A method and system for three-dimensional reconstruction of embossing depth of specialty paper

Abstract

The application provides a special paper embossing depth three-dimensional reconstruction method and system, and relates to the technical field of quality detection.The method comprises the following steps: obtaining original image data of a special paper surface; the original image data contains embossing structure, non-uniform diffuse reflection interference caused by matte coating, and plant fiber protrusion information; performing image preprocessing on the original image data, suppressing the non-uniform diffuse reflection interference of the matte coating, enhancing the contrast of the embossing area and the fiber protrusion area, and obtaining optimized image data; based on the optimized image data, performing three-dimensional reconstruction calculation to obtain an initial three-dimensional point cloud, and the initial three-dimensional point cloud contains embossing depth information; performing data interpolation processing on the initial three-dimensional point cloud, that is, for arc-shaped edge corners and narrow line areas, the data loss is filled by increasing the point cloud density to obtain a complete three-dimensional point cloud.The application can realize accurate measurement, dynamic calibration and automatic quality inspection of the embossing depth.

Description

Technical Field

[0001] This invention relates to the field of quality inspection technology, and in particular to a method and system for three-dimensional reconstruction of the embossing depth of special paper. Background Technology

[0002] Specialty paper, with its unique surface texture and physical properties, occupies a core position in high-end packaging, art printing, anti-counterfeiting labels and other fields. Among them, the embossing depth, as a key indicator that determines the texture and function of a product, directly affects the printing adaptability and anti-counterfeiting effect of downstream applications.

[0003] However, the material characteristics and production environment of specialty paper pose various challenges to existing technologies: First, the non-uniform diffuse reflection caused by the matte coating leads to light intensity fluctuations in the laser stripe image, making it difficult for traditional filtering methods to accurately separate interference signals from effective contour information; Second, the random protrusions of plant fibers cause local light intensity attenuation, blurring the grayscale contrast between the embossed area and the fiber area, directly affecting the accuracy of light stripe center extraction; Third, the curved corners and narrow lines in the embossed structure are prone to point cloud loss due to scanning angle limitations. For example, existing methods often use fixed density interpolation without dynamically adjusting parameters based on local curvature features, resulting in excessive deviation between the fitted surface and the actual contour, making it impossible to accurately restore the embossed details; Fourth, the dynamic deformation of paper caused by high-speed movement on the production line lacks a real-time compensation mechanism linked to the detection system. Traditional fixed reference point measurement easily introduces very small systematic errors. For example, most existing technologies use static reference points to establish the measurement coordinate system without considering the stretching and shaking during paper movement. Feature evaluation lacks dual correction of spatial attitude and time dimensions, and the repeatability error of depth measurement within continuous acquisition cycles is far higher than the upper limit of process requirements. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method and system for three-dimensional reconstruction of embossing depth of special paper, which can realize accurate measurement, dynamic calibration and automated quality inspection of embossing depth, and meet the quality control requirements of fine production of special paper.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0006] In a first aspect, a method for three-dimensional reconstruction of the embossing depth of special paper, the method comprising:

[0007] Acquire raw image data of the surface of specialty paper; the raw image data includes information on the embossed structure, non-uniform diffuse reflection interference caused by the matte coating, and plant fiber protrusions.

[0008] Image preprocessing is performed on the original image data to suppress the non-uniform diffuse reflection interference of the matte coating and enhance the contrast between the embossed area and the fiber protrusion area, thereby obtaining optimized image data.

[0009] Based on optimized image data, an initial 3D point cloud is obtained by performing 3D reconstruction calculations. The initial 3D point cloud contains embossing depth information.

[0010] The initial 3D point cloud is subjected to data interpolation processing, that is, for curved corners and narrow texture areas, the point cloud density is increased to fill the data gaps and obtain a complete 3D point cloud;

[0011] Based on the complete 3D point cloud, three reference coordinate points are located at the intersection of the curvature extremum point of the embossed arc corner and the central axis of the narrow texture; a dynamic feature evaluation triangle is constructed based on the three reference coordinate points.

[0012] Based on the feature evaluation of the normal vector and area change rate of the triangular face, a multidimensional depth compensation coefficient is calculated. Based on the multidimensional depth compensation coefficient, the embossing depth, fiber protrusion height and local curvature in the complete 3D point cloud are uniformly corrected to obtain the calibrated 3D depth data.

[0013] The calibrated 3D depth data is evaluated for compliance in real time. Based on the evaluation results, the cycle time requirements of the production line are matched to complete the automated quality inspection process.

[0014] Furthermore, the raw image data of the special paper surface is acquired; the raw image data includes information on the embossed structure, non-uniform diffuse reflection interference caused by the matte coating, and plant fiber protrusions, including:

[0015] Linear laser stripes of a specific wavelength are projected onto the surface of a moving special paper.

[0016] Based on linear laser stripes, a laser stripe image is obtained by modulating the surface of a special paper. The modulation process of the special paper surface includes the deformation of the laser stripes by the embossing structure, the non-uniform diffuse reflection of the matte coating, and the local light intensity attenuation caused by the protrusion of plant fibers.

[0017] Laser stripe images are acquired synchronously at a specific angle. The acquired laser stripe images are then format-converted and cached to form a raw image data sequence containing spatial intensity information.

[0018] Furthermore, image preprocessing is performed on the original image data to suppress the non-uniform diffuse reflection interference of the matte coating and enhance the contrast between the embossed area and the fiber protrusion area, resulting in optimized image data, including:

[0019] The original image data sequence is subjected to frequency domain filtering to suppress non-uniform diffuse reflection interference in a specific frequency band caused by the matte coating, thus obtaining the filtered image data.

[0020] Multi-scale morphological operations are applied to the filtered image data, and the gray-scale contrast features between the embossed contour and the fiber protrusion area are enhanced by structuring element matching to obtain an enhanced contrast image.

[0021] The enhanced contrast image is normalized and denoised to output optimized image data suitable for 3D reconstruction calculations.

[0022] Furthermore, based on the optimized image data, an initial 3D point cloud is obtained by performing 3D reconstruction calculations. The initial 3D point cloud contains embossing depth information, including:

[0023] Laser stripe centerline extraction is performed on optimized image data to obtain a sub-pixel precision light stripe center coordinate sequence;

[0024] Based on the light stripe center coordinate sequence, and combined with the pre-calibrated camera and laser system parameters, a projection mapping relationship between the image coordinate system and the laser plane coordinate system is established.

[0025] Based on the principle of triangulation, the projection mapping relationship is transformed into spatial geometric constraints. The three-dimensional spatial coordinate set is determined by solving the intersection point of the laser plane equation and the projection ray equation emitted by the camera optical center.

[0026] The three-dimensional spatial coordinate set is stitched together and processed with coordinate system unification to obtain an initial three-dimensional point cloud containing embossing depth information.

[0027] Furthermore, data interpolation is performed on the initial 3D point cloud. Specifically, for curved corners and narrow textured areas, the point cloud density is increased to fill in data gaps, resulting in a complete 3D point cloud, including:

[0028] Based on the initial 3D point cloud, the missing data areas of curved corners and narrow texture regions are identified through curvature analysis;

[0029] Based on the geometric features of the identified missing data areas, the point cloud interpolation density and interpolation direction are adaptively determined to form interpolation parameters;

[0030] Based on the interpolation parameters, a moving least squares surface fitting process is performed in the neighborhood of the missing data area. That is, a local reference plane is established with the missing area as the center, and the weight influence value of each neighboring point is allocated based on the spatial distance decay law to establish the weight distribution.

[0031] Based on the weight distribution, the local surface is approximated by the least squares criterion to obtain the fitted surface; based on the fitted surface, new three-dimensional coordinate points are formed according to the preset density, thus obtaining the interpolated point cloud.

[0032] The interpolated point cloud is spatially registered and fused with the initial 3D point cloud to form a complete 3D point cloud.

[0033] Furthermore, based on the complete 3D point cloud, three reference coordinate points are located at the intersection of the curvature extremum point of the embossed arc corner and the central axis of the narrow texture; based on the three reference coordinate points, a dynamic feature evaluation triangle is constructed, including:

[0034] Based on the complete 3D point cloud, the curvature distribution characteristics of the embossed arc corner area are analyzed to locate the extreme points of curvature change;

[0035] Based on the complete 3D point cloud, the center direction feature line of the narrow texture region is extracted to determine the key intersection position of the center direction feature line and the arc-shaped corner contour.

