Flexible circuit board three-dimensional shape laser scanning measurement and reconstruction method and system

Abstract

The invention provides a flexible circuit board three-dimensional shape laser scanning measurement and reconstruction method and system, and relates to the technical field of three-dimensional measurement, and the method comprises the steps: obtaining a strain distribution diagram of a flexible circuit board through applying a preset stress field and constructing a strain monitoring point array, and dividing a deformation sensitive region; arranging a plurality of groups of binocular laser scanning devices based on the distribution of the deformation sensitive area, collecting an orthogonal grid projection image, and obtaining a space mapping relation matrix through a self-adaptive multi-step calibration algorithm; constructing a depth information mapping model in combination with the laser stripe center line, and generating initial three-dimensional point cloud data; point cloud registration fusion is carried out through a local feature preserving algorithm, noise reduction and abnormal point elimination are carried out through a local curvature adaptive algorithm, and optimized point cloud data are obtained; and block fitting is carried out based on the deformation sensitive area, a fitting result is optimized and corrected by combining the strain distribution diagram and the curvature distribution, and a three-dimensional geometrical morphology of the flexible circuit board is generated.

Description

Technical Field

[0001] The present invention relates to the field of three-dimensional measurement technology, and in particular to a three-dimensional shape laser scanning measurement and reconstruction method and system for a flexible circuit board. Background Art

[0002] Flexible circuit boards are widely used in smart wearables, aerospace and other fields due to their thinness and bendability. In practical applications, flexible circuit boards are prone to deformation, and the accurate measurement and reconstruction of their three-dimensional shape is of great significance to product quality control and reliability assessment. At present, the commonly used three-dimensional shape measurement methods include contact measurement and non-contact measurement. Among them, non-contact measurement based on binocular laser scanning has gradually become the main method for three-dimensional shape measurement of flexible circuit boards due to its non-destructive and efficient characteristics.

[0003] However, the existing measurement methods have the following shortcomings: the traditional scanning measurement adopts a fixed arrangement, which is difficult to adapt to the complex deformation characteristics of flexible circuit boards, resulting in measurement blind areas and data loss; the existing point cloud data processing method does not consider the particularity of deformation-sensitive areas, and adopts a unified noise reduction and reconstruction strategy, resulting in distortion of geometric features in key areas; conventional surface fitting algorithms lack consideration of material mechanical properties and cannot guarantee the consistency of reconstruction results with actual deformation states. These problems seriously affect the accuracy and reliability of flexible circuit board 3D topography reconstruction.

[0004] In summary, there is an urgent need for a method for measuring and reconstructing the three-dimensional shape of a flexible circuit board based on deformation-sensitive areas by laser scanning. The deformation-sensitive areas are determined by monitoring the strain distribution, the arrangement position of the laser scanning device is optimized, and the data acquisition quality is improved by using an adaptive exposure strategy; the point cloud data is optimized by combining the local feature preservation algorithm and the curvature adaptive algorithm; block fitting is performed based on the deformation-sensitive areas, and optimization correction is performed in combination with the strain distribution information, so as to achieve high-precision reconstruction of the three-dimensional shape of the flexible circuit board. The present invention can solve the problems in the prior art. Summary of the invention

[0005] The embodiment of the present invention provides a method and system for laser scanning measurement and reconstruction of the three-dimensional shape of a flexible circuit board, which can solve the problems in the prior art.

[0006] According to a first aspect of the embodiments of the present invention,

[0007] A method for measuring and reconstructing the three-dimensional shape of a flexible circuit board by laser scanning is provided, comprising:

[0008] A preset stress field is applied to the flexible circuit board to be measured, a gridded strain monitoring point array is constructed on the surface of the flexible circuit board, a strain distribution map of the flexible circuit board is obtained through a multispectral camera, and deformation sensitive areas are divided based on the strain distribution map; based on the distribution of the deformation sensitive areas, the optimal arrangement positions of four groups of binocular laser scanning devices are calculated, and an orthogonal grid projection image group of the flexible circuit board is collected using an adaptive exposure control strategy;

[0009] Input the calibration data of the four groups of binocular laser scanning devices into an adaptive multi-step calibration algorithm to obtain a spatial mapping relationship matrix; perform double Gaussian fitting on the orthogonal grid projection image group to extract the center line of the laser stripe, combine the center line of the laser stripe with the spatial mapping relationship matrix to construct a depth information mapping model, calculate the three-dimensional coordinates according to the depth information mapping model, generate initial three-dimensional point cloud data, and resample and align and fuse the initial three-dimensional point cloud data through a local feature preservation algorithm to obtain fused point cloud data;

[0010] The fused point cloud data is input into a local curvature adaptive algorithm for noise reduction and abnormal point removal to obtain optimized point cloud data; the surface of the flexible circuit board is divided into multiple characteristic sub-areas according to the deformation sensitive area, and the characteristic sub-areas are fitted in blocks using an adaptive spline surface to determine the fitting results; the curvature distribution of the flexible circuit board is calculated based on the optimized point cloud data, and the fitting results are optimized and corrected in combination with the strain distribution map to generate the three-dimensional geometric morphology of the flexible circuit board.

[0011] In an optional embodiment,

[0012] Applying a preset stress field to the flexible circuit board to be measured, constructing a gridded strain monitoring point array on the surface of the flexible circuit board, obtaining a strain distribution map of the flexible circuit board through a multispectral camera, and dividing the deformation sensitive area based on the strain distribution map includes:

[0013] Applying a preset stress field to a flexible circuit board, constructing a gridded strain monitoring point array on the surface of the flexible circuit board, and obtaining a stressed flexible circuit board with the gridded strain monitoring point array; collecting multispectral images of the stressed flexible circuit board before and after the preset stress field is applied, and performing registration processing on the multispectral images to obtain a registered deformation image;

[0014] Divide the registered deformation image into a plurality of sub-regions, each of which contains at least one monitoring point in the gridded strain monitoring point array; construct a displacement correlation function between adjacent sub-regions, wherein the displacement correlation function includes a gray value distribution and a displacement increment of the sub-region; solve the displacement correlation function by an iterative optimization algorithm to obtain a displacement vector of the sub-region, calculate a displacement gradient of each sub-region based on the displacement vector, and obtain a displacement field distribution of the flexible circuit board;

[0015] Using a preset radial basis function to interpolate the displacement field distribution, obtain a continuous displacement field of the flexible circuit board, calculate the strain tensor of the flexible circuit board based on the continuous displacement field, obtain a strain distribution map of the flexible circuit board, calculate the principal strain amplitude according to the strain distribution map, determine the area where the principal strain amplitude exceeds a preset strain threshold as a candidate sensitive area, calculate the strain gradient value in the candidate sensitive area, and obtain a strain gradient value distribution map;

[0016] Based on the strain gradient value distribution map, a strain gradient maximum point is determined as a seed point in the candidate sensitive area, and a strain monitoring network is constructed with the seed point as the center, wherein the strain monitoring network includes a plurality of concentric annular monitoring areas, and a radius difference between adjacent concentric annular monitoring areas is determined according to the strain gradient value distribution map, and a plurality of strain monitoring nodes are set in each of the concentric annular monitoring areas;

[0017] The correlation between the strain value of each strain monitoring node and the strain value of the strain monitoring node in the adjacent concentric annular monitoring area is calculated, and the strain weight value of each strain monitoring node is determined based on the correlation; the area corresponding to the strain monitoring node whose strain weight value is greater than a preset weight threshold is determined as the deformation sensitive area of ​​the flexible circuit board.

[0018] In an optional embodiment,

[0019] Based on the distribution of the deformation sensitive areas, calculating the optimal arrangement positions of the four sets of binocular laser scanning devices includes:

[0020] The deformation sensitive area is represented as a plurality of plane sub-areas, each of which includes a center coordinate and an area weight, and based on the center coordinate and the area weight, four sets of position parameter sets of the binocular laser scanning device are constructed;

[0021] Calculating the observation angle of each group of binocular laser scanning devices for each of the plane sub-areas, evaluating the laser stripe imaging quality coefficient at each of the observation angles, and calculating the occlusion coefficient of each of the plane sub-areas at each of the observation angles; determining the visibility score of the plane sub-area based on the product of the observation angle, the imaging quality coefficient and the occlusion coefficient, and constructing an optimization objective function based on the area weight and the visibility score;

[0022] Construct a first candidate solution group and a second candidate solution group based on the position parameter set, each of the candidate solutions carries a set of the position parameter set; the first search radius corresponding to the first candidate solution group is greater than the second search radius corresponding to the second candidate solution group; calculate the optimization objective function value corresponding to each candidate solution; select the candidate solution with the highest optimization objective function value in the first candidate solution group as the first optimal solution, and select the candidate solution with the highest optimization objective function value in the second candidate solution group as the second optimal solution;

[0023] Taking the first optimal solution as the center, within the scope of the first search radius, randomly generating a corresponding number of new first position parameter sets according to a preset first number, updating the candidate solutions carrying the new first position parameter sets to the first candidate solution group, and obtaining an updated first candidate solution group; taking the second optimal solution as the center, within the scope of the second search radius, randomly generating a corresponding number of new second position parameter sets according to a preset second number, updating the candidate solutions carrying the new second position parameter sets to the second candidate solution group, and obtaining an updated second candidate solution group;

[0024] Calculating the optimization objective function value of each candidate solution in the updated first candidate solution group and the updated second candidate solution group; if the optimization objective function value of the first optimal solution is greater than the optimization objective function value of the second optimal solution, replacing the candidate solution with the lowest optimization objective function value in the second candidate solution group with the position parameter set of the first optimal solution;

[0025] The iterative process is repeatedly performed. Based on a preset number of continuous iterations, when the maximum improvement of the optimization objective function value is less than a preset improvement threshold, the position parameter set corresponding to the first optimal solution is determined to be the optimal arrangement scheme of the four groups of binocular laser scanning devices.

[0026] In an optional embodiment,

[0027] Inputting the calibration data of the four groups of binocular laser scanning devices into the adaptive multi-step calibration algorithm to obtain the spatial mapping relationship matrix includes:

[0028] Acquire a calibration image and calibration data of a binocular laser scanning device, extract initial values ​​of intrinsic parameters of the binocular laser scanning device based on the calibration data, wherein the initial values ​​of intrinsic parameters include focal length parameters, principal point coordinate parameters, and radial distortion coefficient parameters; construct a camera intrinsic parameter matrix based on the focal length parameters and the principal point coordinate parameters, and construct a distortion coefficient matrix based on the radial distortion coefficient parameters; and combine the camera intrinsic parameter matrix with the distortion coefficient matrix to construct an initial projection matrix;

[0029] Extracting a projection point set from the calibration image, dividing the projection points in the projection point set into a central area projection point subset and an edge area projection point subset according to the distance from the image center point; applying the initial projection matrix to the central area projection point subset and the edge area projection point subset, respectively, and calculating the reprojection error mean and reprojection error standard deviation of each area;

[0030] Determine a regional weight item according to a reprojection error mean and a reprojection error standard deviation corresponding to the region where the projection point is located, determine a distance weight item according to a distance from the projection point to the center point of the image, and determine a dual adaptive weight factor based on the regional weight item and the distance weight item;

[0031] Multiplying the dual adaptive weight factor by the square of the reprojection error of the corresponding projection point to construct a weighted projection error, using the weighted projection error to construct an optimization objective function, optimizing the projection points of the central area projection point subset and the edge area projection point subset based on the optimization objective function, updating the initial value of the internal parameter, and obtaining an optimized internal parameter matrix;

[0032] Based on the central area projection point subset, the initial value of the rotation matrix is ​​calculated, and the initial value of the rotation matrix is ​​used as the initial value. In combination with the radial distortion characteristics of the edge area projection point subset, an external parameter matrix is ​​calculated, and the external parameter matrix includes final rotation matrix parameters and translation vector parameters; the optimized internal parameter matrix is ​​combined with the external parameter matrix to construct a spatial mapping relationship matrix;

[0033] Reprojection errors of the projection points in the central area projection point subset and the edge area projection point subset are calculated respectively. When the reprojection error of the central area projection point subset is greater than a first preset threshold or the reprojection error of the edge area projection point subset is greater than a second preset threshold, the dual adaptive weight factor is updated according to the current reprojection error distribution, and the iteration is continued until the reprojection error of the central area projection point subset is less than the first preset threshold and the reprojection error of the edge area projection point subset is less than the second preset threshold, and the current spatial mapping relationship matrix is ​​determined as the final spatial mapping relationship matrix.

