Braille printing method and system combined with 3D relief

Abstract

The invention relates to computer aided design, and discloses a Braille printing method and system combined with 3D relief, and the method comprises the steps: carrying out the connection and fusion of a concave-convex topographic map and a multi-level depth map corresponding to each Braille dot matrix, and generating a dual-channel depth map as a printing pattern; controlling the 3D printing system to spray resin on the printing substrate based on the printing pattern read by the light field modulator, and scanning the surface of the resin by laser aiming at the braille dot matrix area to form a concave-convex structure corresponding to the concave-convex topographic map; on the adjacent side of the braille dot matrix area, a multi-wavelength DLP projection system is used for executing layered exposure, the embossment structure represented by the multi-level depth map is projected to the resin layer, and an embossment model is formed; and performing global curing on the whole printed piece based on ultraviolet light to form an integrated finished product of braille alphabets and 3D embossments. The invention further discloses a computer readable storage medium. The method aims at efficiently manufacturing seamless and high-precision multi-layer 3D embossment integrated finished products of braille alphabets and related semantics.

Description

Technical Field

[0001] This application relates to the field of computer-aided design, and more particularly to a Braille printing method, a 3D printing system, and a computer-readable storage medium that incorporates 3D relief printing. Background Technology

[0002] Braille, as the primary medium for visually impaired individuals to access written information, is primarily produced using techniques that revolve around the raised dots of the Braille dot matrix. Traditional Braille products contain only discrete dot matrix information, resulting in a limited range of representations. For complex visual content such as graphics, charts, maps, artworks, or scientific models that contain rich spatial, textural, and hierarchical information, visually impaired users struggle to understand and perceive the complete semantic and aesthetic information conveyed by the graphics, leading to information silos.

[0003] Existing technologies for creating Braille and graphic reliefs on the same substrate typically employ a step-by-step processing, splicing, or pasting process. For example, a Braille board is first printed, and then a rough relief is created in adjacent areas using another set of equipment or by hand. This process is not only cumbersome and inefficient, but more importantly, it often results in noticeable physical breaks or seams between the Braille and graphic areas, disrupting the overall smoothness and consistency of tactile exploration and impacting the user experience.

[0004] The above content is only used to help understand the technical solution of this application and does not represent an admission that the above content is prior art. Summary of the Invention

[0005] The main purpose of this application is to provide a Braille printing method, a 3D printing system and a computer-readable storage medium that combine 3D relief printing, aiming to efficiently manufacture seamless, high-precision Braille and multi-layered 3D relief integrated products with associated semantics, so as to ensure the overall smoothness and consistency of the user's tactile exploration and improve the user experience.

[0006] To achieve the above objectives, this application provides a Braille printing method combining 3D relief, comprising the following steps: The Braille dot matrix is ​​converted into a concave-convex topographic map, and the semantic graphics associated with the Braille dot matrix are used to generate a multi-level depth map through neural style transfer. Using an adaptive fusion algorithm, the concave and convex topographic maps and multi-level depth maps corresponding to each Braille dot matrix are connected and fused to generate a dual-channel depth map as a printing pattern; The printed pattern is input into the light field modulator of the 3D printing system; the 3D printing system also includes a laser and a multi-wavelength DLP (Digital Light Processing) projection system; The 3D printing system controls the printing pattern read by the light field modulator, sprays resin onto the printing substrate, and uses a laser to scan the resin surface for the Braille dot matrix area to form a concave-convex structure corresponding to the concave-convex topographic map; and, on the side adjacent to the Braille dot matrix area, a multi-wavelength DLP projection system is used to perform layered exposure to project the relief structure represented by multi-level depth maps onto the resin layer to form a relief shape; wherein, the multi-wavelength DLP projection system uses light of different wavelengths to control the curing depth of the resin for the same relief structure. The printed parts are globally cured using ultraviolet light, forming a single finished product that integrates Braille and 3D relief.

[0007] To achieve the above objectives, this application also provides a 3D printing system. The main control device of the 3D printing system includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the steps of the Braille printing method combined with 3D relief as described above.

[0008] To achieve the above objectives, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the Braille printing method combined with 3D relief as described above.

[0009] This application provides a method, system, and computer-readable storage medium for printing Braille using 3D relief, enabling the production of seamless, high-precision, multi-layered 3D relief integrated products with associated semantics within an automated 3D printing process. This not only improves production efficiency but also ensures the overall smoothness and consistency of the user's tactile exploration, enhancing the user experience. Furthermore, by generating multi-level depth maps through neural style transfer and controlling the curing depth with multi-wavelength light, it can precisely reproduce the complex spatial layers and textures from shallow to deep undulations. This provides an efficient and precise production foundation for realizing personalized, highly expressive tactile content. Attached Figure Description

[0010] Figure 1 This is a schematic diagram of the steps of a Braille printing method incorporating 3D relief in one embodiment of this application; Figure 2 This is a schematic diagram of the internal architecture of the main control device of a 3D printing system according to an embodiment of this application.

[0011] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0012] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0013] Furthermore, descriptions using terms such as "first" and "second" in this application are for descriptive purposes only (e.g., to distinguish identical or similar features) and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, features defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, technical solutions from different embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. If a combination of technical solutions is contradictory or impossible to implement, such a combination should be considered nonexistent and not within the scope of protection claimed in this application.