[0036] The located curvature extrema points and the determined key intersection points are combined into three spatial reference points to establish a set of reference coordinate points;

[0037] Based on a set of reference coordinate points, a spatial triangulation method is used to connect three spatial reference points to form the smallest closed unit, thereby constructing a dynamic feature evaluation triangle that covers key feature regions.

[0038] Furthermore, based on the feature evaluation of the normal vector and area change rate of the triangular facets, a multidimensional depth compensation coefficient is calculated. Based on the multidimensional depth compensation coefficient, the embossing depth, fiber protrusion height, and local curvature in the complete 3D point cloud are uniformly corrected to obtain calibrated 3D depth data, including:

[0039] The spatial pose characteristics of the triangular facets are obtained by calculating the unit normal vector of the triangular facets.

[0040] By analyzing the features, the area change rate of the triangular facets during continuous acquisition cycles is evaluated, and the dynamic deformation characteristics of the embossed surface are obtained.

[0041] Based on spatial attitude features and dynamic deformation features, a multidimensional depth compensation coefficient is obtained through weighted fusion calculation.

[0042] Based on the multidimensional depth compensation coefficient, the embossing depth, fiber protrusion height and local curvature in the complete 3D point cloud are collaboratively corrected to form calibrated 3D depth data.

[0043] Furthermore, the calibrated 3D depth data undergoes real-time conformity assessment. Based on the assessment results, the production line's cycle time requirements are matched to complete the automated quality inspection process, including:

[0044] Feature analysis is performed on the calibrated 3D depth data to extract embossing depth distribution parameters and key geometric features;

[0045] The embossing depth distribution parameters are compared point by point with the preset process specification threshold to identify abnormal areas that exceed the tolerance range;

[0046] Based on the identified abnormal regions, the spatial boundaries of the abnormal regions are determined by performing minimum bounding box calculations, and the corresponding quality judgment results are obtained based on the geometric dimensions and positions of the spatial boundaries.

[0047] The quality judgment results are converted into production line control signals. Based on the production line control signals, automatic sorting operations are achieved by adjusting the conveyor cycle.

[0048] Secondly, a three-dimensional reconstruction system for the embossing depth of special paper includes:

[0049] The acquisition module is used to acquire the original image data of the special paper surface; the original image data includes the embossed structure, the non-uniform diffuse reflection interference caused by the matte coating, and the information on the plant fiber protrusions;

[0050] The processing module is used to perform image preprocessing on the original image data. By suppressing the non-uniform diffuse reflection interference of the matte coating and enhancing the contrast between the embossed area and the fiber protrusion area, optimized image data is obtained.

[0051] The calculation module is used to obtain an initial 3D point cloud by performing 3D reconstruction calculations based on optimized image data. The initial 3D point cloud contains embossing depth information. The initial 3D point cloud is then subjected to data interpolation processing, that is, for curved corners and narrow texture areas, the point cloud density is increased to fill the data gaps and obtain a complete 3D point cloud.

[0052] The module is used to locate three reference coordinate points at the intersection of the curvature extremum point of the embossed arc corner and the central axis of the narrow texture, based on the complete 3D point cloud; and to construct a dynamic feature evaluation triangle based on the three reference coordinate points.

[0053] The calibration module is used to evaluate the normal vector and area change rate of the triangular face based on the features, and calculate a multi-dimensional depth compensation coefficient. Based on the multi-dimensional depth compensation coefficient, the embossing depth, fiber protrusion height and local curvature in the complete 3D point cloud are corrected in an integrated manner to obtain the calibrated 3D depth data.

[0054] The execution module is used to perform real-time qualification judgment on the calibrated 3D depth data. Based on the judgment result, it matches the cycle time requirements of the production line to complete the automated quality inspection process.

[0055] Thirdly, a computing device includes:

[0056] One or more processors;

[0057] A storage device for storing one or more programs that, when executed by one or more processors, cause the one or more processors to implement the method.

[0058] Fourthly, a computer-readable storage medium storing a program that, when executed by a processor, implements the method.

[0059] The above-described solution of the present invention has at least the following beneficial effects:

[0060] By employing image preprocessing techniques combining frequency domain filtering and multi-scale morphological operations, the technical problems of non-uniform diffuse reflection interference from matte coatings and blurred grayscale contrast between embossed and fiber areas are overcome, thereby improving the signal-to-noise ratio and regional contrast of laser stripe images and providing a high-precision data foundation for 3D reconstruction. By using adaptive point cloud interpolation based on curvature analysis and moving least squares surface fitting, the technical problems of missing point clouds in embossed arc corners and narrow texture areas, and large fitting deviations in fixed-density interpolation are overcome, thereby controlling the fitting deviation within a small range and accurately restoring the detailed morphology of the embossed areas. By employing multi-dimensional depth compensation techniques based on dynamic feature evaluation of triangular face normal vectors and area change rates, the technical problems of no real-time compensation for dynamic paper deformation on the production line and static benchmark measurement errors exceeding process requirements are overcome, thereby reducing the repeatability error of depth measurement within continuous acquisition cycles. By employing automated judgment methods based on feature analysis, minimum bounding box algorithms, and production line cycle time matching, the technical problems of disconnect between quality inspection and production line linkage and high misjudgment rates are overcome, thereby improving quality inspection response speed and quality control stability, and effectively reducing product misjudgment rates. Attached Figure Description

[0061] Figure 1 This is a flowchart illustrating a method for three-dimensional reconstruction of embossing depth in special paper, provided by an embodiment of the present invention.

[0062] Figure 2 This is a schematic diagram of a three-dimensional reconstruction system for the embossing depth of special paper provided in an embodiment of the present invention. Detailed Implementation

[0063] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0064] like Figure 1 As shown, an embodiment of the present invention proposes a method for three-dimensional reconstruction of the embossing depth of special paper, the method comprising the following steps:

[0065] Step 1, Step 1, acquire the original image data of the special paper surface; the original image data includes the embossed structure, the non-uniform diffuse reflection interference caused by the matte coating, and the information of plant fiber protrusions;

[0066] Step 2: Perform image preprocessing on the original image data. By suppressing the non-uniform diffuse reflection interference of the matte coating and enhancing the contrast between the embossed area and the fiber protrusion area, optimized image data is obtained.

[0067] Step 3: Based on the optimized image data, an initial 3D point cloud is obtained by performing 3D reconstruction calculations. The initial 3D point cloud contains embossing depth information.

[0068] Step 4: Perform data interpolation on the initial 3D point cloud, that is, for curved corners and narrow texture areas, increase the point cloud density to fill the data gaps and obtain a complete 3D point cloud.

[0069] Step 5: Based on the complete 3D point cloud, locate three reference coordinate points at the intersection of the curvature extremum point of the embossed arc corner and the central axis of the narrow texture; construct a dynamic feature evaluation triangle based on the three reference coordinate points.

[0070] Step 6: Based on the feature evaluation of the normal vector and area change rate of the triangular face, calculate a multidimensional depth compensation coefficient; based on the multidimensional depth compensation coefficient, perform integrated correction on the embossing depth, fiber protrusion height and local curvature in the complete 3D point cloud to obtain calibrated 3D depth data.

[0071] Step 7: Perform real-time qualification assessment on the calibrated 3D depth data. Based on the assessment results, match the cycle time requirements of the production line to complete the automated quality inspection process.

[0072] In this embodiment of the invention, by acquiring original image data containing information on embossing structure, non-uniform diffuse reflection interference of matte coating, and plant fiber protrusions, and then preprocessing the image to suppress diffuse reflection interference and enhance the contrast between embossed and fiber areas, high-quality optimized image data is provided. Based on the optimized image, three-dimensional reconstruction is completed to obtain an initial point cloud containing embossing depth. Data interpolation is then performed on curved corners and narrow texture areas to fill in the missing point cloud and obtain a complete three-dimensional point cloud. Next, a dynamic feature evaluation triangle is constructed by locating reference coordinate points. The multi-dimensional depth compensation coefficient is calculated by combining the normal vector and area change rate of the triangle, and the embossing depth, fiber protrusion height, and local curvature are corrected in an integrated manner, significantly improving the accuracy of the three-dimensional depth data. Finally, the calibration data is judged for compliance in real time and matched with the production line cycle, realizing automated quality inspection. Overall, this effectively solves the problems of multiple interferences, incomplete point clouds, large measurement errors, and low quality inspection efficiency in the detection of embossing depth in specialty paper, meeting the quality control requirements of refined production of specialty paper.