[0034] In an optional embodiment,

[0035] The orthogonal grid projection image group is subjected to double Gaussian fitting to extract the center line of the laser stripe, the center line of the laser stripe is combined with the spatial mapping relationship matrix to construct a depth information mapping model, three-dimensional coordinates are calculated according to the depth information mapping model, and initial three-dimensional point cloud data is generated. The initial three-dimensional point cloud data is resampled and fused by a local feature preservation algorithm to obtain fused point cloud data including:

[0036] grayscale the orthogonal grid projection image group to obtain a grayscale image, decompose the grayscale image into multiple layers of sub-images by a multi-scale decomposition method, and determine a corresponding adaptive window based on a local grayscale gradient amplitude of each layer of the multiple layers of sub-images;

[0037] Sampling points of the grayscale distribution of the laser stripes are collected within the adaptive window, outlier detection is performed on the sampling points to obtain optimized sampling points, and a dual Gaussian distribution model is established based on the optimized sampling points, wherein the dual Gaussian distribution model includes a first Gaussian distribution function and a second Gaussian distribution function; a weighted least squares objective function with a regularization term is constructed, and parameters of the first Gaussian distribution function and the second Gaussian distribution function are solved by minimizing the weighted least squares objective function to obtain a multi-scale laser stripe grayscale distribution fitting model;

[0038] Extracting the laser stripe center line of each layer of sub-images in the multi-layer sub-images based on the multi-scale laser stripe grayscale distribution fitting model to obtain the multi-scale laser stripe center line, adaptively fusing the multi-scale laser stripe center line to obtain the laser stripe fusion center line, combining the laser stripe fusion center line with the spatial mapping relationship matrix to establish a nonlinear spatial mapping relationship, and calculating the spatial coordinate point corresponding to the laser stripe fusion center line according to the nonlinear spatial mapping relationship to obtain initial three-dimensional point cloud data;

[0039] Calculate the principal curvature parameters and principal direction parameters of each of the spatial coordinate points in the initial three-dimensional point cloud data, and calculate the point distribution entropy of the spatial coordinate point based on the principal curvature parameters and the principal direction parameters; construct a feature descriptor by combining the principal curvature parameters, the principal direction parameters and the point distribution entropy; calculate the adaptive weight coefficient of the spatial coordinate point based on the feature descriptor; perform hierarchical resampling on the initial three-dimensional point cloud data according to the adaptive weight coefficient to obtain resampled point cloud data;

[0040] A local consistency constraint including a point constraint, a normal vector constraint and a curvature constraint is constructed, the local consistency constraint is integrated into an iterative closest point algorithm, and the resampled point cloud data is registered and locally fused to obtain fused three-dimensional point cloud data.

[0041] A local consistency constraint including a point position constraint, a normal vector constraint and a curvature constraint is constructed, and the local consistency constraint is integrated into an iterative closest point algorithm to obtain a constraint optimization model; based on the constraint optimization model, the resampled point cloud data is iteratively registered to obtain registered point cloud data; and the registered point cloud data is locally fused according to the feature descriptor to obtain fused three-dimensional point cloud data.

[0042] In an optional embodiment,

[0043] The fused point cloud data is input into the local curvature adaptive algorithm for noise reduction and abnormal point removal, and the optimized point cloud data obtained includes:

[0044] The fused point cloud data is spatially segmented based on the Euclidean distance between points and the point cloud density to obtain multiple point cloud subsets, a KD tree index structure is constructed for each of the point cloud subsets, and a neighbor point set of each point is calculated based on the KD tree index structure;

[0045] For each of the neighboring point sets, a local structure tensor is calculated using a preset tensor voting algorithm, and based on the eigenvalues ​​of the local structure tensor, a normal vector and a tangent plane of each point are decomposed to obtain; an orthogonal basis function family is constructed on the tangent plane, and the orthogonal basis function family is used to fit a local surface, and a first principal curvature and a second principal curvature of each point are calculated according to the local surface;

[0046] Calculate the variance of the first principal curvature and the second principal curvature of each point in the neighboring point set, and when the variance is greater than the preset threshold, mark the corresponding point as a candidate abnormal point; calculate the spatial distribution density of the points in the neighboring point set, verify the candidate abnormal point based on the spatial distribution density, and obtain an abnormal point set;

[0047] For points that do not belong to the abnormal point set, a Gaussian curvature map is constructed based on the first principal curvature and the second principal curvature, a local entropy value of the Gaussian curvature map is calculated, the local entropy value is determined as the local feature strength, a shape parameter and a scale parameter of a smoothing kernel function are calculated according to the local feature strength, and an adaptive smoothing kernel function is constructed; the adaptive smoothing kernel function is applied to the neighboring point set, and a contribution weight of each neighboring point is calculated based on the Euclidean distance between the points and the corresponding angle of the normal vector, wherein the contribution weight decreases as the Euclidean distance between the points increases, and decreases as the corresponding angle of the normal vector increases;

[0048] A geometric feature preserving function is constructed, wherein the geometric feature preserving function includes a position constraint term and a normal vector consistency constraint term, wherein the position constraint term calculates a weighted average deviation of the point position based on the contribution weight, and the normal vector consistency constraint term calculates a weighted average deviation of the normal vector based on the contribution weight; the position and normal vector of the point are updated by iteratively optimizing the geometric feature preserving function, and points in the abnormal point set are eliminated to obtain optimized point cloud data.

[0049] In an optional embodiment,

[0050] Dividing the surface of the flexible circuit board into a plurality of characteristic sub-areas according to the deformation sensitive area, using an adaptive spline surface to perform block fitting on the characteristic sub-areas, and determining the fitting results; calculating the curvature distribution of the flexible circuit board based on the optimized point cloud data, optimizing and correcting the fitting results in combination with the strain distribution map, and generating the three-dimensional geometric morphology of the flexible circuit board includes:

[0051] Acquire point cloud data of the surface of the flexible circuit board and position data of the deformation sensitive area on the surface of the flexible circuit board; extract a feature enhancement point set based on the position data of the deformation sensitive area, calculate the feature importance value of each point in the feature enhancement point set, and determine a grid division parameter according to the feature importance value, wherein the grid division parameter decreases as the feature importance value increases;

[0052] The point cloud data is adaptively quad-tree partitioned using the grid partitioning parameters to obtain a plurality of feature sub-regions; a local coordinate system is established in each of the feature sub-regions, the point cloud data in the feature sub-region is projected onto the local coordinate system, and the local curvature distribution of the projected point cloud data in the feature sub-region is calculated;

[0053] A weight coefficient is constructed according to the local curvature distribution, and the weight coefficient increases as the local curvature distribution increases; an adaptive spline surface model is constructed based on the weight coefficient, and a surface fitting is performed on the characteristic sub-region through a B-spline basis function to obtain an initial fitting surface of the characteristic sub-region;

[0054] Acquiring strain distribution data of the flexible circuit board, calculating the Gaussian curvature value and the average curvature value of the characteristic sub-region based on the initial fitting surface, and constructing a surface correction function according to the strain distribution data, the Gaussian curvature value and the average curvature value, wherein the correction amount of the surface correction function is positively correlated with the strain distribution data;

[0055] The normal vector of the characteristic sub-region is calculated, and the correction amount of the surface correction function is applied to the initial fitting surface along the direction of the normal vector to obtain an optimized fitting surface; the parameters of the surface correction function are iteratively optimized to match the optimized fitting surface with the strain distribution data, thereby generating the three-dimensional geometric morphology of the flexible circuit board.

[0056] According to a second aspect of the embodiments of the present invention,

[0057] Provided is a flexible circuit board three-dimensional shape laser scanning measurement and reconstruction system, comprising:

[0058] The first unit is used to apply a preset stress field to the flexible circuit board to be measured, construct a gridded strain monitoring point array on the surface of the flexible circuit board, obtain a strain distribution map of the flexible circuit board through a multispectral camera, and divide the deformation sensitive area based on the strain distribution map; based on the distribution of the deformation sensitive area, calculate the optimal arrangement position of four groups of binocular laser scanning devices, and use an adaptive exposure control strategy to collect an orthogonal grid projection image group of the flexible circuit board;

[0059] The second unit is used to input the calibration data of the four groups of binocular laser scanning devices into an adaptive multi-step calibration algorithm to obtain a spatial mapping relationship matrix; perform double Gaussian fitting on the orthogonal grid projection image group to extract the center line of the laser stripe, combine the center line of the laser stripe with the spatial mapping relationship matrix to construct a depth information mapping model, calculate the three-dimensional coordinates according to the depth information mapping model, generate initial three-dimensional point cloud data, and resample and align and fuse the initial three-dimensional point cloud data through a local feature preservation algorithm to obtain fused point cloud data;

[0060] The third unit is used to input the fused point cloud data into a local curvature adaptive algorithm for noise reduction and abnormal point removal to obtain optimized point cloud data; divide the surface of the flexible circuit board into multiple characteristic sub-areas according to the deformation sensitive area, and use an adaptive spline surface to perform block fitting on the characteristic sub-areas to determine the fitting results; calculate the curvature distribution of the flexible circuit board based on the optimized point cloud data, optimize and correct the fitting results in combination with the strain distribution map, and generate the three-dimensional geometric morphology of the flexible circuit board.

[0061] According to a third aspect of the embodiments of the present invention,

[0062] An electronic device is provided, comprising:

[0063] processor;

[0064] a memory for storing processor-executable instructions;

[0065] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0066] A fourth aspect of the embodiments of the present invention is:

[0067] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0068] In the embodiment of the present invention, by constructing a gridded strain monitoring point array and a multispectral camera to obtain a strain distribution map, dividing the deformation sensitive area, and optimizing the layout position of the binocular laser scanning device based on this, the subtle deformation of the flexible circuit board can be effectively captured, thereby improving the accuracy of three-dimensional shape measurement. The adaptive multi-step calibration algorithm and dual Gaussian fitting further improve the accuracy of the data; the use of four groups of binocular laser scanning devices and adaptive exposure control strategies can quickly collect orthogonal grid projection image groups and accelerate the speed of three-dimensional shape reconstruction; the application of local feature preservation algorithm and local curvature adaptive algorithm optimizes the point cloud data processing process and further improves the measurement efficiency; based on the deformation sensitive area, block fitting is performed, and combined with the curvature distribution and strain distribution map for optimization and correction, the complex three-dimensional shape of the flexible circuit board can be more accurately restored, especially in the area with large deformation; it effectively avoids the problems of error accumulation and distortion that occur in traditional methods when dealing with complex deformations, and enhances the integrity and robustness of shape reconstruction. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 It is a schematic diagram of the process of the three-dimensional shape laser scanning measurement and reconstruction method of a flexible circuit board according to an embodiment of the present invention;

[0070] Figure 2 It is the observation angle quality assessment and optimal layout simulation diagram;

[0071] Figure 3 Comparison chart of simulation effects for 3D point cloud reconstruction;

[0072] Figure 4 It is a line chart comparing the convergence performance of surface fitting algorithms;

[0073] Figure 5 It is a schematic structural diagram of a flexible circuit board three-dimensional topography laser scanning measurement and reconstruction system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0074] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0075] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0076] Figure 1 FIG. 1 is a flow chart of a method for measuring and reconstructing the three-dimensional shape of a flexible circuit board using laser scanning according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0077] S101. Apply a preset stress field to the flexible circuit board to be measured, construct a gridded strain monitoring point array on the surface of the flexible circuit board, obtain a strain distribution map of the flexible circuit board through a multispectral camera, and divide the deformation sensitive areas based on the strain distribution map; based on the distribution of the deformation sensitive areas, calculate the optimal arrangement positions of four groups of binocular laser scanning devices, and use an adaptive exposure control strategy to collect an orthogonal grid projection image group of the flexible circuit board;

[0078] In this embodiment, a preset stress field is applied to the flexible circuit board and a gridded strain monitoring point array is constructed to achieve accurate measurement of its strain distribution; a multispectral camera is used to obtain the strain distribution map and divide the deformation sensitive areas to improve the recognition ability of key deformation areas; the layout position of the binocular laser scanning device is optimized based on the deformation sensitive areas to ensure the efficiency and accuracy of data acquisition; an adaptive exposure control strategy is used to obtain an orthogonal grid projection image group to improve imaging quality and data stability, providing high-precision data support for subsequent three-dimensional reconstruction and morphology analysis.

