP exr is the external energy of the right camera grayscale image corresponding to the contour of the left camera grayscale image. The external energy of the contour curve on the left camera grayscale image is specifically: E GP exl = ∫ε left (X(s))ds == ∫ -[ω l ε l + ω e ε e + ω t ε t + ω p ε p ds The external energy of the grayscale image of the right camera corresponding to the contour of the grayscale image of the left camera is specifically: E GP exr = ∫ε right (X(s))ds = ∫ -[ω l ε l + ω e ε e + ω t ε t ds Among them, X(s) represents the two-dimensional contour coordinate points, s is the normalized arc length, ω l 、ω e 、ω t 、ω p are the weight coefficients corresponding to ε l 、ε e 、ε t 、ε p respectively. ε l represents the image gray-scale information, ε t represents the end-point information of the gray-scale image, ε e represents the magnitude information of the second-order gray-scale gradient of the image, and ε p represents the second-order phase gradient information; Then, construct a gradient vector field based on the external energy field, combine it with the internal energy field to construct an overall energy function, and iteratively solve for the minimum energy to obtain a sub-pixel accurate contour; S4. Implement sub-pixel point stereo matching and 3D reconstruction of the contour based on phase information.

2. The method according to claim 1, wherein Mapping the grayscale image of the right camera to the left camera based on phase matching specifically includes: traversing each pixel of the grayscale image of the left camera, searching for the corresponding matching point on the grayscale image of the right camera according to the phase information, and then performing bilinear interpolation according to the sub-pixel coordinates to obtain the corresponding grayscale value on the grayscale image of the right camera; finally, after the traversal ends, obtain the mapping image of the grayscale image of the right camera on the left camera.

3. The method according to claim 1, wherein ε l Represents the grayscale information of the image, attracting the contour to the position with the maximum grayscale, specifically: ε l = I ε t Indicates the end point information of the grayscale image, attracting the contour to the line end point or corner point, specifically expressed as: Is = G σ *I where the subscripts x and y respectively represent the derivatives in the x and y directions; ε e represents the second-order gray-scale gradient amplitude information of the image, attracting the contour to the position with the maximum second-order gray-scale gradient amplitude, specifically: ε p represents the second-order phase gradient information, attracting the contour to the position where the second-order phase gradient is the largest, specifically: Among them, I represents the grayscale image, Is represents the grayscale image after Gaussian filtering, G σ represents the Gaussian kernel, σ represents the variance, and P represents the phase diagram.

4. The method according to claim 1, wherein The internal energy of the contour curve is specifically: where α(s) and β(s) are weighting coefficients; f s and f ss are the first derivative and the second derivative of X(s), respectively, representing the length change rate of the contour curve and the curvature vector.

5. The method according to claim 4, characterized in that The overall energy function is specifically: E GP (C) = E GP in + E GP ex Among them, E GP (C) represents the gray-phase energy of the contour curve; E GP in represents the internal energy of the contour curve; E GP ex represents the external gray-phase energy of the contour curve; C is the contour curve, defined as the set of X(s) = (x(s), y(s)), where s is the normalized arc length.

6. A three-dimensional measurement system for chemically milled adhesive film profiling that couples gray-scale and phase information, characterized in that, Including: A computer-readable storage medium and a processor; The computer-readable storage medium is used to store executable instructions; The processor is used to read the executable instructions stored in the computer-readable storage medium and execute the chemical milling film pattern three-dimensional measurement method with coupled grayscale-phase information according to any one of claims 1 to 5.

Images