[0073] In a preferred embodiment of the present invention, step 1 above may include:

[0074] Step 1.1 involves projecting linear laser stripes of a specific wavelength onto the moving surface of the specialty paper. Specifically, to address the issue of non-uniform diffuse reflection caused by the matte coating of the specialty paper and the dynamic movement of the paper on the production line, the specific wavelength of the linear laser stripes is first determined. Considering that the matte coating is prone to strong diffuse reflection interference in the 400-700nm visible light band, a linear laser in the 850nm near-infrared band is selected. This band effectively reduces the initial reflection interference of the matte coating while ensuring moderate laser penetration, clearly revealing the outline of the embossed structure and the raised plant fibers. Subsequently, a high-power laser emitter is installed above the conveyor rollers of the production line, 30-50cm away from the surface of the specialty paper. The scanning frequency of the emitter must match the conveying speed of the specialty paper. For example, when the conveying speed is 120 meters per minute, the scanning frequency is set to 1000Hz, meaning one continuous linear laser stripe is projected onto the moving surface of the specialty paper every millisecond, ensuring that each embossed area on the paper surface is completely covered by the laser stripe during the paper's movement.

[0075] Step 1.2: Based on linear laser stripes, a laser stripe image is obtained by modulating the surface of a special paper. The modulation process on the special paper surface includes the deformation of the laser stripes by the embossed structure, the non-uniform diffuse reflection of the matte coating, and the local intensity attenuation caused by plant fiber protrusions. Specifically, after the projected linear laser stripes contact the moving special paper surface, they undergo a multi-dimensional modulation process. This process directly corresponds to three key interference and feature sources: the embossed structure, the diffuse reflection of the matte coating, and the plant fiber protrusions. When the laser stripes fall on the embossed structure, the protruding parts of the embossing cause the laser stripes to bend upwards, while the concave parts cause the stripes to bend downwards. The degree of deformation is positively correlated with the embossing depth. This deformation can be used to... The initial characteristics of the embossing are reflected. Simultaneously, the matte coating on the surface of the specialty paper produces non-uniform diffuse reflection of the laser light. Areas with thicker coatings show increased light intensity reflection, while thinner areas show decreased light intensity reflection, resulting in alternating bright and dark light intensity fluctuations in the laser stripe image. This fluctuation is a direct manifestation of interference from the matte coating. Furthermore, the plant fiber protrusions inside the specialty paper partially block the laser light, causing localized light intensity attenuation in the laser stripe segments corresponding to the protrusions, forming dark spots or segments. These dark features precisely record the position and height information of the plant fiber protrusions. Through this complete modulation process, the originally uniform linear laser stripes are transformed into a laser stripe image that includes embossing deformation, diffuse reflection intensity fluctuations, and fiber protrusion attenuation characteristics.

[0076] Step 1.3: Synchronously acquire laser stripe images at a specific angle, convert the acquired laser stripe images into a format and cache the data to form a raw image data sequence containing spatial intensity information. Specifically, this includes: to accurately capture the modulated laser stripe images, a suitable acquisition angle and synchronous acquisition mechanism must be determined first; the determined acquisition angle is pre-calibrated to minimize interference from strong light reflected directly from the matte coating while clearly capturing the deformation details and intensity changes of the laser stripes; simultaneously, the frame rate of image acquisition is matched with the scanning frequency of the laser emitter to ensure that each laser stripe modulated by the special paper surface can be accurately and synchronously acquired, avoiding misalignment between the laser stripes and the acquired image frames; the acquired laser stripe images are initially in RAW format. Although this format can completely preserve the original intensity data, its data processing convenience is low, so it needs to be converted to 166kbps. The grayscale image format retains a sufficient range of light intensity gradients, effectively avoiding the loss of light intensity details during format conversion. This accurately presents the light intensity fluctuations caused by the matte coating and the local light intensity attenuation characteristics caused by plant fiber protrusions. After format conversion, the resulting grayscale images are stored in the cache unit in chronological order of acquisition time to ensure that the acquired image data will not be interrupted, lost, or delayed during continuous operation of the production line. Based on this, combined with the movement state of the special paper, spatial coordinate labels are added to each grayscale image according to the positional relationship between the acquisition time and the corresponding area on the paper surface. Finally, an original image data sequence containing the spatial position information and light intensity information of each image frame is formed.

[0077] In this embodiment of the invention, by projecting linear laser stripes of a specific wavelength onto the surface of moving special paper, the dynamic movement of the paper in the production line can be accurately matched, avoiding the problem that static acquisition methods are difficult to match with the production rhythm. When the laser stripes are modulated by the surface of the special paper, the stripe deformation caused by the embossing structure, the non-uniform diffuse reflection caused by the matte coating, and the local light intensity attenuation caused by the protrusion of plant fibers are completely captured, ensuring that no original information is missed, laying the foundation for distinguishing effective features from interference factors. Then, the laser stripe image is acquired synchronously at a specific angle, and after format conversion and data caching, an original image data sequence containing spatial intensity information is formed. This not only ensures the timeliness matching of the acquired data with the paper movement, but also gives the original data a clear spatial feature dimension, which can directly support image preprocessing and three-dimensional reconstruction calculation. Finally, it provides comprehensive, accurate and production-adaptable original data support for the entire embossing depth three-dimensional reconstruction process, effectively solving the problems of incomplete information and poor adaptability to dynamic production in existing original data acquisition.

[0078] In a preferred embodiment of the present invention, step 2 above may include:

[0079] Step 2.1 involves performing frequency domain filtering on the original image data sequence. By suppressing non-uniform diffuse reflection interference in a specific frequency band caused by the matte coating, filtered image data is obtained. Specifically, this includes: performing frequency domain feature analysis on the original image data sequence; converting each frame of the original image from the spatial domain to the frequency domain to observe the frequency distribution pattern of the grayscale signal; combining the characteristics of non-uniform diffuse reflection interference from the matte coating, it is found that this type of interference forms a specific continuous frequency range in the frequency domain, and this range does not overlap with the frequency band of the effective contour information of the laser stripes; based on this analysis result, the suppression frequency band of the filter is precisely set to the frequency band range corresponding to the matte coating interference; during the filtering process, the attenuation amplitude of the filter is strictly controlled to ensure that only the signal within the interference frequency band is suppressed, without affecting the signal integrity of the frequency band where the effective contour information is located; after filtering, the image is converted back from the frequency domain to the spatial domain to obtain filtered image data.

[0080] Step 2.2 involves applying multi-scale morphological operations to the filtered image data. This is achieved by using structuring element matching to enhance the grayscale contrast between the embossed outline and the fiber protrusion area, resulting in an enhanced contrast image. Specifically, based on the size difference between the embossed structure and the plant fiber protrusion, two sets of structuring elements at different scales are designed: small-scale structuring elements, such as 3×3 pixel rectangles, are used to match the fine features of the plant fiber protrusion, while large-scale structuring elements, such as 7×7 pixel elliptical structures, are used to match the macroscopic morphology of the embossed outline. First, a morphological opening operation using small-scale structuring elements is applied to the filtered image to remove grayscale noise around the fiber protrusion area, allowing the edges of the fiber protrusion to initially appear. Then, a morphological closing operation using large-scale structuring elements is performed to fill in the small depressions at the edges of the embossed outline, enhancing the continuity and integrity of the outline. Through this precise matching of multi-scale structuring elements, the grayscale value of the embossed outline area is specifically enhanced, while the grayscale value of the plant fiber protrusion area is relatively reduced, significantly widening the grayscale difference between the two. After processing, an enhanced contrast image is obtained.

[0081] Step 2.3 involves normalizing and denoising the enhanced contrast image to produce optimized image data suitable for 3D reconstruction calculations. This includes: calculating the maximum and minimum grayscale values ​​for each frame of the enhanced contrast image; mapping all pixel grayscale values ​​to a standard grayscale range of 0 to 255 using linear transformation to ensure consistent grayscale benchmarks across images acquired at different times and on different paper areas, avoiding deviations in laser stripe centerline extraction due to grayscale range differences; after normalization, adaptive median filtering is used for denoising. This filtering method dynamically adjusts the filter window size based on the noise density around the pixel: a larger window is used to filter noise in noisy areas, while a smaller window is used to preserve edge information in areas with detailed features, effectively balancing denoising performance and feature integrity; after normalization and denoising, the image exhibits uniform grayscale distribution, clear details, and minimal noise interference, fully meeting the accuracy requirements of 3D reconstruction calculations, ultimately forming optimized image data suitable for 3D reconstruction calculations.