[0079] S102. Input the calibration data of the four groups of binocular laser scanning devices into an adaptive multi-step calibration algorithm to obtain a spatial mapping relationship matrix; perform double Gaussian fitting on the orthogonal grid projection image group to extract the center line of the laser stripe, combine the center line of the laser stripe with the spatial mapping relationship matrix to construct a depth information mapping model, calculate the three-dimensional coordinates according to the depth information mapping model, generate initial three-dimensional point cloud data, and resample and align the initial three-dimensional point cloud data through a local feature preservation algorithm to obtain fused point cloud data;

[0080] In this embodiment, an adaptive multi-step calibration algorithm is used to accurately obtain the spatial mapping relationship of the binocular laser scanning device, thereby improving the accuracy of three-dimensional measurement. A dual Gaussian fitting is used to extract the center line of the laser stripe, thereby enhancing the accuracy and robustness of feature point extraction. A depth information mapping model is used to calculate the three-dimensional coordinates and generate initial point cloud data, thereby achieving efficient three-dimensional reconstruction. A local feature preserving algorithm is used for resampling and registration fusion, thereby effectively improving the integrity and consistency of the point cloud data, thereby providing a high-quality data foundation for subsequent geometric analysis and morphology optimization.

[0081] S103. Input the fused point cloud data into the local curvature adaptive algorithm for noise reduction and abnormal point removal to obtain optimized point cloud data; divide the surface of the flexible circuit board into multiple characteristic sub-areas according to the deformation sensitive area, and use adaptive spline surface to perform block fitting on the characteristic sub-areas to determine the fitting results; calculate the curvature distribution of the flexible circuit board based on the optimized point cloud data, optimize and correct the fitting results in combination with the strain distribution map, and generate the three-dimensional geometric morphology of the flexible circuit board.

[0082] In this embodiment, the fused point cloud data is denoised and abnormal points are removed through a local curvature adaptive algorithm to improve the purity and reliability of the point cloud data; the surface of the flexible circuit board is characterized based on the deformation sensitive area, and adaptive spline surfaces are used for block fitting to ensure accurate description of complex morphologies; the curvature distribution is calculated using the optimized point cloud data, and the fitting results are optimized and corrected in combination with the strain distribution diagram, thereby improving the accuracy of three-dimensional geometric morphology reconstruction and providing high-precision data support for quality inspection and performance evaluation of flexible circuit boards.

[0083] In an optional implementation, a preset stress field is applied to the flexible circuit board to be measured, a gridded strain monitoring point array is constructed on the surface of the flexible circuit board, a strain distribution map of the flexible circuit board is obtained by a multi-spectral camera, and the deformation sensitive area is divided based on the strain distribution map, including:

[0084] Applying a preset stress field to a flexible circuit board, constructing a gridded strain monitoring point array on the surface of the flexible circuit board, and obtaining a stressed flexible circuit board with the gridded strain monitoring point array; collecting multispectral images of the stressed flexible circuit board before and after the preset stress field is applied, and performing registration processing on the multispectral images to obtain a registered deformation image;

[0085] Divide the registered deformation image into a plurality of sub-regions, each of which contains at least one monitoring point in the gridded strain monitoring point array; construct a displacement correlation function between adjacent sub-regions, wherein the displacement correlation function includes a gray value distribution and a displacement increment of the sub-region; solve the displacement correlation function by an iterative optimization algorithm to obtain a displacement vector of the sub-region, calculate a displacement gradient of each sub-region based on the displacement vector, and obtain a displacement field distribution of the flexible circuit board;

[0086] Using a preset radial basis function to interpolate the displacement field distribution, obtain a continuous displacement field of the flexible circuit board, calculate the strain tensor of the flexible circuit board based on the continuous displacement field, obtain a strain distribution map of the flexible circuit board, calculate the principal strain amplitude according to the strain distribution map, determine the area where the principal strain amplitude exceeds a preset strain threshold as a candidate sensitive area, calculate the strain gradient value in the candidate sensitive area, and obtain a strain gradient value distribution map;

[0087] Based on the strain gradient value distribution map, a strain gradient maximum point is determined as a seed point in the candidate sensitive area, and a strain monitoring network is constructed with the seed point as the center, wherein the strain monitoring network includes a plurality of concentric annular monitoring areas, and a radius difference between adjacent concentric annular monitoring areas is determined according to the strain gradient value distribution map, and a plurality of strain monitoring nodes are set in each of the concentric annular monitoring areas;

[0088] The correlation between the strain value of each strain monitoring node and the strain value of the strain monitoring node in the adjacent concentric annular monitoring area is calculated, and the strain weight value of each strain monitoring node is determined based on the correlation; the area corresponding to the strain monitoring node whose strain weight value is greater than a preset weight threshold is determined as the deformation sensitive area of ​​the flexible circuit board.

[0089] In a specific implementation, it is first necessary to apply a preset stress field to the flexible circuit board and construct a strain monitoring point array, obtain deformation information through multi-spectral image acquisition, and then accurately identify deformation sensitive areas through displacement field analysis and strain monitoring network construction.

[0090] When constructing an array of strain monitoring points on the surface of a flexible circuit board, a photolithography process is used to coat the surface with photosensitive materials, and regular grid-like monitoring points are made through exposure and development. The grid spacing is 2 mm, covering the entire surface of the flexible circuit board. A customized multi-degree-of-freedom load application device is used to apply a preset stress field to the flexible circuit board, applying a tensile force of 5 Newtons at one end and a torque of 0.5 Newton-meters in the middle, causing the flexible circuit board to produce a complex deformation.

[0091] A multispectral camera equipped with three bands of 540 nm, 680 nm and 820 nm was used to collect image groups before and after deformation. The collected images were preprocessed, including lens distortion correction and dark field correction, to eliminate system errors and random noise. The images before and after deformation were registered by feature point matching method to obtain a registered image group that accurately reflects the deformation information.

[0092] The registered image is divided into overlapping sub-region grids with a sub-region size of 5×5 mm and an overlap rate of 50% for adjacent sub-regions. A correlation analysis model considering the gray value distribution and displacement increment is established between adjacent sub-regions. The correlation function is solved by an iterative optimization method to obtain the displacement vector of the sub-region. The displacement gradient field is calculated based on the displacement vector to obtain the displacement field distribution reflecting the overall deformation characteristics.

[0093] The radial basis function interpolation is performed on the displacement field distribution to obtain a continuous displacement field function. The components of the strain tensor are calculated based on the continuous displacement field to obtain a strain distribution map. The principal strain direction and amplitude are extracted, and the area where the principal strain amplitude exceeds 3% is marked as a candidate sensitive area. The strain gradient value is calculated in the candidate sensitive area to obtain a strain gradient distribution map.

[0094] In the strain gradient distribution map, local maximum points are found as seed points, and three concentric annular monitoring areas are constructed with the seed points as the center, with radii of 3 mm, 6 mm, and 9 mm respectively. The spacing between adjacent annular areas is determined according to the variation law of the strain gradient, and 12 monitoring nodes are evenly arranged in each annular area.

[0095] The strain correlation between monitoring nodes is calculated, and the strain weight value of each monitoring node is determined by using a weight evaluation method based on strain gradient. The area where the monitoring node with a strain weight value greater than 0.8 is located is determined as a deformation sensitive area.

[0096] In practical applications, taking a flexible circuit board for a smart bracelet as an example, the above method successfully identified three main deformation-sensitive areas located near the bending area and solder joints of the device. The maximum strain gradient value of these areas reached 0.15 / mm, and the strain weight value exceeded 0.85, indicating that these areas are the most significant locations for deformation. The identification results of deformation-sensitive areas provide an important basis for subsequent 3D morphology reconstruction, effectively improving the reconstruction accuracy.

[0097] This method based on multispectral image analysis and strain monitoring network achieves accurate identification of deformation-sensitive areas through multi-level strain information extraction and correlation analysis, providing effective technical support for the morphology reconstruction and reliability assessment of flexible circuit boards.

[0098] In this embodiment, by constructing a strain monitoring network and calculating the strain weight value, the deformation sensitive area of ​​the flexible circuit board can be identified more accurately, avoiding the error caused by relying solely on the strain amplitude for judgment; through multi-spectral imaging and displacement field calculation, a high-resolution strain distribution map can be obtained, thereby realizing the refined division of the deformation sensitive area and better guiding the design and manufacture of the flexible circuit board; by identifying and monitoring the deformation sensitive area, the failure position of the flexible circuit board can be predicted in advance, and corresponding measures can be taken to improve the reliability and service life of the flexible circuit board.

[0099] In an optional implementation, based on the distribution of the deformation sensitive area, calculating the optimal arrangement positions of four groups of binocular laser scanning devices includes:

[0100] The deformation sensitive area is represented as a plurality of plane sub-areas, each of which includes a center coordinate and an area weight, and based on the center coordinate and the area weight, four sets of position parameter sets of the binocular laser scanning device are constructed;

[0101] Calculating the observation angle of each group of binocular laser scanning devices for each of the plane sub-areas, evaluating the laser stripe imaging quality coefficient at each of the observation angles, and calculating the occlusion coefficient of each of the plane sub-areas at each of the observation angles; determining the visibility score of the plane sub-area based on the product of the observation angle, the imaging quality coefficient and the occlusion coefficient, and constructing an optimization objective function based on the area weight and the visibility score;

[0102] Construct a first candidate solution group and a second candidate solution group based on the position parameter set, each of the candidate solutions carries a set of the position parameter set; the first search radius corresponding to the first candidate solution group is greater than the second search radius corresponding to the second candidate solution group; calculate the optimization objective function value corresponding to each candidate solution; select the candidate solution with the highest optimization objective function value in the first candidate solution group as the first optimal solution, and select the candidate solution with the highest optimization objective function value in the second candidate solution group as the second optimal solution;

[0103] Taking the first optimal solution as the center, within the scope of the first search radius, randomly generating a corresponding number of new first position parameter sets according to a preset first number, updating the candidate solutions carrying the new first position parameter sets to the first candidate solution group, and obtaining an updated first candidate solution group; taking the second optimal solution as the center, within the scope of the second search radius, randomly generating a corresponding number of new second position parameter sets according to a preset second number, updating the candidate solutions carrying the new second position parameter sets to the second candidate solution group, and obtaining an updated second candidate solution group;

[0104] Calculating the optimization objective function value of each candidate solution in the updated first candidate solution group and the updated second candidate solution group; if the optimization objective function value of the first optimal solution is greater than the optimization objective function value of the second optimal solution, replacing the candidate solution with the lowest optimization objective function value in the second candidate solution group with the position parameter set of the first optimal solution;

[0105] The iterative process is repeatedly performed. Based on a preset number of continuous iterations, when the maximum improvement of the optimization objective function value is less than a preset improvement threshold, the position parameter set corresponding to the first optimal solution is determined to be the optimal arrangement scheme of the four groups of binocular laser scanning devices.

[0106] In a specific implementation, first, the deformation sensitive area to be measured is divided into multiple planar sub-areas. Each sub-area is represented by its central coordinates and area weight. The area weight reflects the importance of the sub-area in the overall deformation monitoring. For example, a sub-area with a larger area or a sub-area in a key position will be given a larger weight. For example, a 10m×10m deformation sensitive area can be divided into 100 1m×1m sub-areas, and the central coordinates of each sub-area are its geometric center. The area weight can be set to the same value or different values ​​according to the actual situation. For example, a sub-area in the central area can be set with a larger weight.

[0107] Next, construct a set of position parameters for four sets of binocular laser scanning devices. Each set of binocular laser scanning devices contains two laser transmitters and two receivers, and its position parameters include three-dimensional coordinates and rotation angles. These parameters will be used as variables in the subsequent optimization process. The initial set of position parameters can be determined by experience or rough estimation in advance. For example, the four sets of devices can be evenly distributed around the deformation-sensitive area and roughly oriented toward the center of the area.

[0108] Then, for each sub-area, the observation angle of each binocular laser scanning device is calculated. The observation angle refers to the angle between the laser beam and the normal of the sub-area. At the same time, the laser stripe imaging quality coefficient at each observation angle is evaluated. This coefficient reflects the clarity and stability of the laser stripe at this angle, and its value range is 0 to 1. The larger the value, the better the imaging quality. The imaging quality coefficient can be obtained by simulation or experiment. For example, a vertically incident laser usually has a good imaging quality, and its imaging quality coefficient can be set to 1; while a laser with a large tilt angle may be affected by scattering and reflection, and its imaging quality coefficient will be reduced accordingly, for example, set to 0.5 or lower. In addition, it is necessary to calculate the occlusion coefficient of each sub-area at each observation angle. The occlusion coefficient indicates the degree to which the sub-area is blocked by other objects at this angle, and its value range is also 0 to 1. The larger the value, the more serious the occlusion. The occlusion coefficient can be calculated by establishing a three-dimensional model and performing ray tracing.