[0082] In this embodiment of the invention, frequency domain filtering of the original image data sequence precisely suppresses the non-uniform diffuse reflection interference in a specific frequency band caused by the matte coating, effectively separating the interference signal from the effective contour information of the laser stripes, providing a clear data foundation for enhancing the features of the embossed and fiber regions. Multi-scale morphological operations are applied to the filtered image data, and structural element matching is used to specifically enhance the grayscale contrast features between the embossed contour and the fiber protrusion region, solving the problem of blurred grayscale contrast caused by random protrusions of plant fibers, making the contour boundaries of the embossed and fiber easier to identify. Furthermore, the enhanced contrast image is normalized and denoised, eliminating residual grayscale fluctuations and noise interference in the image, resulting in a unified grayscale range and stable details in the processed image data, ultimately forming optimized image data suitable for 3D reconstruction calculations. Through progressive processing, from suppressing interference and enhancing contrast to optimizing output, the image data quality is gradually improved, effectively solving the problems of excessive image interference and unclear features in existing technologies.

[0083] In a preferred embodiment of the present invention, step 3 above may include:

[0084] Step 3.1 involves extracting the laser stripe centerline from the optimized image data to obtain a sub-pixel precision laser stripe center coordinate sequence. Specifically, this includes: based on the optimized image data, preliminary stripe region localization is performed on each frame of the optimized image; high grayscale regions corresponding to the laser stripes are filtered out by setting a grayscale threshold, eliminating interference from image background and a small amount of residual fiber protrusions; next, for the located stripe regions, sub-pixel centerline extraction is performed using the grayscale moment method: first, the stripe region is divided into several pixel segments, the distribution moment of grayscale values ​​within each pixel segment is calculated, and then the sub-pixel center position of each pixel segment is determined by moment value calculation, ensuring that the center coordinate accuracy is within 0.1 pixels; finally, all sub-pixel center positions are arranged according to the acquisition time sequence, based on the row or column order of the image, forming a continuous laser stripe center coordinate sequence. This laser stripe center coordinate sequence can accurately reflect the deformation trajectory of the laser stripes on the surface of the special paper.

[0085] Step 3.2: Based on the light stripe center coordinate sequence and combined with pre-calibrated camera and laser system parameters, establish the projection mapping relationship between the image coordinate system and the laser plane coordinate system. Specifically, the light stripe center coordinate sequence only corresponds to the pixel position in the image coordinate system. To convert it into spatial position, it needs to be combined with pre-calibrated camera and laser system parameters. These pre-calibrated camera and laser system parameters include camera intrinsic parameters and camera extrinsic parameters, such as the camera's position and attitude relative to the laser emitter, and laser plane parameters, such as the equation coefficients of the laser plane in the world coordinate system. The calibration process has been pre-calibrated using standard calibration. After the plate setting and laser projection calibration are completed, ensuring that the parameter error is controlled within the allowable range of the process, based on the above parameters and the coordinate sequence of the light stripe center, the pixel coordinates of the light stripe center in the image coordinate system are first converted into normalized image coordinates in the camera coordinate system through the camera's internal parameters. Then, combined with the camera's external parameters and the laser plane parameters, a correspondence between the normalized image coordinates and the coordinates in the laser plane coordinate system is established. That is, each light stripe center pixel can find its unique projection point on the laser plane through this correspondence. Through this transformation, the projection mapping relationship between the image coordinate system and the laser plane coordinate system is finally established.

[0086] Step 3.3: Based on the principle of triangulation, the projection mapping relationship is transformed into spatial geometric constraints. The three-dimensional spatial coordinate set is determined by solving the intersection point of the laser plane equation and the projection ray equation emitted from the camera's optical center. Specifically, this includes: determining the projection ray equation pointing from the camera's optical center to the center pixel of the light stripe based on the pre-calibrated camera optical center position and normalized image coordinates. The ray in the projection ray equation represents the spatial direction of the camera's observation of that pixel. Simultaneously, the determined laser plane equation is called in conjunction with the laser plane parameters. Since the actual spatial point corresponding to the center of the light stripe lies both on the projection ray and on the laser plane, the simultaneous equations of the projection ray equation and the laser plane equation are solved. The resulting intersection point coordinates are the three-dimensional spatial coordinates corresponding to the center pixel of the light stripe. During the solution process, the validity of each intersection point coordinate is verified, eliminating abnormal intersection points caused by slight paper vibrations, such as coordinate points deviating from the normal embossing depth range, to avoid initial errors introduced by dynamic deformation. All valid intersection point coordinates are summarized to form a three-dimensional spatial coordinate set covering the current acquisition area.

[0087] The specific construction process of the projection ray equation and the laser plane equation is as follows:

[0088] The spatial coordinates of the camera's optical center are retrieved from pre-calibrated parameters. These coordinates are fixed values ​​determined in the world coordinate system, representing the core observation position of the camera lens. Simultaneously, the normalized image coordinates corresponding to the central pixel of the light stripe are extracted. These coordinates reflect the relative position of the pixel on the camera's imaging plane, and their values ​​have been correlated and calibrated with camera intrinsic parameters such as focal length and principal point coordinates. Using the camera's optical center as the starting point of the ray, the direction vector corresponding to the normalized image coordinates is used as the extension direction of the ray. Specifically, the horizontal and vertical coordinate values ​​of the normalized image coordinates are converted into the angles between the ray and the X and Y axes of the camera coordinate system in space. Combined with the camera's extrinsic parameters, the direction is transformed in the world coordinate system, ultimately forming a projection ray equation that originates from the camera's optical center and extends along a specific spatial direction. This ray precisely corresponds to the actual spatial path of the camera's observation of the central pixel of the light stripe.

[0089] Next, the laser plane equation is constructed. From the pre-calibrated laser plane parameters, the coordinates of three non-collinear feature points on the laser plane are extracted. These three feature points were determined during the calibration stage by projecting the laser onto a standard calibration plate, and their spatial positions have been precisely calibrated. Alternatively, the calibrated laser plane normal vector and the coordinates of a fixed point on the plane can be directly retrieved. If the feature point method is used, the position and tilt of the laser plane in the world coordinate system are determined by calculating the vector relationship between the three points based on the spatial coordinates of these three feature points. If the normal vector method is used, the fixed point coordinates are used as the reference point on the plane, and the spatial orientation of the plane is determined by combining the normal vector. Finally, a laser plane equation that can completely describe the laser irradiation area is formed. This equation covers all areas on the surface of the special paper that are projected by the laser stripes, and the actual spatial point corresponding to the center of the light stripe must fall on this plane.

[0090] The simultaneous system of two equations is then solved to obtain the three-dimensional spatial coordinates. Since the actual spatial point corresponding to the center pixel of the light stripe must lie both on the projected ray observed by the camera and on the plane illuminated by the laser, the constructed projection ray equation and the laser plane equation are combined. During the solution process, the intersection of the two equations is analyzed in conjunction with spatial geometric relationships, i.e., a unique spatial point that simultaneously satisfies the constraints of the ray extension direction and the plane position is found. The coordinates of this point are the actual three-dimensional spatial coordinates corresponding to the center pixel of the light stripe. The validity of each obtained intersection point coordinate is then verified. The range of normal embossing depth is retrieved from the process requirements; this range is determined according to the design standards for special paper embossing. The Z-axis value of each intersection point coordinate is compared with this range value. If the value exceeds the upper limit or falls below the lower limit, it indicates that the point may have been observed incorrectly due to slight paper vibration, and is considered an abnormal intersection point. Simultaneously, the spatial distance between the intersection point coordinates and surrounding adjacent points is checked. If the distance between a point and its adjacent points far exceeds the detail size of normal embossing, it is also determined to be an abnormal intersection point.

[0091] Step 3.4 involves performing point cloud stitching and coordinate system unification on the 3D spatial coordinate set to obtain an initial 3D point cloud containing embossing depth information. Specifically, this includes: analyzing the 3D spatial coordinate sets corresponding to adjacent acquisition frames, finding the overlapping area between the two coordinate sets. This overlapping area typically contains the same embossing feature points, such as obvious embossing protrusions or depressions. Using these feature points as a reference, the coordinate sets of adjacent frames are aligned to ensure that the stitched point cloud has no significant misalignment in the overlapping area. Next, all aligned local coordinate sets are uniformly transformed to a preset world coordinate system. This preset world coordinate system uses the central axis of the production line conveyor rollers as the reference axis, ensuring that the coordinates of all acquisition areas reflect their actual positions on the production line. During coordinate unification, the coordinate offset caused by the paper's movement is compensated based on the movement speed and acquisition frame rate of the special paper to avoid point cloud discontinuities caused by continuous acquisition. Finally, the unified 3D spatial coordinate set is deduplicated, removing duplicate coordinate points to form point cloud data continuously covering the surface of the special paper, i.e., the initial 3D point cloud.

[0092] In an embodiment of the present invention,

[0093] In a preferred embodiment of the present invention, step 4 above may include:

[0094] Step 4.1: Based on the initial 3D point cloud, identify data gaps in curved corners and narrow texture areas through curvature analysis. Specifically, this includes: dividing the initial 3D point cloud into several local point cloud blocks according to a preset spatial grid, ensuring each block covers an independent embossing feature area; calculating the spatial curvature value and rate of change of curvature for each point in each local point cloud block; for curved corner areas, the curvature value is significantly higher than other flat areas and the curvature change is continuous. When the curvature values ​​of multiple consecutive points within a local block exceed a preset threshold, the area is marked as a potential curved corner missing area. The preset threshold is based on normal embossing... The curvature range of the smooth texture area is set in advance; for the narrow texture area, its point cloud density is significantly lower than the surrounding area, and there is a point cloud break along the texture direction. When the point cloud density in a certain local block is lower than the preset density threshold, and there is a point cloud discontinuity along a specific direction, the area is marked as a potential missing area of ​​narrow texture. The preset density threshold is set in advance based on the average point cloud density of the complete embossed area. Finally, edge verification is performed on the two types of potential missing areas. By comparing the feature continuity of adjacent point cloud blocks, the mislabeled normal areas are excluded, and the data missing parts of the arc corners and narrow texture areas are accurately determined.

[0095] Step 4.2: Based on the geometric features of the identified missing data areas, adaptively determine the point cloud interpolation density and direction to form interpolation parameters. Specifically, for missing arc-shaped corners, first measure the radius of curvature; the smaller the radius of curvature, the steeper the arc, indicating more refined embossing details in the area, and the higher the interpolation density should be set. At the same time, determine the interpolation direction based on the tangent direction of the curvature to ensure that the interpolation points are continuously distributed along the arc contour. For missing areas with narrow textures, first measure the texture width and the direction of the central axis; the narrower the texture width, the more refined the embossing structure in the area, and the higher the interpolation density should be set. The interpolation direction is along the direction of the texture's central axis to avoid the interpolation points deviating from the texture contour. The determined interpolation density and direction are associated and stored according to the location of the missing area to form exclusive interpolation parameters for each missing area.

[0096] Step 4.3: Based on the interpolation parameters, perform a moving least squares surface fitting process in the neighborhood of the missing data area. This involves establishing a local reference plane centered on the missing area and allocating weights to neighboring points based on spatial distance decay principles to establish a weight distribution. Specifically, this includes: using the center of each missing area as the origin, determining the neighborhood range based on the interpolation density in the interpolation parameters; the higher the interpolation density, the smaller the neighborhood range. Extracting all valid point clouds within this neighborhood from the initial 3D point cloud, excluding points from other missing areas, and then using the average normal vector of these valid point clouds... Using the target as a reference, a local reference plane is established to ensure that the plane fits the actual contour trend around the missing part. Then, the weight influence value of each neighboring valid point is assigned based on the spatial distance decay law. The spatial distance from each neighboring valid point to the center of the missing part is calculated. The closer the point is, the greater its influence on the fitted surface, and the higher the weight value is set. Through such weight allocation, the fitted surface is made closer to the actual contour shape around the missing part, avoiding fitting deviation caused by not considering the influence of distance. The coordinates of each neighboring valid point are associated with the corresponding weight value to form a weight distribution covering the neighborhood range.

[0097] Step 4.4: Based on the weight distribution, the local surface is optimally approximated using the least squares criterion to obtain a fitted surface. Based on the fitted surface, new three-dimensional coordinate points are formed according to a preset density, thus obtaining the interpolated point cloud. Specifically, this includes: extracting the coordinates and corresponding weight values ​​of neighboring valid points in the weight distribution, analyzing the spatial distribution trend of these points, and then adjusting the surface shape using the least squares criterion. This involves comparing the error between the actual coordinates of each neighboring valid point and the predicted position of the current surface, continuously fine-tuning the spatial orientation of the surface to minimize the sum of squared errors of all neighboring valid points, ultimately obtaining a fitted surface that closely matches the actual contour of the neighborhood. Subsequently, based on the interpolation density and interpolation direction determined in the interpolation parameters, new three-dimensional coordinate points are selected on the fitted surface. The selection positions are uniformly determined along the interpolation direction at intervals set by the interpolation density. The coordinates of each selected position are calculated by referring to the spatial positions of surrounding neighboring valid points and the contour trend of the fitted surface, ensuring that these new coordinate points can continuously fill the gaps in the point cloud of the missing parts. All newly formed three-dimensional coordinate points are organized according to the positions of the missing parts to obtain the interpolated point cloud for the missing parts.

[0098] Step 4.5 involves spatially registering and fusing the interpolated point cloud with the initial 3D point cloud to form a complete 3D point cloud. This includes: first, performing spatial registration by extracting feature points from the overlapping areas of both the initial 3D point cloud and the interpolated point cloud. These feature points include valid points in the neighborhood of the missing area and points in the interpolated point cloud that are close to their neighbors. By calculating the spatial transformation matrices between these feature points (commonly translation and rotation matrices), the coordinates of the interpolated point cloud are adjusted to ensure that the coordinates of the interpolated point cloud and the initial 3D point cloud are completely consistent in the overlapping area, avoiding point cloud misalignment due to coordinate deviations. Next, point cloud fusion is performed by adding the registered interpolated point cloud to the initial 3D point cloud. The fused point cloud is then deduplicated by removing duplicate points with identical coordinates or very close distances to ensure uniform point cloud density and no redundant data. After fusion, point cloud data covering all embossed areas on the surface of the special paper is obtained, i.e., the complete 3D point cloud.

[0099] In this embodiment of the invention, based on the initial 3D point cloud, curvature analysis is used to identify data gaps in curved corners and narrow texture areas, effectively overcoming the problem of inaccurate positioning of point cloud gaps due to scanning angle limitations. Based on the identified geometric features of the gaps, the interpolation density and direction are adaptively determined to form interpolation parameters, avoiding the shortcomings of existing technologies that use fixed-density interpolation and mismatch with local features, allowing the interpolation parameters to accurately adapt to the morphological requirements of different gap areas. Based on the interpolation parameters, moving least-squares surface fitting is performed in the neighborhood of the gap. By establishing a local reference plane and allocating weights to neighboring points according to spatial distance decay, the problem of uneven influence of neighboring points on the fitting results is overcome. According to the weight distribution, the least-squares criterion is used to achieve optimal approximation of the local surface and generate an interpolated point cloud, filling the coordinate data gaps of the gaps and solving the problem of excessive deviation between existing fitting methods and the actual contour. The interpolated point cloud is spatially registered and fused with the initial 3D point cloud to finally form a complete 3D point cloud, thoroughly improving the incompleteness of the initial 3D point cloud.

[0100] In a preferred embodiment of the present invention, step 5 above may include:

[0101] Step 5.1: Based on the complete 3D point cloud, analyze the curvature distribution characteristics of the embossed arc-shaped corner region to locate the extreme points of curvature change. Specifically, this includes: selecting all point clouds belonging to the arc-shaped corner region from the complete 3D point cloud, which are the identified and interpolated parts; arranging these points continuously according to the arc direction, calculating the curvature value of each point and the rate of curvature change of adjacent points; observing the curvature change trend, when the rate of curvature change of a certain point changes from a positive value to a negative value, indicating that the point is the turning point of the arc curvature, and marking it as the extreme point of curvature change; locating at least one such extreme point for each arc-shaped corner region, these points can accurately reflect the key morphological characteristics of the arc-shaped corner.