[0109] The visibility score of each sub-area is determined based on the product of the observation angle, the imaging quality coefficient, and the occlusion coefficient. The higher the visibility score, the better the deformation monitoring effect of the sub-area at the observation angle. Then, based on the area weight and visibility score of each sub-area, an optimization objective function is constructed. The goal of the optimization objective function is to maximize the sum of the weighted visibility scores of all sub-areas.

[0110] In order to find the optimal device layout solution, a two-layer iterative optimization strategy is adopted. First, a first candidate solution group and a second candidate solution group are constructed. Each candidate solution carries a set of position parameter sets. The first search radius corresponding to the first candidate solution group is larger than the second search radius corresponding to the second candidate solution group. The optimization objective function value corresponding to each candidate solution is calculated. The candidate solution with the highest optimization objective function value in the first candidate solution group is selected as the first optimal solution, and the candidate solution with the highest optimization objective function value in the second candidate solution group is selected as the second optimal solution.

[0111] Then, enter the iterative process. With the first optimal solution as the center, within the scope of the first search radius, randomly generate a preset number of new position parameter sets, and update these new candidate solutions to the first candidate solution group. Similarly, with the second optimal solution as the center, within the scope of the second search radius, randomly generate a preset number of new position parameter sets, and update these new candidate solutions to the second candidate solution group. Calculate the optimization objective function value of each candidate solution in the two updated candidate solution groups. If the optimization objective function value of the first optimal solution is greater than the optimization objective function value of the second optimal solution, the position parameter set of the first optimal solution is used to replace the candidate solution with the lowest optimization objective function value in the second candidate solution group. Repeat the above iterative process. When the maximum improvement of the optimization objective function value is less than the preset improvement threshold in the preset number of consecutive iterations, it is determined that the position parameter set corresponding to the first optimal solution is the optimal layout scheme of the four groups of binocular laser scanning devices. For example, the number of iterations is set to 100 times and the improvement threshold is 0.01. If the improvement of the optimization objective function value does not exceed 0.01 in 100 consecutive iterations, the algorithm is considered to have converged, and the first optimal solution at this time is the final optimization result.

[0112] Said Figure 2The optimal layout results of the binocular laser scanning device based on the proposed technical solution are shown. The figure clearly shows that the deformation sensitive area is divided into 45 planar sub-areas (9×5 grids), each of which contains the center coordinates and area weight information (such as (5,4) and W=0.72, etc.). Four groups of binocular laser scanning devices (devices A, B, C, and D) are arranged at the four corners of the observation space, with coordinates of (100,100), (650,100), (650,450), and (100,450), respectively. Each device is marked with 4 main observation angles, represented by red rays. The observation angles of device A are 28°, 45°, 65°, and 83°, those of device B are 115°, 135°, 152°, and 175°, those of device C are 195°, 215°, 232°, and 255°, and those of device D are 285°, 302°, 322°, and 345°. These angles are calculated by an optimization algorithm to ensure optimal coverage of deformation-sensitive areas. The simulation results show that the layout scheme achieves a comprehensive visibility score of 94.7%, with an average occlusion coefficient of only 0.15, and the observation quality coefficients of each device are 0.92, 0.89, 0.95 and 0.88, respectively, indicating that the scheme can effectively overcome the occlusion problem and provide high-quality laser stripe imaging.

[0113] In this embodiment, by optimizing the layout position of the binocular laser scanning device, the visibility score of the deformation sensitive area can be maximized, thereby improving the accuracy and reliability of deformation measurement; the optimal device layout plan can be automatically determined, avoiding the tedious process of manual trial and error, thereby improving work efficiency; the device layout plan can be flexibly adjusted according to the shape and distribution characteristics of different deformation sensitive areas, thereby enhancing the adaptability of the method.

[0114] In an optional implementation, inputting the calibration data of the four groups of binocular laser scanning devices into an adaptive multi-step calibration algorithm to obtain a spatial mapping relationship matrix includes:

[0115] Acquire a calibration image and calibration data of a binocular laser scanning device, extract initial values ​​of intrinsic parameters of the binocular laser scanning device based on the calibration data, wherein the initial values ​​of intrinsic parameters include focal length parameters, principal point coordinate parameters, and radial distortion coefficient parameters; construct a camera intrinsic parameter matrix based on the focal length parameters and the principal point coordinate parameters, and construct a distortion coefficient matrix based on the radial distortion coefficient parameters; and combine the camera intrinsic parameter matrix with the distortion coefficient matrix to construct an initial projection matrix;

[0116] Extracting a projection point set from the calibration image, dividing the projection points in the projection point set into a central area projection point subset and an edge area projection point subset according to the distance from the image center point; applying the initial projection matrix to the central area projection point subset and the edge area projection point subset, respectively, and calculating the reprojection error mean and reprojection error standard deviation of each area;

[0117] Determine a regional weight item according to a reprojection error mean and a reprojection error standard deviation corresponding to the region where the projection point is located, determine a distance weight item according to a distance from the projection point to the center point of the image, and determine a dual adaptive weight factor based on the regional weight item and the distance weight item;

[0118] Multiplying the dual adaptive weight factor by the square of the reprojection error of the corresponding projection point to construct a weighted projection error, using the weighted projection error to construct an optimization objective function, optimizing the projection points of the central area projection point subset and the edge area projection point subset based on the optimization objective function, updating the initial value of the internal parameter, and obtaining an optimized internal parameter matrix;

[0119] Based on the central area projection point subset, the initial value of the rotation matrix is ​​calculated, and the initial value of the rotation matrix is ​​used as the initial value. In combination with the radial distortion characteristics of the edge area projection point subset, an external parameter matrix is ​​calculated, and the external parameter matrix includes final rotation matrix parameters and translation vector parameters; the optimized internal parameter matrix is ​​combined with the external parameter matrix to construct a spatial mapping relationship matrix;

[0120] Reprojection errors of the projection points in the central area projection point subset and the edge area projection point subset are calculated respectively. When the reprojection error of the central area projection point subset is greater than a first preset threshold or the reprojection error of the edge area projection point subset is greater than a second preset threshold, the dual adaptive weight factor is updated according to the current reprojection error distribution, and the iteration is continued until the reprojection error of the central area projection point subset is less than the first preset threshold and the reprojection error of the edge area projection point subset is less than the second preset threshold, and the current spatial mapping relationship matrix is ​​determined as the final spatial mapping relationship matrix.

[0121] The calibration image specifically refers to a specific image used for camera calibration, which is usually a calibration plate of known structure or a laser scanning projection pattern, which is used to estimate the internal and external parameters of the camera and achieve accurate geometric measurement and three-dimensional reconstruction. The calibration image usually contains a regular pattern, such as a checkerboard, a dot array, or a specific mark, so that feature points can be extracted for calculation. The main function of the calibration image is to obtain pixel coordinate information and cover different areas to evaluate the distortion and projection error within the entire field of view.

[0122] The calibration data specifically refers to the geometric and optical information extracted from the calibration image, which is used to calculate the parameters of the camera or laser scanning device to ensure the accuracy of the measurement. The calibration data mainly includes intrinsic parameters, extrinsic parameters, projection point data and error analysis data. The intrinsic parameters describe the internal optical characteristics of the camera, such as focal length, principal point coordinates and distortion coefficients. The extrinsic parameters are used to represent the position and orientation of the camera in space, including the rotation matrix and the translation vector. The projection point data are the coordinates of the feature points extracted from the calibration image, which are used to calculate the projection error, while the error analysis data are used to optimize the camera parameters and improve the accuracy of the measurement.

[0123] In a specific implementation, a calibration image and calibration data of a binocular laser scanning device are obtained. Taking a checkerboard calibration plate as an example, multiple sets of calibration plate images with different postures and positions are taken, and the pixel coordinates of feature points in the image are extracted using the known size information of the calibration plate as calibration data. For example, 15 sets of calibration plate images are taken, each set of images has a resolution of 1920×1080 pixels, and the calibration plate grid size is 25 mm.

[0124] Extract the initial values ​​of the intrinsic parameters of the binocular laser scanning device based on the calibration data. Using the Zhang Zhengyou calibration method, based on the extracted calibration data, calculate the initial values ​​of the intrinsic parameters of the binocular laser scanning device, including focal length parameters, principal point coordinate parameters, and radial distortion coefficient parameters. For example, the initial value of the left camera focal length is (fx=1200,fy=1200), the initial value of the principal point coordinate is (cx=960,cy=540), and the initial value of the radial distortion coefficient is (k1=0.001,k2=-0.0005).

[0125] Construct the camera intrinsic parameter matrix and the distortion coefficient matrix, and combine them to construct the initial projection matrix. Construct the camera intrinsic parameter matrix based on the acquired focal length parameters and principal point coordinate parameters, and construct the distortion coefficient matrix based on the radial distortion coefficient parameters. Multiply the camera intrinsic parameter matrix and the distortion coefficient matrix to obtain the initial projection matrix.

[0126] Extract the projection point set from the calibration image and divide it into the center area projection point subset and the edge area projection point subset. Extract the pixel coordinates of all feature points in the calibration image to form a projection point set. Calculate the distance from each projection point to the center point of the image, divide the projection points whose distance is less than half of the distance from the center point of the image to the edge of the image into the center area projection point subset, and divide the remaining projection points into the edge area projection point subset. Assuming that the coordinates of the center point of the image are (960,540) and the image size is 1920×1080, the radius of the center area is sqrt((960 / 2) 2 +(540 / 2) 2 )≈540.

[0127] Calculate the mean and standard deviation of the reprojection error for each region. Apply the initial projection matrix to the subset of projection points in the center region and the subset of projection points in the edge region, respectively, and calculate the reprojection error of each projection point, that is, the difference between the actual pixel coordinates of the projection point and the pixel coordinates of the projection point calculated according to the projection matrix. Calculate the mean and standard deviation of the reprojection error for the center region and the edge region respectively. For example, the mean reprojection error for the center region is 0.5 pixels and the standard deviation is 0.2 pixels; the mean reprojection error for the edge region is 1.2 pixels and the standard deviation is 0.5 pixels.

[0128] The dual adaptive weight factor is determined based on the reprojection error and the distance from the projection point to the center of the image. The regional weight term is calculated based on the mean and standard deviation of the regional reprojection error, for example, the regional weight term = 1 / (mean + standard deviation). The distance weight term is calculated based on the distance from the projection point to the center of the image, for example, the distance weight term = 1 / (1+distance / image diagonal length). The dual adaptive weight factor is the product of the regional weight term and the distance weight term.

[0129] Construct the weighted projection error and optimization objective function, perform optimization calculations, and update the internal parameter matrix. Multiply the double adaptive weight factor by the square of the reprojection error of the corresponding projection point to obtain the weighted projection error. The sum of the weighted projection errors of all projection points is used as the optimization objective function. Use a nonlinear optimization algorithm (such as the Levenberg-Marquardt algorithm) to minimize the optimization objective function, update the initial value of the internal parameter, and obtain the optimized internal parameter matrix.

[0130] The initial value of the rotation matrix is ​​calculated based on the subset of projection points in the center area, and the extrinsic parameter matrix is ​​calculated in combination with the subset of projection points in the edge area. The initial value of the rotation matrix is ​​calculated by solving the pose estimation problem, such as using the PnP algorithm, using the optimized intrinsic parameter matrix and the subset of projection points in the center area. The initial value of the rotation matrix is ​​used as the initial value, combined with the radial distortion characteristics of the subset of projection points in the edge area, and a nonlinear optimization algorithm, such as the Bundle Adjustment algorithm, is used to calculate the extrinsic parameter matrix, including the final rotation matrix parameters and translation vector parameters.

[0131] Iteratively optimize the spatial mapping relationship matrix until the preset conditions are met. Calculate the reprojection errors of the projection points of the central area projection point subset and the edge area projection point subset respectively. Set a first preset threshold and a second preset threshold, for example, 0.1 pixels and 0.2 pixels respectively. When the reprojection error of the central area projection point subset is greater than the first preset threshold or the reprojection error of the edge area projection point subset is greater than the second preset threshold, update the dual adaptive weight factor according to the current reprojection error distribution, and return to recalculate the weighted projection error, optimize the internal parameters and external parameters, until the reprojection errors of the two areas are less than the corresponding preset thresholds. At this time, the current spatial mapping relationship matrix is ​​determined as the final spatial mapping relationship matrix.