[0102] Step 5.2: Based on the complete 3D point cloud, extract the center direction feature line of the narrow texture region to determine the key intersection position of the center direction feature line and the arc-shaped corner contour. Specifically, this includes: extracting the center direction feature line of the narrow texture region and determining the key intersection position of the center direction feature line and the arc-shaped corner contour; first, extract all point clouds belonging to the narrow texture region from the complete 3D point cloud, which are continuous point sets formed after interpolation; along the extension direction of the texture, perform straight line fitting on these points to obtain the center direction feature line of the texture, referring to the edge point positions on both sides of the texture during the fitting process to ensure that the center line is at the geometric center of the texture; then, extend the center direction feature line to the arc-shaped corner region and find its intersection point with the arc-shaped corner contour. This intersection point is the key intersection position of the center direction feature line and the arc-shaped corner contour. One such intersection position is determined for each narrow texture region. The key intersection position is associated with the features of both the narrow texture and the arc-shaped corner.

[0103] Step 5.3: Combine the located curvature extreme points and the determined key intersection points into three spatial reference points to establish a set of reference coordinate points. Specifically, this includes: selecting two representative points from the curvature change extreme points, which come from different arc-shaped corner areas and can reflect the main bending characteristics of the embossed structure; selecting one point from the key intersection points, which is located in the central area of ​​the embossed structure and can simultaneously associate the main arc-shaped corners and narrow textures; combining these three points, i.e., the three spatial reference points, to ensure that they form a triangular distribution in space and cover the key feature areas of the embossing, avoiding instability of the reference due to the collinearity of the three points; recording the three-dimensional coordinates of these three spatial reference points to form a set of reference coordinate points, which includes both the feature positions of the arc-shaped corners and the intersection points of the textures and arcs.

[0104] Step 5.4: Based on the set of reference coordinate points, a dynamic feature evaluation triangle covering key feature areas is constructed by using a spatial triangulation method that connects three spatial reference points to form the smallest closed unit. Specifically, this includes: obtaining the real-time three-dimensional coordinates of three spatial reference points in the set of reference coordinate points. These coordinates change dynamically as the special paper moves on the production line, such as during stretching or shaking; using the spatial triangulation method, connecting these three spatial reference points in pairs to form the three sides of a triangle, thus constituting a spatial triangle; the three vertices of this triangle always correspond to the three spatial reference points in the set of reference coordinate points, and can change shape synchronously with the dynamic deformation of the paper; the size and shape of the triangle are adjusted according to the relative positions of the three spatial reference points to ensure that it always covers key feature areas such as the curved corners and narrow lines of the embossing; the feature evaluation triangle constructed in this way is no longer fixed, but can reflect the dynamic posture of the paper in real time.

[0105] In this embodiment of the invention, relying on a complete three-dimensional point cloud, the curvature distribution of the embossed arc-shaped corner area is analyzed to locate the extreme points of curvature change. This solves the problem that static reference points are difficult to accurately capture the dynamic features of arc-shaped corners, providing key positional basis for determining reference coordinate points in the arc-shaped area. Based on the complete three-dimensional point cloud, the central direction feature line of the narrow texture is extracted and its key intersection with the arc-shaped corner is determined, filling the gap of lacking clear reference marks in the narrow texture area and supplementing key positional data associated with texture features. The curvature extreme points and key intersections are combined into three spatial reference points, avoiding the defect of a single reference point being easily affected by local errors, forming a stable reference foundation covering the key area. Based on these spatial reference points, a dynamic feature evaluation triangle is constructed through spatial triangulation. This triangle can be adjusted synchronously with the dynamic deformation of the paper, overcoming the problem that traditional fixed reference surfaces cannot adapt to the stretching and shaking of paper on the production line, ultimately forming a dynamic evaluation carrier covering the key feature area.

[0106] In a preferred embodiment of the present invention, step 6 above may include:

[0107] Step 6.1 involves calculating the unit normal vector of the feature evaluation triangle to obtain its spatial orientation features. This includes: acquiring the real-time 3D coordinates of the three vertices of the feature evaluation triangle; calculating the vectors of two adjacent sides of the triangle based on these coordinates (the vector pointing from the first vertex to the second vertex and the vector pointing from the first vertex to the third vertex); performing a cross product on these two vectors to obtain the original normal vector perpendicular to the triangle, whose direction initially reflects the tilting trend of the triangle in space; then normalizing the original normal vector by adjusting its length to 1 to obtain the unit normal vector. The direction parameter of the unit normal vector, such as its angle with the coordinate system axes, accurately describes the spatial orientation of the triangle. For example, a larger angle indicates that the triangle is tilted upwards, and a smaller angle indicates that it is tilted downwards. Through this calculation, the spatial orientation features of the triangle are finally obtained, which can capture the spatial orientation changes of the embossed surface caused by paper vibration in real time.

[0108] Step 6.2 involves analyzing the area change rate of the triangular facet within a continuous acquisition cycle to obtain the dynamic deformation characteristics of the embossed surface. Specifically, this includes: setting an acquisition cycle; calculating the area of ​​the triangular facet in the current cycle based on the real-time coordinates of the three vertices of the triangular facet; then retrieving the triangular facet area data stored in the previous acquisition cycle, calculating the difference between the current cycle area and the previous cycle area, and dividing this difference by the previous cycle area to obtain the area change rate; a positive change rate indicates that the triangular facet area has increased, corresponding to stretching deformation of the paper in that area; a negative change rate indicates that the triangular facet area has decreased, corresponding to shrinking deformation of the paper in that area; then, trend analysis is performed on the area change rate over multiple consecutive acquisition cycles to eliminate random fluctuations and determine the overall trend of the paper's dynamic deformation, such as continuous stretching or periodic shrinkage; through this analysis, the dynamic deformation characteristics of the embossed surface are finally obtained, which can reflect the dynamic deformation of the paper caused by the high-speed movement of the production line in real time.

[0109] Step 6.3: Based on spatial attitude features and dynamic deformation features, a multi-dimensional depth compensation coefficient is obtained through weighted fusion calculation. Specifically, this includes: setting the weight values ​​of the two features according to the process requirements of specialty paper production; if the process is more sensitive to changes in spatial attitude, such as stricter embossing depth tolerance, the weight of the spatial attitude feature is set higher, and the weight of the dynamic deformation feature is set lower; if the process is more sensitive to dynamic deformation, such as easily stretched paper, the weight ratio is adjusted; then, the attitude compensation parameter corresponding to the spatial attitude feature, such as the compensation value calculated based on the unit normal vector angle, is multiplied by... The weighting process involves multiplying the deformation compensation parameters corresponding to the dynamic deformation features, such as the compensation value calculated based on the area change rate, by their respective weights, and then adding the two products together to obtain a preliminary fusion compensation value. This preliminary fusion compensation value is then corrected by incorporating the correlation errors between attitude and deformation in historical detection data, such as misjudgments of deformation caused by attitude tilt, to ensure that the compensation value can simultaneously adapt to attitude changes and dynamic deformation. Through this weighted fusion calculation, a multidimensional depth compensation coefficient is finally obtained, which includes both attitude compensation and deformation compensation dimensions.

[0110] Step 6.4: Based on the multidimensional depth compensation coefficient, the embossing depth, fiber protrusion height, and local curvature in the complete 3D point cloud are collaboratively corrected to form calibrated 3D depth data. Specifically, this includes: extracting the original embossing depth data of each point in the complete 3D point cloud; adding the attitude compensation component and deformation compensation component from the multidimensional depth compensation coefficient to the original depth data to obtain the corrected embossing depth; if the compensation coefficient is positive, it indicates that the original depth is too small and needs to be corrected upwards; if the compensation coefficient is negative, it indicates that the original depth is too large and needs to be corrected downwards; then, for the fiber protrusion height, the corrected embossing depth is referenced to the average of the embossing around that point. The depth difference, combined with the local correction ratio in the compensation coefficient, is used to adjust the fiber protrusion height data to ensure that the relative relationship between the protrusion height and the embossing depth conforms to the actual shape. For example, after the embossing depth is corrected, the protrusion height needs to be corrected simultaneously to keep the difference between the two constant. Then, the local curvature is corrected. Based on the corrected embossing depth data, the local curvature value of each point is recalculated to eliminate curvature anomalies caused by depth errors, such as curvature being too small due to an excessively large original depth, ensuring that the curvature value can accurately reflect the arc shape of the embossing. Finally, the corrected embossing depth, fiber protrusion height, and local curvature data are integrated to form calibrated three-dimensional depth data.