[0132] In this embodiment, through the dual adaptive weight factors, the influence of edge area projection points and projection points with large distortion on the calibration results can be effectively reduced, thereby improving the calibration accuracy; the quality of the calibration image and the posture of the calibration plate are not required, and it has strong robustness; through regional optimization and adaptive weight factors, the convergence speed of the calibration algorithm can be accelerated and the calibration time can be shortened.

[0133] In an optional implementation, a double Gaussian fitting is performed on the orthogonal grid projection image group to extract the center line of the laser stripe, the center line of the laser stripe is combined with the spatial mapping relationship matrix to construct a depth information mapping model, three-dimensional coordinates are calculated according to the depth information mapping model, and initial three-dimensional point cloud data is generated. The initial three-dimensional point cloud data is resampled and fused by a local feature preservation algorithm to obtain the fused point cloud data including:

[0134] grayscale the orthogonal grid projection image group to obtain a grayscale image, decompose the grayscale image into multiple layers of sub-images by a multi-scale decomposition method, and determine a corresponding adaptive window based on a local grayscale gradient amplitude of each layer of the multiple layers of sub-images;

[0135] Sampling points of the grayscale distribution of the laser stripes are collected within the adaptive window, outlier detection is performed on the sampling points to obtain optimized sampling points, and a dual Gaussian distribution model is established based on the optimized sampling points, wherein the dual Gaussian distribution model includes a first Gaussian distribution function and a second Gaussian distribution function; a weighted least squares objective function with a regularization term is constructed, and parameters of the first Gaussian distribution function and the second Gaussian distribution function are solved by minimizing the weighted least squares objective function to obtain a multi-scale laser stripe grayscale distribution fitting model;

[0136] Extracting the laser stripe center line of each layer of sub-images in the multi-layer sub-images based on the multi-scale laser stripe grayscale distribution fitting model to obtain the multi-scale laser stripe center line, adaptively fusing the multi-scale laser stripe center line to obtain the laser stripe fusion center line, combining the laser stripe fusion center line with the spatial mapping relationship matrix to establish a nonlinear spatial mapping relationship, and calculating the spatial coordinate point corresponding to the laser stripe fusion center line according to the nonlinear spatial mapping relationship to obtain initial three-dimensional point cloud data;

[0137] Calculate the principal curvature parameters and principal direction parameters of each of the spatial coordinate points in the initial three-dimensional point cloud data, and calculate the point distribution entropy of the spatial coordinate point based on the principal curvature parameters and the principal direction parameters; construct a feature descriptor by combining the principal curvature parameters, the principal direction parameters and the point distribution entropy; calculate the adaptive weight coefficient of the spatial coordinate point based on the feature descriptor; perform hierarchical resampling on the initial three-dimensional point cloud data according to the adaptive weight coefficient to obtain resampled point cloud data;

[0138] A local consistency constraint including a point position constraint, a normal vector constraint and a curvature constraint is constructed, and the local consistency constraint is integrated into an iterative closest point algorithm to obtain a constraint optimization model; based on the constraint optimization model, the resampled point cloud data is iteratively registered to obtain registered point cloud data; and the registered point cloud data is locally fused according to the feature descriptor to obtain fused three-dimensional point cloud data.

[0139] In a specific implementation, firstly, an orthogonal grid projection image group is obtained. The image group includes an orthogonal grid pattern projected onto the surface of the object under test at different phases. Take an orthogonal grid projection image group including 10 images as an example.

[0140] Then the acquired image group is grayed out. The color image is converted into a gray image for subsequent processing. For example, the RGB value is converted into a gray value using a weighted average method.

[0141] Next, we use a multi-scale decomposition method to process the grayscale image. For example, we use wavelet transform to decompose the grayscale image into multiple layers of sub-images, assuming that it is decomposed into 3 layers. This can process image features of different scales separately and improve the robustness of the algorithm.

[0142] For each layer of sub-image, the window size is adaptively determined according to the local grayscale gradient amplitude. The window size is smaller in areas with larger gradient amplitudes, and vice versa. For example, in areas with gradient amplitudes between 0 and 10, the window size is 5×5; in areas with gradient amplitudes between 10 and 50, the window size is 3×3; in areas with gradient amplitudes greater than 50, the window size is 1×1.

[0143] The sampling points of the grayscale distribution of the laser stripe are collected in the adaptive window. For example, in a 5×5 window, the grayscale values ​​of 25 sampling points are collected.

[0144] Perform outlier detection on the collected sampling points and remove noise points. For example, using a statistical method, points that deviate from the mean by more than three standard deviations are considered outliers and removed.

[0145] The remaining optimized sampling points are used to establish a dual Gaussian distribution model, which consists of two Gaussian functions, fitting the bright and dark fringes of the laser stripe respectively.

[0146] By minimizing the weighted least squares objective function and solving the parameters of the two Gaussian functions, a multi-scale laser stripe grayscale distribution fitting model is obtained. The objective function contains a regularization term to prevent overfitting.

[0147] The center line of the laser stripe of each layer of sub-image is extracted based on the fitting model to obtain the center line of the multi-scale laser stripe. The position of the center line corresponds to the center point between the two Gaussian function peaks.

[0148] The center lines of the multi-scale laser stripes are adaptively fused to obtain the fused center line of the laser stripes. The fusion process takes into account the weights of the center lines of different scales, for example, using a weighted average method based on confidence.

[0149] The laser stripe fusion center line is combined with the space mapping relationship matrix to establish a nonlinear space mapping relationship. The space mapping relationship matrix describes the transformation relationship between the camera coordinate system and the projector coordinate system.

[0150] The spatial coordinate points corresponding to the laser stripe fusion center line are calculated according to the nonlinear spatial mapping relationship to obtain the initial three-dimensional point cloud data.

[0151] The principal curvature parameters and principal direction parameters of each spatial coordinate point in the initial 3D point cloud data are calculated, for example, by using a local surface fitting method to calculate these parameters.

[0152] The point distribution entropy of the spatial coordinate points is calculated based on the principal curvature parameters and the principal direction parameters. The point distribution entropy reflects the distribution uniformity of the local point cloud.

[0153] The principal curvature parameter, principal direction parameter and point distribution entropy are combined to construct a feature descriptor, which is used to characterize the local geometric features of each point.

[0154] The adaptive weight coefficients of the spatial coordinate points are calculated based on the feature descriptors. For example, points with larger feature differences are given larger weights.

[0155] The initial 3D point cloud data is layered and resampled according to the adaptive weight coefficient to obtain the resampled point cloud data. The resampling process retains important feature points.

[0156] Construct local consistency constraints including point position constraints, normal vector constraints and curvature constraints. These constraints are used to ensure that the registered point cloud data maintains the consistency of local geometric features.

[0157] The local consistency constraint is integrated into the iterative closest point algorithm to obtain a constrained optimization model.

[0158] The resampled point cloud data is iteratively registered based on the constrained optimization model to obtain the registered point cloud data.

[0159] The registered point cloud data is locally fused according to the feature descriptor to obtain the final fused 3D point cloud data. The fusion process takes feature similarity into account and merges similar points.

[0160] Figure 3 The comparison chart of the simulation effect of 3D point cloud reconstruction comprehensively demonstrates the difference in effect between this technical solution and three existing technologies (binocular stereo vision method, traditional structured light method and single Gaussian fitting method) when processing the same complex surface. The results of the four methods are displayed side by side in the figure, and the complete key performance indicators are marked for each method. From the visual effect, the point cloud generated by this technical solution is significantly denser and more evenly distributed, with a point cloud density of 1254 points / cm², which is much higher than other methods. Especially at the edges and details of the surface, the point distribution of this technical solution is more accurate, which is consistent with its 92.7% surface detail retention rate and 89.6% edge sharpness retention rate data. In contrast, the point cloud of the binocular stereo vision method is obviously sparse, with a point cloud density of only 743 points / cm², chaotic point distribution in the edge area, and a surface detail retention rate of only 68.3%, and an edge sharpness retention rate as low as 57.2%, resulting in a reconstruction accuracy error of up to 0.246mm.

[0161] Although the traditional structured light method performs better than the binocular method, with a point cloud density of 895 points / cm², a surface detail retention rate of 76.5%, an edge sharpness retention rate of 69.1%, and an overall score of 7.67, its reconstruction accuracy error of 0.157mm is still significantly higher than the 0.075mm of this technical solution. The single Gaussian fitting rule is between the traditional structured light method and the binocular stereo vision method, with a point cloud density of 862 points / cm², a surface detail retention rate of 72.4%, an edge sharpness retention rate of 64.8%, a reconstruction accuracy error of 0.138mm, and an overall score of 7.29. It can also be clearly seen in the figure that this technical solution not only has an advantage in point cloud density, but also demonstrates its excellent ability to retain the geometric features of objects through the complete presentation of feature lines, which is consistent with its point cloud completeness of 95.3% and an overall quality score of 9.34. Through the intuitive comparison of various methods on the same complex surface and the complete presentation of all key indicators, the figure strongly demonstrates the comprehensive performance advantages achieved by this technical solution through dual Gaussian fitting and local feature preservation algorithm.

[0162] In this embodiment, through double Gaussian fitting and multi-scale processing, the center line of the laser stripe can be extracted more accurately, thereby improving the accuracy of three-dimensional reconstruction; the point cloud resampling and registration fusion strategy based on local features can effectively remove noise points and redundant points, and retain important feature points, thereby improving the integrity of the point cloud data; the application of strategies such as adaptive window, adaptive fusion and local consistency constraint makes this method more robust to noise and complex scenes, and can obtain more stable and reliable three-dimensional reconstruction results.

[0163] In an optional implementation, the fused point cloud data is input into a local curvature adaptive algorithm for noise reduction and outlier removal, and the optimized point cloud data is obtained, including:

[0164] The fused point cloud data is spatially segmented based on the Euclidean distance between points and the point cloud density to obtain multiple point cloud subsets, a KD tree index structure is constructed for each of the point cloud subsets, and a neighbor point set of each point is calculated based on the KD tree index structure;

[0165] For each of the neighboring point sets, a local structure tensor is calculated using a preset tensor voting algorithm, and based on the eigenvalues ​​of the local structure tensor, a normal vector and a tangent plane of each point are decomposed to obtain; an orthogonal basis function family is constructed on the tangent plane, and the orthogonal basis function family is used to fit a local surface, and a first principal curvature and a second principal curvature of each point are calculated according to the local surface;

[0166] Calculate the variance of the first principal curvature and the second principal curvature of each point in the neighboring point set, and when the variance is greater than a preset threshold, mark the corresponding point as a candidate outlier point; calculate the spatial distribution density of the points in the neighboring point set, verify the candidate outlier point based on the spatial distribution density, and obtain an outlier point set;

[0167] For points that do not belong to the abnormal point set, a Gaussian curvature map is constructed based on the first principal curvature and the second principal curvature, a local entropy value of the Gaussian curvature map is calculated, the local entropy value is determined as the local feature strength, a shape parameter and a scale parameter of a smoothing kernel function are calculated according to the local feature strength, and an adaptive smoothing kernel function is constructed; the adaptive smoothing kernel function is applied to the neighboring point set, and a contribution weight of each neighboring point is calculated based on the Euclidean distance between the points and the corresponding angle of the normal vector, wherein the contribution weight decreases as the Euclidean distance between the points increases, and decreases as the corresponding angle of the normal vector increases;

[0168] A geometric feature preserving function is constructed, wherein the geometric feature preserving function includes a position constraint term and a normal vector consistency constraint term, wherein the position constraint term calculates a weighted average deviation of the point position based on the contribution weight, and the normal vector consistency constraint term calculates a weighted average deviation of the normal vector based on the contribution weight; the position and normal vector of the point are updated by iteratively optimizing the geometric feature preserving function, and points in the abnormal point set are eliminated to obtain optimized point cloud data.

[0169] The smooth kernel function specifically refers to giving different weights to data points within a certain range, so that the points close to the center have a greater impact, while the points far from the center have a smaller impact; in point cloud data processing, the smooth kernel function is used to optimize the distribution of points, reduce the impact of noise, and try to retain the details of the surface structure. It adjusts the contribution of each point in the point cloud to make the overall shape smoother, and continuously optimizes the data in the iterative process to make it more consistent with the real geometric features; a good smooth kernel function needs to take into account both noise reduction and feature retention. If the kernel function is too smooth, it may cause the loss of edges and details, making the geometric features blurred; if the smoothing effect is too weak, it may not be able to effectively remove noise. In order to strike a balance between smoothing and feature retention, the influence of the kernel function is usually adjusted in combination with the spatial distance between points and the direction change of the normal vector; in addition, the smooth kernel function can adaptively adjust its parameters to adapt to the characteristics of different regions. For example, in flat areas, a stronger smoothing effect can be used, while in areas with large curvature, such as corners and boundaries, the smoothing degree is reduced to retain more details. This adaptive adjustment method can improve the accuracy of point cloud optimization and make the final three-dimensional morphology more realistic and reliable.

[0170] In a specific implementation, first, the fused point cloud data is spatially segmented. The Euclidean distance between all point pairs in the point cloud is calculated, and the number of points in the neighborhood of each point is counted as the point cloud density. Combining the distance and density information between points, the point cloud is divided into multiple small point cloud subsets. In order to speed up subsequent processing, a KD tree index structure is constructed for each point cloud subset to quickly search for the neighboring points of each point.

[0171] Next, local surface fitting is performed on the points in each point cloud subset. Using the KD tree index structure, the K nearest neighbor points of each point are found. For each set of nearest neighbor points, the local structure tensor is calculated using the tensor voting algorithm. The normal vector and tangent plane of each point are obtained by eigenvalue decomposition of the local structure tensor. A set of orthogonal basis functions, such as polynomial basis functions, are constructed on the tangent plane, and these basis functions are used to fit the local surface. Based on the fitted local surface, the two principal curvatures of each point are calculated: the first principal curvature and the second principal curvature.

[0172] Then, outlier detection is performed. The variance of the two principal curvatures of all points in the neighbor point set is calculated. If the variance exceeds the preset threshold, the corresponding point is marked as a candidate outlier. In order to avoid misjudgment, the spatial distribution density of the points in the neighbor point set is further calculated. If the candidate outlier is located in a low-density area, it is confirmed as an outlier and added to the outlier set.

[0173] For non-abnormal points, local curvature adaptive smoothing is performed. Based on the two principal curvatures of each point, a Gaussian curvature map is constructed. The local entropy value of the Gaussian curvature map is calculated and defined as the local feature strength. According to the local feature strength, the shape parameter and scale parameter of the smoothing kernel function are calculated. The higher the local feature strength, the more drastic the change in local curvature. The shape parameter of the smoothing kernel function should be smaller, and the scale parameter should be smaller to maintain the detailed features. Conversely, the lower the local feature strength, the more gradual the change in local curvature. The shape parameter of the smoothing kernel function should be larger, and the scale parameter should be larger to achieve a stronger smoothing effect.

[0174] After constructing the adaptive smoothing kernel function, apply it to the set of neighboring points. Calculate the contribution weight of each neighboring point. The contribution weight is determined by two parts: the Euclidean distance between points and the angle between the normal vectors. The greater the distance, the greater the angle, and the smaller the contribution weight.

[0175] Finally, the position and normal vector of the point are updated by iteratively optimizing the geometric feature preserving function. The geometric feature preserving function contains position constraint terms and normal vector consistency constraint terms. The position constraint term calculates the weighted average deviation of the point position based on the contribution weight, and the normal vector consistency constraint term calculates the weighted average deviation of the normal vector based on the contribution weight. By minimizing the geometric feature preserving function, the position and normal vector of the point are made smoother, and the points with abnormal point concentration are eliminated, and finally the optimized point cloud data is obtained.

[0176] For example, for a point cloud subset containing 1000 points, set the K value to 20. Calculate the 20 nearest neighbor points of each point, use the tensor voting algorithm to calculate the local structure tensor, and fit the local surface. Assume that the two principal curvatures of one of the points are 0.1 and 0.05 respectively, and the variance of the two principal curvatures of its 20 nearest neighbor points is 0.001, which is less than the preset threshold of 0.01, then the point is not a candidate outlier. The local entropy value of the Gaussian curvature map of the area where the point is located is calculated to be 0.5, and the shape parameter of the smoothing kernel function is determined to be 2 and the scale parameter is 0.1 based on the local entropy value. Calculate the distance and normal vector angle between the point and its 20 nearest neighbor points to obtain the contribution weight of each neighbor point. Update the position and normal vector of the point by iteratively optimizing the geometric feature preservation function.

[0177] In this embodiment, by removing outliers and smoothing noise, the quality of point cloud data can be effectively improved, providing a reliable data basis for subsequent 3D reconstruction, model analysis and other applications; by using a local curvature adaptive smoothing kernel function, the smoothing strength can be adjusted according to the local feature intensity, while removing noise, the geometric features of the point cloud are retained to the maximum extent, avoiding the loss of details caused by excessive smoothing; combining point cloud density and local curvature information for outlier detection can effectively identify various types of outliers, and has strong adaptability to changes in noise levels, ensuring the robustness of the algorithm.

[0178] In an optional implementation, the surface of the flexible circuit board is divided into a plurality of characteristic sub-areas according to the deformation sensitive area, and the characteristic sub-areas are fitted in blocks using an adaptive spline surface to determine the fitting results; the curvature distribution of the flexible circuit board is calculated based on the optimized point cloud data, and the fitting results are optimized and corrected in combination with the strain distribution map to generate the three-dimensional geometric morphology of the flexible circuit board, including:

[0179] Acquire point cloud data of the surface of the flexible circuit board and position data of the deformation sensitive area on the surface of the flexible circuit board; extract a feature enhancement point set based on the position data of the deformation sensitive area, calculate the feature importance value of each point in the feature enhancement point set, and determine a grid division parameter according to the feature importance value, wherein the grid division parameter decreases as the feature importance value increases;

[0180] The point cloud data is adaptively quad-tree partitioned using the grid partitioning parameters to obtain a plurality of feature sub-regions; a local coordinate system is established in each of the feature sub-regions, the point cloud data in the feature sub-region is projected onto the local coordinate system, and the local curvature distribution of the projected point cloud data in the feature sub-region is calculated;

[0181] A weight coefficient is constructed according to the local curvature distribution, and the weight coefficient increases as the local curvature distribution increases; an adaptive spline surface model is constructed based on the weight coefficient, and a surface fitting is performed on the characteristic sub-region through a B-spline basis function to obtain an initial fitting surface of the characteristic sub-region;

[0182] Acquiring strain distribution data of the flexible circuit board, calculating the Gaussian curvature value and the average curvature value of the characteristic sub-region based on the initial fitting surface, and constructing a surface correction function according to the strain distribution data, the Gaussian curvature value and the average curvature value, wherein the correction amount of the surface correction function is positively correlated with the strain distribution data;

[0183] The normal vector of the characteristic sub-region is calculated, and the correction amount of the surface correction function is applied to the initial fitting surface along the direction of the normal vector to obtain an optimized fitting surface; the parameters of the surface correction function are iteratively optimized to match the optimized fitting surface with the strain distribution data, thereby generating the three-dimensional geometric morphology of the flexible circuit board.

[0184] In a specific embodiment, point cloud data of the surface of the flexible circuit board and position data of the deformation sensitive area are obtained. The point cloud data can be obtained by a three-dimensional scanner or other depth camera, and contains the three-dimensional coordinate information of each point on the surface of the flexible circuit board. The position data of the deformation sensitive area can be obtained by manual marking or image processing algorithms. For example, the edge detection algorithm can identify the edges of components on the circuit board, which are usually deformation sensitive areas. Assume that the acquired point cloud data contains 10,000 points, each of which contains three coordinate values ​​of x, y, and z. The position data of the deformation sensitive area contains 100 points, each of which also contains three coordinate values ​​of x, y, and z.

[0185] A feature enhancement point set is extracted based on the position data of the deformation sensitive area, and the feature importance value of each point is calculated. The feature enhancement point set can be the point set of the deformation sensitive area itself, or it can be an extended point set generated based on the position data of the deformation sensitive area. The calculation of the feature importance value can take into account the distance from the point to the deformation sensitive area. The closer the distance, the higher the importance value. For example, a Gaussian function can be used to calculate the feature importance value. The closer the distance to the deformation sensitive area, the larger the Gaussian function value, that is, the higher the feature importance value. Assume that the feature enhancement point set contains 200 points, and the feature importance value of each point is between 0 and 1.

[0186] The meshing parameters are determined based on the feature importance value. The meshing parameters decrease as the feature importance value increases. This means that in areas with high feature importance values, the meshing is finer, and in areas with low feature importance values, the meshing is coarser. For example, a linear function can be used to define the relationship between the meshing parameters and the feature importance value. Assume that when the feature importance value is 1, the meshing parameter is 0.1, and when the feature importance value is 0, the meshing parameter is 1.

[0187] Adaptive quadtree partitioning is performed on the point cloud data using meshing parameters to obtain multiple feature sub-regions. Adaptive quadtree partitioning is a recursive meshing method that divides the point cloud data into multiple sub-regions based on meshing parameters. In areas with high feature importance values, the meshing is finer, resulting in more and smaller sub-regions. Assume that the point cloud data is divided into 100 feature sub-regions.

[0188] A local coordinate system is established in each feature sub-region, and the point cloud data in the feature sub-region is projected to the local coordinate system. The establishment of the local coordinate system can be based on the distribution of the point cloud data in the feature sub-region. For example, the center point of the feature sub-region can be used as the origin of the local coordinate system. Assume that each feature sub-region contains 100 points.

[0189] Calculate the local curvature distribution of the projected point cloud data in the feature sub-region. The local curvature distribution can reflect the curvature of the point cloud data in the feature sub-region. For example, the principal curvature can be used to represent the local curvature distribution.

[0190] The weight coefficient is constructed according to the local curvature distribution. The weight coefficient increases as the local curvature distribution increases. This means that in areas with greater curvature, the weight coefficient is greater, and in areas with less curvature, the weight coefficient is smaller. For example, a linear function can be used to define the relationship between the weight coefficient and the local curvature.

[0191] An adaptive spline surface model is constructed based on the weight coefficient, and the surface fitting is performed on the characteristic sub-region through the B-spline basis function to obtain the initial fitting surface of the characteristic sub-region. The B-spline basis function is a commonly used surface fitting method, which can perform smooth surface fitting on the characteristic sub-region according to the weight coefficient.

[0192] When performing adaptive spline surface block fitting on the feature sub-region, data preparation for the feature sub-region is performed. Obtain the point cloud data set in each feature sub-region, establish a local coordinate system for the point cloud data set, and project all point cloud data into the coordinate system. By analyzing the projected point cloud data, the local curvature values ​​of all points in the sub-region are calculated, which will serve as an important basis for subsequent weight distribution. For example, when a feature sub-region containing 100 points is obtained, first obtain the precise three-dimensional coordinate data of these 100 points, then establish a local coordinate system with the geometric center point of the region as the origin, project the data of these 100 points into the coordinate system, and calculate the local curvature value of each point, which is usually distributed between 0 and 1.

[0193] After completing the data preparation, enter the weight coefficient construction stage. Analyze the local curvature of each point in detail, and assign corresponding weight values ​​to each point based on the curvature value, following the principle that the greater the curvature, the greater the weight. Finally, normalize all weight values ​​to ensure the stability of the value. In the specific implementation, different weight levels can be set according to the curvature value of the point. For example, points with a curvature value of more than 0.8 are assigned a weight of 1.0, points with a curvature value between 0.5 and 0.8 are assigned a weight of 0.8, points with a curvature value between 0.2 and 0.5 are assigned a weight of 0.5, and points with a curvature value below 0.2 are assigned a weight of 0.3. Differentiated weight allocation can better maintain the geometric characteristics of the feature area.

[0194] In the B-spline surface fitting stage, it is necessary to determine the control point grid size of the B-spline surface, evenly arrange the control points within the sub-region, calculate the corresponding parameter coordinates for each data point, and establish a B-spline basis function system. Based on this, a weighted least squares equation group is constructed and solved to obtain the optimal position of the control points, and finally generate an initial fitting surface. For example, a 6×6 control point grid can be set, 36 control points can be evenly distributed within the sub-region, and then the parameter coordinates of 100 data points are calculated respectively, and a third-order B-spline basis function is established. The optimal position of the 36 control points is obtained by solving the weighted equation group, and finally a smooth initial fitting surface is generated.

[0195] After obtaining the initial fitting surface, strain data fusion optimization is performed to obtain the strain distribution data corresponding to the sub-area, calculate the geometric features of the initial fitting surface, including Gaussian curvature and mean curvature, and then construct a surface correction function that takes strain into account. Calculate the normal vector of each point on the surface, apply the correction amount along the direction of the normal vector, and continuously update the shape of the fitting surface. Specifically, a correction rule can be formulated: when the strain value increases by 0.1, the corresponding correction amount increases by 0.05 mm, so that a larger correction will be applied in the high strain area to obtain a more accurate optimized surface.