[0111] In this embodiment of the invention, the unit normal vector of the triangular facet is calculated based on dynamic feature evaluation to obtain spatial attitude features. This overcomes the deficiency of traditional static benchmarks in lacking spatial attitude evaluation, accurately captures the tilt and deflection state of the embossed surface, and provides attitude basis for depth compensation. The area change rate of the triangular facet is analyzed during the continuous acquisition cycle, and the dynamic deformation of the paper caused by high-speed movement is monitored in real time, solving the problem of lack of dynamic deformation perception and supplementing dynamic state data. The spatial attitude and dynamic deformation features are fused to obtain a multi-dimensional depth compensation coefficient, avoiding the limitation of incomplete compensation under a single feature dimension and significantly improving the compensation accuracy. Based on the multi-dimensional depth compensation coefficient, the embossing depth, fiber protrusion height and local curvature are synergistically corrected to overcome the data contradictions that are easily caused by traditional single parameter correction. Finally, calibrated three-dimensional depth data is formed, which effectively reduces the repeatability error of depth measurement during the continuous acquisition cycle and ensures that the data truly reflects the actual state of the product.

[0112] In a preferred embodiment of the present invention, step 7 above may include:

[0113] Step 7.1 involves performing feature analysis on the calibrated 3D depth data to extract embossing depth distribution parameters and key geometric features. Specifically, this includes: first, traversing the calibrated 3D depth data, calculating the depth values ​​of all points within the embossing area, and statistically determining the maximum, minimum, and average depth values. These values ​​collectively constitute the embossing depth distribution parameters, which can intuitively reflect whether the overall embossing depth meets the process expectations; next, for the curved corner areas of the embossing, extracting the curvature value of each arc to verify whether the curvature matches the designed bending shape; for narrow texture areas, measuring the width and length of the texture to confirm whether the texture size meets the standards; and finally, integrating these depth distribution parameters with the arc curvature, texture size, etc., to form key geometric features.

[0114] Step 7.2 involves comparing the embossing depth distribution parameters with preset process specification thresholds point by point to identify abnormal areas that exceed the tolerance range. Specifically, this includes: retrieving preset thresholds based on the special paper production process, including the acceptable range threshold for embossing depth, the allowable threshold for curved corner curvature, and the tolerance threshold for narrow texture width; then examining each point in the calibrated 3D depth data point by point, comparing the actual depth value of that point with the acceptable depth range threshold, and marking the point as a depth anomaly if the actual depth exceeds the upper limit or falls below the lower limit of the threshold; simultaneously comparing the actual curved curvature with the allowable curvature threshold and the actual texture width with the texture tolerance threshold, marking curvature anomalies and texture size anomalies; finally, classifying all anomalies by spatial location to form continuous anomaly areas, avoiding misjudging a single isolated anomaly as an overall non-compliance.

[0115] Step 7.3: Based on the identified abnormal regions, the spatial boundaries of the abnormal regions are determined by performing minimum bounding box calculations. The corresponding quality judgment results are obtained based on the geometric dimensions and location of the spatial boundaries. Specifically, this includes: statistically analyzing the spatial coordinates of all abnormal points in each abnormal region to find the maximum and minimum coordinate values ​​of the abnormal points in the X, Y, and Z axes; performing minimum bounding box calculations, i.e., using these maximum and minimum coordinate values ​​as boundaries, constructing a minimum cuboid that can completely enclose the abnormal region. The length, width, and height of this cuboid are the geometric dimensions of the spatial boundary of the abnormal region; then checking the embossing position corresponding to this spatial boundary to determine if it is located in a critical area of ​​the product, such as the area where the anti-counterfeiting label is located or a critical area for printing adaptation; if the geometric dimensions of the abnormal region are smaller than the preset allowable tolerance size and are not in a critical area, the product is judged as qualified; if the geometric dimensions exceed the allowable tolerance size or are in a critical area, the product is judged as unqualified, forming a clear quality judgment result.

[0116] Step 7.4 converts the quality judgment result into a production line control signal. Based on the production line control signal, automatic sorting is achieved by adjusting the conveyor cycle time. Specifically, this includes: first, establishing the correspondence between the quality judgment result and the production line control signal; if the judgment result is qualified, a normal conveying control signal is generated; if the judgment result is unqualified, a deceleration sorting control signal is generated; the control signal is transmitted to the production line; when a normal conveying signal is received, the current conveying cycle time is maintained, allowing qualified products to continue flowing to subsequent processing stages; when a deceleration sorting signal is received, the conveying speed at the location of the product is appropriately reduced to give the sorting mechanism more time to operate; the sorting mechanism identifies the location of unqualified products according to the control signal, and when the product arrives at the sorting station, it is transferred from the normal conveyor line to the unqualified product collection area, completing automatic sorting and realizing real-time linkage between the quality inspection process and the production line.

[0117] In this embodiment of the invention, feature analysis is performed on the calibrated three-dimensional depth data to extract embossing depth distribution parameters and key geometric features, providing accurate feature basis for quality judgment. These parameters are compared point by point with preset process specification thresholds to accurately identify abnormal areas that exceed tolerances, avoiding missed or false judgments in traditional inspections. Based on the abnormal areas, the spatial boundary is determined by minimum bounding box calculation, and the quality judgment result is obtained by combining the geometric dimensions and position of the boundary, making the judgment more objective and accurate. The judgment result is converted into a production line control signal, and automatic sorting is achieved by adjusting the conveyor cycle, solving the problem of poor linkage between traditional quality inspection and production line. The whole process forms a closed-loop quality inspection process from data analysis to automatic sorting, effectively improving the quality control efficiency and automation level of specialty paper production, and meeting the quality inspection needs of refined production.

[0118] like Figure 2As shown, embodiments of the present invention also provide a three-dimensional reconstruction system for the embossing depth of special paper, comprising:

[0119] The acquisition module is used to acquire the original image data of the special paper surface; the original image data includes the embossed structure, the non-uniform diffuse reflection interference caused by the matte coating, and the information on the plant fiber protrusions;

[0120] The processing module is used to perform image preprocessing on the original image data. By suppressing the non-uniform diffuse reflection interference of the matte coating and enhancing the contrast between the embossed area and the fiber protrusion area, optimized image data is obtained.

[0121] The calculation module is used to obtain an initial 3D point cloud by performing 3D reconstruction calculations based on optimized image data. The initial 3D point cloud contains embossing depth information. The initial 3D point cloud is then subjected to data interpolation processing, that is, for curved corners and narrow texture areas, the point cloud density is increased to fill the data gaps and obtain a complete 3D point cloud.

[0122] The module is used to locate three reference coordinate points at the intersection of the curvature extremum point of the embossed arc corner and the central axis of the narrow texture, based on the complete 3D point cloud; and to construct a dynamic feature evaluation triangle based on the three reference coordinate points.

[0123] The calibration module is used to evaluate the normal vector and area change rate of the triangular face based on the features, and calculate a multi-dimensional depth compensation coefficient. Based on the multi-dimensional depth compensation coefficient, the embossing depth, fiber protrusion height and local curvature in the complete 3D point cloud are corrected in an integrated manner to obtain the calibrated 3D depth data.

[0124] The execution module is used to perform real-time qualification judgment on the calibrated 3D depth data. Based on the judgment result, it matches the cycle time requirements of the production line to complete the automated quality inspection process.

[0125] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for three-dimensional reconstruction of embossing depth in special paper, characterized in that, The method includes: Acquire raw image data of the surface of specialty paper; the raw image data includes information on the embossed structure, non-uniform diffuse reflection interference caused by the matte coating, and plant fiber protrusions. Image preprocessing is performed on the original image data to suppress non-uniform diffuse reflection interference from the matte coating and enhance the contrast between the embossed area and the fiber protrusion area, resulting in optimized image data. This includes: performing frequency domain filtering on the original image data sequence to suppress non-uniform diffuse reflection interference in specific frequency bands caused by the matte coating, resulting in filtered image data; applying multi-scale morphological operations to the filtered image data to enhance the gray-level contrast features between the embossed contour and the fiber protrusion area through structuring element matching, resulting in an enhanced contrast image; and normalizing and denoising the enhanced contrast image for output, forming optimized image data suitable for 3D reconstruction calculations. Based on optimized image data, an initial 3D point cloud is obtained by performing 3D reconstruction calculations. The initial 3D point cloud contains embossing depth information. The initial 3D point cloud is subjected to data interpolation processing, that is, for curved corners and narrow texture areas, the point cloud density is increased to fill the data gaps and obtain a complete 3D point cloud; Based on the complete 3D point cloud, three reference coordinate points are located at the intersection of the curvature extremum point of the embossed arc corner and the central axis of the narrow texture; a dynamic feature evaluation triangle is constructed based on the three reference coordinate points. Based on the feature evaluation of the normal vector and area change rate of the triangular face, a multidimensional depth compensation coefficient is calculated. Based on the multidimensional depth compensation coefficient, the embossing depth, fiber protrusion height and local curvature in the complete 3D point cloud are uniformly corrected to obtain the calibrated 3D depth data. The calibrated 3D depth data is evaluated for compliance in real time. Based on the evaluation results, the cycle time requirements of the production line are matched to complete the automated quality inspection process.