[0196] During iterative optimization, the surface fitting optimization function is set, and the matching degree between the fitting surface and the strain data is evaluated by continuously adjusting the parameters until the convergence condition is reached and the final fitting result is output. Specifically, when performing iterative optimization of the fitting surface, the surface fitting optimization target is first defined as the minimum difference between the fitting surface and the actual strain data, which is represented by the error between the predicted value and the actual value at each measurement point. The measurement points in important areas can be given higher weight values.

[0197] Initialize the adjustable parameters in the surface correction function. These parameters are used to determine the correction method of the initial fitting surface. They are usually set to decimals of the order of one percent, starting with small corrections and gradually adjusting to the optimal state.

[0198] At the beginning of the optimization phase, the fitting effect under the current parameter settings is calculated, the difference between the fitting value and the actual value at all sampling points is checked, and the weighted sum of the differences is obtained to obtain the overall error value. The sensitivity of the current error to each parameter is analyzed, the impact of parameter changes on the error is judged, and the direction of parameter adjustment is determined accordingly.

[0199] Modify the parameters according to the determined direction, and the adjustment range needs to be moderate to avoid instability or too slow optimization process. Setting optimization termination conditions includes: the overall error is less than the preset standard, the parameter adjustment amount is extremely small, or the maximum number of optimization times is reached.

[0200] Repeat the evaluation, adjustment direction, and adjustment steps until the termination condition is met. Check the results after each round of optimization, and adjust the optimization strategy when the results deteriorate. Reduce the adjustment range when optimization fluctuations occur, increase the adjustment range when progress is slow, and reset the initial value of the parameter when local optimization is in trouble. At the same time, focus on the fitting effect of important areas.

[0201] After the optimization is completed, compare the error improvement before and after the optimization, check the smoothness of the fitting surface, verify the fitting accuracy of the key area, and ensure that the results have physical significance. Apply the final parameters to the correction function to generate the final fitting surface, and record the key parameter settings and effect data of the optimization process.

[0202] This parameter optimization method improves the matching degree between the fitting surface and the actual strain data through systematic adjustment, ensures the optimization convergence and result accuracy, and can obtain accurate three-dimensional geometric morphology reconstruction results.

[0203] Obtain the strain distribution data of the flexible circuit board. The strain distribution data can be obtained through a strain sensor or finite element analysis software. Assume that the strain distribution data is a two-dimensional array, representing the strain value of each point on the surface of the flexible circuit board.

[0204] The Gaussian curvature and mean curvature of the characteristic sub-region are calculated based on the initial fitting surface. Gaussian curvature and mean curvature are important indicators of surface geometry and can reflect the curvature of the surface.

[0205] The surface correction function is constructed based on the strain distribution data, Gaussian curvature value and mean curvature value. The correction amount of the surface correction function is positively correlated with the strain distribution data. This means that in areas with larger strains, the surface correction amount is larger, while in areas with smaller strains, the surface correction amount is smaller.

[0206] The normal vector of the feature sub-region is calculated, and the correction amount of the surface correction function is applied to the initial fitting surface along the direction of the normal vector to obtain the optimized fitting surface. The normal vector is a vector perpendicular to the surface and can indicate the direction of the surface.

[0207] By iteratively optimizing the parameters of the surface correction function, the optimized fitting surface is matched with the strain distribution data to generate the three-dimensional geometric morphology of the flexible circuit board. The iterative optimization process can be achieved by minimizing the error between the optimized fitting surface and the strain distribution data.

[0208] exist Figure 4 In the figure, the line graphs are compared in detail to compare the convergence performance of the four iterative optimization methods in the surface fitting process. The technical solution (triangle mark) shows a rapid error decrease in the early iteration, from 0.165mm in the 0th iteration to 0.0265mm in the 5th iteration, a decrease of about 84%; by the 14th iteration, the error has dropped to 0.0042mm, reaching below the convergence threshold (0.005mm), requiring a total of 14 iterations, and each iteration takes 28.5ms. The gradient descent method (square mark) converges slowly, and does not reach the convergence threshold until the 16th iteration. The convergence speed is about 14% slower than this solution. Although each iteration only takes 19.2ms, the total calculation time is still higher than this solution. The Newton method (circle mark) performs well in the first 4 iterations, quickly reducing the error to 0.0242mm, but then the convergence speed slows down significantly, stabilizing at around 0.007mm, and cannot reach the convergence threshold; and each iteration takes 42.7ms, with the highest calculation cost. The fixed step size method (diamond mark) performed the worst. Even after 20 iterations, the error was still 0.0201mm, which was much higher than the convergence threshold. From the perspective of the change in optimization step size, this technical solution adopted an adaptive step size strategy with an initial step size of 0.085, which gradually decreased with the iteration process, maintaining fast and stable convergence; the Newton method's step size changed dramatically, rapidly decreasing from 0.163 to close to 0, resulting in stagnation in the later optimization; the fixed step size method always remained unchanged at 0.01 and lacked adaptability. These data fully demonstrate the superiority of this technical solution in the iterative optimization process. Through the strategy of "iterative optimization of surface correction function parameters", the convergence speed and computational efficiency were successfully balanced, and the ideal optimization effect was achieved with a smaller number of iterations.

[0209] In this embodiment, through adaptive spline surface fitting and surface correction, the complex shape of the flexible circuit board can be described more accurately, especially in deformation-sensitive areas; adaptive quadtree partitioning can block the point cloud data according to the feature importance value, thereby improving the computational efficiency; and by combining the strain distribution data for surface correction, the influence of noise and error can be effectively eliminated, thereby improving the reliability of the reconstruction results.

[0210] Figure 5 FIG. 1 is a schematic diagram of the structure of a flexible circuit board three-dimensional shape laser scanning measurement and reconstruction system according to an embodiment of the present invention. Figure 5 As shown, the system comprises:

[0211] The first unit is used to apply a preset stress field to the flexible circuit board to be measured, construct a gridded strain monitoring point array on the surface of the flexible circuit board, obtain a strain distribution map of the flexible circuit board through a multispectral camera, and divide the deformation sensitive area based on the strain distribution map; based on the distribution of the deformation sensitive area, calculate the optimal arrangement position of four groups of binocular laser scanning devices, and use an adaptive exposure control strategy to collect an orthogonal grid projection image group of the flexible circuit board;

[0212] The second unit is used to input the calibration data of the four groups of binocular laser scanning devices into an adaptive multi-step calibration algorithm to obtain a spatial mapping relationship matrix; perform double Gaussian fitting on the orthogonal grid projection image group to extract the center line of the laser stripe, combine the center line of the laser stripe with the spatial mapping relationship matrix to construct a depth information mapping model, calculate the three-dimensional coordinates according to the depth information mapping model, generate initial three-dimensional point cloud data, and resample and align and fuse the initial three-dimensional point cloud data through a local feature preservation algorithm to obtain fused point cloud data;

[0213] The third unit is used to input the fused point cloud data into a local curvature adaptive algorithm for noise reduction and abnormal point removal to obtain optimized point cloud data; divide the surface of the flexible circuit board into multiple characteristic sub-areas according to the deformation sensitive area, and use an adaptive spline surface to perform block fitting on the characteristic sub-areas to determine the fitting results; calculate the curvature distribution of the flexible circuit board based on the optimized point cloud data, optimize and correct the fitting results in combination with the strain distribution map, and generate the three-dimensional geometric morphology of the flexible circuit board.

[0214] According to a third aspect of the embodiments of the present invention,

[0215] An electronic device is provided, comprising:

[0216] processor;

[0217] a memory for storing processor-executable instructions;

[0218] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0219] A fourth aspect of the embodiments of the present invention is:

[0220] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0221] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0222] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for measuring and reconstructing the three-dimensional shape of a flexible circuit board by laser scanning, characterized in that: include: Applying a preset stress field to the flexible circuit board to be measured, constructing a gridded strain monitoring point array on the surface of the flexible circuit board, acquiring a strain distribution map of the flexible circuit board through a multispectral camera, and dividing the deformation sensitive area based on the strain distribution map; Based on the distribution of the deformation sensitive area, the optimal arrangement positions of four groups of binocular laser scanning devices are calculated, and an adaptive exposure control strategy is adopted to collect an orthogonal grid projection image group of the flexible circuit board; Input the calibration data of the four groups of binocular laser scanning devices into an adaptive multi-step calibration algorithm to obtain a spatial mapping relationship matrix; perform double Gaussian fitting on the orthogonal grid projection image group to extract the center line of the laser stripe, combine the center line of the laser stripe with the spatial mapping relationship matrix to construct a depth information mapping model, calculate the three-dimensional coordinates according to the depth information mapping model, generate initial three-dimensional point cloud data, and resample and align and fuse the initial three-dimensional point cloud data through a local feature preservation algorithm to obtain fused point cloud data; Input the fused point cloud data into a local curvature adaptive algorithm for noise reduction and abnormal point removal to obtain optimized point cloud data; divide the surface of the flexible circuit board into a plurality of characteristic sub-areas according to the deformation sensitive area, use an adaptive spline surface to perform block fitting on the characteristic sub-areas, and determine the fitting results; The curvature distribution of the flexible circuit board is calculated based on the optimized point cloud data, and the fitting result is optimized and corrected in combination with the strain distribution diagram to generate the three-dimensional geometric morphology of the flexible circuit board.

2. The method according to claim 1, characterized in that Applying a preset stress field to the flexible circuit board to be measured, constructing a gridded strain monitoring point array on the surface of the flexible circuit board, obtaining a strain distribution map of the flexible circuit board through a multispectral camera, and dividing the deformation sensitive area based on the strain distribution map includes: Applying a preset stress field to a flexible circuit board, constructing a gridded strain monitoring point array on the surface of the flexible circuit board, and obtaining a stressed flexible circuit board with the gridded strain monitoring point array; collecting multispectral images of the stressed flexible circuit board before and after the preset stress field is applied, and performing registration processing on the multispectral images to obtain a registered deformation image; Divide the registered deformation image into a plurality of sub-regions, each of which contains at least one monitoring point in the gridded strain monitoring point array; construct a displacement correlation function between adjacent sub-regions, wherein the displacement correlation function includes a gray value distribution and a displacement increment of the sub-region; solve the displacement correlation function by an iterative optimization algorithm to obtain a displacement vector of the sub-region, calculate a displacement gradient of each sub-region based on the displacement vector, and obtain a displacement field distribution of the flexible circuit board; Using a preset radial basis function to interpolate the displacement field distribution, obtain a continuous displacement field of the flexible circuit board, calculate the strain tensor of the flexible circuit board based on the continuous displacement field, obtain a strain distribution map of the flexible circuit board, calculate the principal strain amplitude according to the strain distribution map, determine the area where the principal strain amplitude exceeds a preset strain threshold as a candidate sensitive area, calculate the strain gradient value in the candidate sensitive area, and obtain a strain gradient value distribution map; Based on the strain gradient value distribution map, a strain gradient maximum point is determined as a seed point in the candidate sensitive area, and a strain monitoring network is constructed with the seed point as the center, wherein the strain monitoring network includes a plurality of concentric annular monitoring areas, and a radius difference between adjacent concentric annular monitoring areas is determined according to the strain gradient value distribution map, and a plurality of strain monitoring nodes are set in each of the concentric annular monitoring areas; The correlation between the strain value of each strain monitoring node and the strain value of the strain monitoring node in the adjacent concentric annular monitoring area is calculated, and the strain weight value of each strain monitoring node is determined based on the correlation; the area corresponding to the strain monitoring node whose strain weight value is greater than a preset weight threshold is determined as the deformation sensitive area of ​​the flexible circuit board.