2. The method for three-dimensional reconstruction of embossing depth in special paper according to claim 1, characterized in that, Acquire raw image data of the special paper surface; the raw image data includes information on the embossed structure, non-uniform diffuse reflection interference caused by the matte coating, and plant fiber protrusions, including: Linear laser stripes of a specific wavelength are projected onto the surface of a moving special paper. Based on linear laser stripes, a laser stripe image is obtained by modulating the surface of a special paper. The modulation process of the special paper surface includes the deformation of the laser stripes by the embossing structure, the non-uniform diffuse reflection of the matte coating, and the local light intensity attenuation caused by the protrusion of plant fibers. Laser stripe images are acquired synchronously at a specific angle. The acquired laser stripe images are then format-converted and cached to form a raw image data sequence containing spatial intensity information.

3. The method for three-dimensional reconstruction of embossing depth in special paper according to claim 2, characterized in that, Based on optimized image data, an initial 3D point cloud is obtained by performing 3D reconstruction calculations. The initial 3D point cloud contains embossing depth information, including: Laser stripe centerline extraction is performed on optimized image data to obtain a sub-pixel precision light stripe center coordinate sequence; Based on the light stripe center coordinate sequence, and combined with the pre-calibrated camera and laser system parameters, a projection mapping relationship between the image coordinate system and the laser plane coordinate system is established. Based on the principle of triangulation, the projection mapping relationship is transformed into spatial geometric constraints. The three-dimensional spatial coordinate set is determined by solving the intersection point of the laser plane equation and the projection ray equation emitted by the camera optical center. The three-dimensional spatial coordinate set is stitched together and processed with coordinate system unification to obtain an initial three-dimensional point cloud containing embossing depth information.

4. The method for three-dimensional reconstruction of embossing depth in special paper according to claim 3, characterized in that, The initial 3D point cloud undergoes data interpolation processing. Specifically, for curved corners and narrow textured areas, the point cloud density is increased to fill in data gaps, resulting in a complete 3D point cloud, including: Based on the initial 3D point cloud, the missing data areas of curved corners and narrow texture regions are identified through curvature analysis; Based on the geometric features of the identified missing data areas, the point cloud interpolation density and interpolation direction are adaptively determined to form interpolation parameters; Based on the interpolation parameters, a moving least squares surface fitting process is performed in the neighborhood of the missing data area. That is, a local reference plane is established with the missing area as the center, and the weight influence value of each neighboring point is allocated based on the spatial distance decay law to establish the weight distribution. Based on the weight distribution, the local surface is approximated by the least squares criterion to obtain the fitted surface; based on the fitted surface, new three-dimensional coordinate points are formed according to a preset density, thus obtaining the interpolated point cloud; The interpolated point cloud is spatially registered and fused with the initial 3D point cloud to form a complete 3D point cloud.

5. The method for three-dimensional reconstruction of embossing depth in special paper according to claim 4, characterized in that, Based on the complete 3D point cloud, three reference coordinate points are located at the intersection of the curvature extremum point of the embossed arc corner and the central axis of the narrow texture. A dynamic feature evaluation triangle is constructed based on these three reference coordinate points, including: Based on the complete 3D point cloud, the curvature distribution characteristics of the embossed arc corner area are analyzed to locate the extreme points of curvature change; Based on the complete 3D point cloud, the center direction feature line of the narrow texture region is extracted to determine the key intersection position of the center direction feature line and the arc-shaped corner contour. The located curvature extrema points and the determined key intersection points are combined into three spatial reference points to establish a set of reference coordinate points; Based on a set of reference coordinate points, a spatial triangulation method is used to connect three spatial reference points to form the smallest closed unit, thereby constructing a dynamic feature evaluation triangle that covers key feature regions.

6. The method for three-dimensional reconstruction of embossing depth in special paper according to claim 5, characterized in that, Based on the feature evaluation of the triangular facet's normal vector and area change rate, a multidimensional depth compensation coefficient is calculated. Based on this coefficient, the embossing depth, fiber protrusion height, and local curvature in the complete 3D point cloud are uniformly corrected to obtain calibrated 3D depth data, including: The spatial pose characteristics of the triangular facets are obtained by calculating the unit normal vector of the triangular facets. By analyzing the features, the area change rate of the triangular facets during continuous acquisition cycles is evaluated, and the dynamic deformation characteristics of the embossed surface are obtained. Based on spatial attitude features and dynamic deformation features, a multidimensional depth compensation coefficient is obtained through weighted fusion calculation. Based on the multidimensional depth compensation coefficient, the embossing depth, fiber protrusion height and local curvature in the complete 3D point cloud are collaboratively corrected to form calibrated 3D depth data.

7. The method for three-dimensional reconstruction of embossing depth in special paper according to claim 6, characterized in that, Real-time conformity assessment is performed on the calibrated 3D depth data. Based on the assessment results, the cycle time requirements of the production line are matched to complete the automated quality inspection process, including: Feature analysis is performed on the calibrated 3D depth data to extract embossing depth distribution parameters and key geometric features; The embossing depth distribution parameters are compared point by point with the preset process specification threshold to identify abnormal areas that exceed the tolerance range; Based on the identified abnormal regions, the spatial boundaries of the abnormal regions are determined by performing minimum bounding box calculations, and the corresponding quality judgment results are obtained based on the geometric dimensions and positions of the spatial boundaries. The quality judgment results are converted into production line control signals. Based on the production line control signals, automatic sorting operations are achieved by adjusting the conveyor cycle.

8. A three-dimensional reconstruction system for the embossing depth of special paper, wherein the system implements the method as described in any one of claims 1 to 7, characterized in that, include: The acquisition module is used to acquire the original image data of the special paper surface; the original image data includes the embossed structure, the non-uniform diffuse reflection interference caused by the matte coating, and the information on the plant fiber protrusions; The processing module is used to preprocess the original image data. By suppressing the non-uniform diffuse reflection interference of the matte coating and enhancing the contrast between the embossed area and the fiber protrusion area, optimized image data is obtained. This includes: performing frequency domain filtering on the original image data sequence to suppress the non-uniform diffuse reflection interference in a specific frequency band caused by the matte coating, resulting in filtered image data; applying multi-scale morphological operations to the filtered image data to enhance the gray-level contrast features between the embossed contour and the fiber protrusion area through structuring element matching, resulting in an enhanced contrast image; and normalizing and denoising the enhanced contrast image to output optimized image data suitable for 3D reconstruction calculations. The calculation module is used to obtain an initial 3D point cloud by performing 3D reconstruction calculations based on optimized image data. The initial 3D point cloud contains embossing depth information. The initial 3D point cloud is then subjected to data interpolation processing, that is, for curved corners and narrow texture areas, the point cloud density is increased to fill the data gaps and obtain a complete 3D point cloud. The module is used to locate three reference coordinate points at the intersection of the curvature extremum point of the embossed arc corner and the central axis of the narrow texture, based on the complete 3D point cloud; and to construct a dynamic feature evaluation triangle based on the three reference coordinate points. The calibration module is used to evaluate the normal vector and area change rate of the triangular face based on the features, and calculate a multi-dimensional depth compensation coefficient. Based on the multi-dimensional depth compensation coefficient, the embossing depth, fiber protrusion height and local curvature in the complete 3D point cloud are corrected in an integrated manner to obtain the calibrated 3D depth data. The execution module is used to perform real-time qualification judgment on the calibrated 3D depth data. Based on the judgment result, it matches the cycle time requirements of the production line to complete the automated quality inspection process.

Images