3. The method according to claim 1, characterized in that Based on the distribution of the deformation sensitive areas, calculating the optimal arrangement positions of the four sets of binocular laser scanning devices includes: The deformation sensitive area is represented as a plurality of plane sub-areas, each of which includes a center coordinate and an area weight, and based on the center coordinate and the area weight, four sets of position parameter sets of the binocular laser scanning device are constructed; Calculating the observation angle of each group of binocular laser scanning devices for each of the plane sub-areas, evaluating the laser stripe imaging quality coefficient at each of the observation angles, and calculating the occlusion coefficient of each of the plane sub-areas at each of the observation angles; determining the visibility score of the plane sub-area based on the product of the observation angle, the imaging quality coefficient and the occlusion coefficient, and constructing an optimization objective function based on the area weight and the visibility score; Construct a first candidate solution group and a second candidate solution group based on the position parameter set, each of the candidate solutions carries a set of the position parameter set; the first search radius corresponding to the first candidate solution group is greater than the second search radius corresponding to the second candidate solution group; calculate the optimization objective function value corresponding to each candidate solution; select the candidate solution with the highest optimization objective function value in the first candidate solution group as the first optimal solution, and select the candidate solution with the highest optimization objective function value in the second candidate solution group as the second optimal solution; Taking the first optimal solution as the center, within the scope of the first search radius, randomly generating a corresponding number of new first position parameter sets according to a preset first number, updating the candidate solutions carrying the new first position parameter sets to the first candidate solution group, and obtaining an updated first candidate solution group; taking the second optimal solution as the center, within the scope of the second search radius, randomly generating a corresponding number of new second position parameter sets according to a preset second number, updating the candidate solutions carrying the new second position parameter sets to the second candidate solution group, and obtaining an updated second candidate solution group; Calculating the optimization objective function value of each candidate solution in the updated first candidate solution group and the updated second candidate solution group; if the optimization objective function value of the first optimal solution is greater than the optimization objective function value of the second optimal solution, replacing the candidate solution with the lowest optimization objective function value in the second candidate solution group with the position parameter set of the first optimal solution; The iterative process is repeatedly performed. Based on a preset number of continuous iterations, when the maximum improvement of the optimization objective function value is less than a preset improvement threshold, the position parameter set corresponding to the first optimal solution is determined to be the optimal arrangement scheme of the four groups of binocular laser scanning devices.

4. The method according to claim 1, characterized in that: Inputting the calibration data of the four groups of binocular laser scanning devices into the adaptive multi-step calibration algorithm to obtain the spatial mapping relationship matrix includes: Acquire a calibration image and calibration data of a binocular laser scanning device, extract initial values ​​of intrinsic parameters of the binocular laser scanning device based on the calibration data, wherein the initial values ​​of intrinsic parameters include focal length parameters, principal point coordinate parameters, and radial distortion coefficient parameters; construct a camera intrinsic parameter matrix based on the focal length parameters and the principal point coordinate parameters, and construct a distortion coefficient matrix based on the radial distortion coefficient parameters; and combine the camera intrinsic parameter matrix with the distortion coefficient matrix to construct an initial projection matrix; Extracting a projection point set from the calibration image, dividing the projection points in the projection point set into a central area projection point subset and an edge area projection point subset according to the distance from the image center point; applying the initial projection matrix to the central area projection point subset and the edge area projection point subset, respectively, and calculating the reprojection error mean and reprojection error standard deviation of each area; Determine a regional weight item according to a reprojection error mean and a reprojection error standard deviation corresponding to the region where the projection point is located, determine a distance weight item according to a distance from the projection point to the center point of the image, and determine a dual adaptive weight factor based on the regional weight item and the distance weight item; Multiplying the dual adaptive weight factor by the square of the reprojection error of the corresponding projection point to construct a weighted projection error, using the weighted projection error to construct an optimization objective function, optimizing the projection points of the central area projection point subset and the edge area projection point subset based on the optimization objective function, updating the initial value of the internal parameter, and obtaining an optimized internal parameter matrix; Based on the central area projection point subset, the initial value of the rotation matrix is ​​calculated, and the initial value of the rotation matrix is ​​used as the initial value. In combination with the radial distortion characteristics of the edge area projection point subset, an external parameter matrix is ​​calculated, and the external parameter matrix includes final rotation matrix parameters and translation vector parameters; the optimized internal parameter matrix is ​​combined with the external parameter matrix to construct a spatial mapping relationship matrix; Reprojection errors of the projection points in the central area projection point subset and the edge area projection point subset are calculated respectively. When the reprojection error of the central area projection point subset is greater than a first preset threshold or the reprojection error of the edge area projection point subset is greater than a second preset threshold, the dual adaptive weight factor is updated according to the current reprojection error distribution, and the iteration is continued until the reprojection error of the central area projection point subset is less than the first preset threshold and the reprojection error of the edge area projection point subset is less than the second preset threshold, and the current spatial mapping relationship matrix is ​​determined as the final spatial mapping relationship matrix.

5. The method according to claim 1, characterized in that The orthogonal grid projection image group is subjected to double Gaussian fitting to extract the center line of the laser stripe, the center line of the laser stripe is combined with the spatial mapping relationship matrix to construct a depth information mapping model, three-dimensional coordinates are calculated according to the depth information mapping model, and initial three-dimensional point cloud data is generated. The initial three-dimensional point cloud data is resampled and fused by a local feature preservation algorithm to obtain fused point cloud data including: grayscale the orthogonal grid projection image group to obtain a grayscale image, decompose the grayscale image into multiple layers of sub-images by a multi-scale decomposition method, and determine a corresponding adaptive window based on a local grayscale gradient amplitude of each layer of the multiple layers of sub-images; Sampling points of the grayscale distribution of the laser stripes are collected within the adaptive window, outlier detection is performed on the sampling points to obtain optimized sampling points, and a dual Gaussian distribution model is established based on the optimized sampling points, wherein the dual Gaussian distribution model includes a first Gaussian distribution function and a second Gaussian distribution function; a weighted least squares objective function with a regularization term is constructed, and parameters of the first Gaussian distribution function and the second Gaussian distribution function are solved by minimizing the weighted least squares objective function to obtain a multi-scale laser stripe grayscale distribution fitting model; Extracting the laser stripe center line of each layer of sub-images in the multi-layer sub-images based on the multi-scale laser stripe grayscale distribution fitting model to obtain the multi-scale laser stripe center line, adaptively fusing the multi-scale laser stripe center line to obtain the laser stripe fusion center line, combining the laser stripe fusion center line with the spatial mapping relationship matrix to establish a nonlinear spatial mapping relationship, and calculating the spatial coordinate point corresponding to the laser stripe fusion center line according to the nonlinear spatial mapping relationship to obtain initial three-dimensional point cloud data; Calculate the principal curvature parameters and principal direction parameters of each of the spatial coordinate points in the initial three-dimensional point cloud data, and calculate the point distribution entropy of the spatial coordinate point based on the principal curvature parameters and the principal direction parameters; construct a feature descriptor by combining the principal curvature parameters, the principal direction parameters and the point distribution entropy; calculate the adaptive weight coefficient of the spatial coordinate point based on the feature descriptor; perform hierarchical resampling on the initial three-dimensional point cloud data according to the adaptive weight coefficient to obtain resampled point cloud data; A local consistency constraint including a point position constraint, a normal vector constraint and a curvature constraint is constructed, and the local consistency constraint is integrated into an iterative closest point algorithm to obtain a constraint optimization model; based on the constraint optimization model, the resampled point cloud data is iteratively registered to obtain registered point cloud data; and the registered point cloud data is locally fused according to the feature descriptor to obtain fused three-dimensional point cloud data.

6. The method according to claim 1, characterized in that The fused point cloud data is input into the local curvature adaptive algorithm for noise reduction and abnormal point removal, and the optimized point cloud data obtained includes: The fused point cloud data is spatially segmented based on the Euclidean distance between points and the point cloud density to obtain multiple point cloud subsets, a KD tree index structure is constructed for each of the point cloud subsets, and a neighbor point set of each point is calculated based on the KD tree index structure; For each of the neighboring point sets, a local structure tensor is calculated using a preset tensor voting algorithm, and based on the eigenvalues ​​of the local structure tensor, a normal vector and a tangent plane of each point are decomposed to obtain; an orthogonal basis function family is constructed on the tangent plane, and the orthogonal basis function family is used to fit a local surface, and a first principal curvature and a second principal curvature of each point are calculated according to the local surface; Calculate the variance of the first principal curvature and the second principal curvature of each point in the neighboring point set, and when the variance is greater than a preset threshold, mark the corresponding point as a candidate outlier point; calculate the spatial distribution density of the points in the neighboring point set, verify the candidate outlier point based on the spatial distribution density, and obtain an outlier point set; For points that do not belong to the abnormal point set, a Gaussian curvature map is constructed based on the first principal curvature and the second principal curvature, a local entropy value of the Gaussian curvature map is calculated, the local entropy value is determined as the local feature strength, a shape parameter and a scale parameter of a smoothing kernel function are calculated according to the local feature strength, and an adaptive smoothing kernel function is constructed; the adaptive smoothing kernel function is applied to the neighboring point set, and a contribution weight of each neighboring point is calculated based on the Euclidean distance between the points and the corresponding angle of the normal vector, wherein the contribution weight decreases as the Euclidean distance between the points increases, and decreases as the corresponding angle of the normal vector increases; A geometric feature preserving function is constructed, wherein the geometric feature preserving function includes a position constraint term and a normal vector consistency constraint term, wherein the position constraint term calculates a weighted average deviation of the point position based on the contribution weight, and the normal vector consistency constraint term calculates a weighted average deviation of the normal vector based on the contribution weight; the position and normal vector of the point are updated by iteratively optimizing the geometric feature preserving function, and points in the abnormal point set are eliminated to obtain optimized point cloud data.

7. The method according to claim 1, characterized in that Dividing the surface of the flexible circuit board into a plurality of characteristic sub-areas according to the deformation sensitive area, performing block fitting on the characteristic sub-areas using an adaptive spline surface, and determining fitting results; Calculating the curvature distribution of the flexible circuit board based on the optimized point cloud data, optimizing and correcting the fitting result in combination with the strain distribution diagram, and generating the three-dimensional geometric morphology of the flexible circuit board includes: Acquire point cloud data of the surface of the flexible circuit board and position data of the deformation sensitive area on the surface of the flexible circuit board; extract a feature enhancement point set based on the position data of the deformation sensitive area, calculate the feature importance value of each point in the feature enhancement point set, and determine a grid division parameter according to the feature importance value, wherein the grid division parameter decreases as the feature importance value increases; The point cloud data is adaptively quad-tree partitioned using the grid partitioning parameters to obtain a plurality of feature sub-regions; a local coordinate system is established in each of the feature sub-regions, the point cloud data in the feature sub-region is projected onto the local coordinate system, and the local curvature distribution of the projected point cloud data in the feature sub-region is calculated; A weight coefficient is constructed according to the local curvature distribution, and the weight coefficient increases as the local curvature distribution increases; an adaptive spline surface model is constructed based on the weight coefficient, and a surface fitting is performed on the characteristic sub-region through a B-spline basis function to obtain an initial fitting surface of the characteristic sub-region; Acquiring strain distribution data of the flexible circuit board, calculating the Gaussian curvature value and the average curvature value of the characteristic sub-region based on the initial fitting surface, and constructing a surface correction function according to the strain distribution data, the Gaussian curvature value and the average curvature value, wherein the correction amount of the surface correction function is positively correlated with the strain distribution data; The normal vector of the characteristic sub-region is calculated, and the correction amount of the surface correction function is applied to the initial fitting surface along the direction of the normal vector to obtain an optimized fitting surface; the parameters of the surface correction function are iteratively optimized to match the optimized fitting surface with the strain distribution data, thereby generating the three-dimensional geometric morphology of the flexible circuit board.

8. A flexible circuit board three-dimensional shape laser scanning measurement and reconstruction system, used to implement the method described in any one of claims 1 to 7, characterized in that: include: The first unit is used to apply a preset stress field to the flexible circuit board to be measured, construct a gridded strain monitoring point array on the surface of the flexible circuit board, obtain a strain distribution map of the flexible circuit board through a multi-spectral camera, and divide the deformation sensitive area based on the strain distribution map; Based on the distribution of the deformation sensitive area, the optimal arrangement positions of four groups of binocular laser scanning devices are calculated, and an adaptive exposure control strategy is adopted to collect an orthogonal grid projection image group of the flexible circuit board; The second unit is used to input the calibration data of the four groups of binocular laser scanning devices into an adaptive multi-step calibration algorithm to obtain a spatial mapping relationship matrix; perform double Gaussian fitting on the orthogonal grid projection image group to extract the center line of the laser stripe, combine the center line of the laser stripe with the spatial mapping relationship matrix to construct a depth information mapping model, calculate the three-dimensional coordinates according to the depth information mapping model, generate initial three-dimensional point cloud data, and resample and align and fuse the initial three-dimensional point cloud data through a local feature preservation algorithm to obtain fused point cloud data; The third unit is used to input the fused point cloud data into a local curvature adaptive algorithm for noise reduction and abnormal point removal to obtain optimized point cloud data; divide the surface of the flexible circuit board into multiple characteristic sub-areas according to the deformation sensitive area, use an adaptive spline surface to perform block fitting on the characteristic sub-areas, and determine the fitting results; The curvature distribution of the flexible circuit board is calculated based on the optimized point cloud data, and the fitting result is optimized and corrected in combination with the strain distribution diagram to generate the three-dimensional geometric morphology of the flexible circuit board.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method described in any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images