[0014] Reference Figure 1 In one embodiment, the Braille printing method incorporating 3D relief includes: Step S10: Convert the Braille dot matrix into a concave-convex topographic map, and generate a multi-level depth map by using neural style transfer to associate the semantic graphics of the Braille dot matrix. Step S20: Using an adaptive fusion algorithm, the concave and convex topographic maps and multi-level depth maps corresponding to each Braille dot matrix are connected and fused to generate a dual-channel depth map as a printing pattern. Step S30: Input the printing pattern into the light field modulator of the 3D printing system; wherein, the 3D printing system also includes a laser and a multi-wavelength DLP projection system; Step S40: The 3D printing system, based on the printing pattern read by the light field modulator, sprays resin onto the printing substrate, and uses a laser to scan the resin surface for the Braille dot matrix area to form a raised structure corresponding to the raised topographic map; and, on the side adjacent to the Braille dot matrix area, a multi-wavelength DLP projection system is used to perform layered exposure to project the relief structure represented by the multi-level depth map onto the resin layer to form a relief shape; wherein, the multi-wavelength DLP projection system uses different wavelengths of light to control the curing depth of the resin for the same relief structure. Step S50: Globally solidify the printed part using ultraviolet light to form an integrated finished product with Braille and 3D relief.

[0015] In this embodiment, the execution terminal can be a 3D printing system, or other equipment or devices (such as control devices) that control the 3D printing system.

[0016] As described in step S10, input standard Braille dot matrix data: standard Braille encoding (such as Unicode Braille blocks U+2800 to U+28FF), or a dot sequence encoded according to national / language standards (such as Chinese double-pinyin Braille, English level 2 Braille).

[0017] Among them, the Braille dot matrix follows the international or industry standard Braille physical size specifications, and the relevant parameters include: dot base diameter (1.5~2 mm), dot height (0.5~1.5 mm), dot spacing (the distance between the center of the dots in the horizontal and vertical directions, which can be 2~3 mm), and cell spacing (the distance between characters).

[0018] The Braille conversion algorithm first parses the Braille code to determine which dots in each character (6-dot or 8-dot cell) are "raised". Then, for each "raised" dot, a standardized microscopic 3D model (such as a hemisphere) is generated at its coordinate center. Next, all the 3D models are precisely arranged on a 2D plane according to the row and column layout of the Braille cells. Finally, this 3D model set is rasterized or voxelized to calculate a 2D height field map, i.e., a topographic map.

[0019] Optionally, the topographic map can be a grayscale image, where the grayscale value of each pixel represents the height (Z value) of that point. Non-point areas are black (0, the base plane), and convex points are white or have high grayscale values ​​(e.g., 255, representing the standard point height). This topographic map is binarized or has extremely low grayscale, with drastic and discontinuous height variations, presenting a regular dot matrix pattern at a microscale.

[0020] Simultaneously, semantic information corresponding to the Braille (such as concepts like "trees" and "animals") is extracted to generate associated 2D vector or raster graphics. A neural style transfer network (such as VGG-19+Gram matrix loss) combined with a depth estimation model is used to transform the semantic graphics into multi-level depth maps with an artistic style.

[0021] The multi-level depth map includes: (1) Shallow details: texture, outline (e.g., depth can be selected from 0.1 to 0.5 mm); (2) Intermediate layer transition: main structure (e.g., depth can be selected from 0.5 to 1 mm); (3) Deep background: Three-dimensional scene (e.g., depth can be selected from 1 to 2 mm).

[0022] Optionally, input any digital image (such as JPG or PNG) that is semantically related to the Braille content. For example, the Braille word "elephant" corresponds to a side profile of an elephant.

[0023] Optionally, a pre-trained deep neural network can be used to fuse the content of the input image with a specific art style (e.g., woodcut, stone relief, or line drawing). The output image will enhance edges and contours, simplify or texture internal areas, and reduce complex color gradations and details. This makes subsequent depth estimation more stable, resulting in clearer and more aesthetically pleasing relief lines when printed.

[0024] Optionally, a depth estimation deep learning model is used to infer the relative distance of the object portion represented by each pixel from the (potentially stylized) image to the observer (relief surface), and the continuous depth is discretized into a finite, printable hierarchy.

[0025] Alternatively, the deep learning model for depth estimation can be MiDaS, DPT, MegaDepth, etc. These models are pre-trained on millions of images with depth labels and have learned to predict patterns of depth information from a single image.

[0026] Optionally, the image can be input into a depth estimation network, which outputs a depth map with continuous values. The value of each pixel in the map is a floating-point number representing the relative depth (larger / whiter foreground values, smaller / blacker for background values).

[0027] Optionally, the original depth value range output by the network can be linearly mapped to a preset physical height range (e.g., 0~2mm) for normalization. The normalized depth values ​​are then divided by the depth step size for each level and rounded down. For example, if the total target height is 2mm and divided into 16 levels, the step size for each level is 0.125mm. All depth values ​​will be "snap-to" to the nearest multiple of 0.125mm.

[0028] The final result is an index map that serves as a multi-level depth map, where the value of each pixel is an integer between 1 and N, representing which depth level it belongs to.

[0029] Optionally, further selective post-processing operations such as structural optimization and detail enhancement can be performed on the generated multi-level depth maps: smoothing depth values ​​in small local areas to eliminate noise and jitter that may be generated by network prediction, making the relief surface smooth; non-linear adjustment of the depth range (such as gamma correction), for example, increasing the depth difference of the foreground to highlight the subject, and compressing the depth difference of the background to save printing height and material; sharpening the outline edges of the main objects to ensure clear boundaries; setting a base plane to ensure that the depth of the background area of ​​the image is 0 (base plane), and all objects "grow" from the base.

[0030] As described in step S20, an adaptive fusion algorithm (such as weighted image fusion) is used to spatially align the convex topographic map with the multi-level depth map (with Braille dots located in the center or a designated area, and semantic reliefs located on the adjacent side).

[0031] Optionally, the adaptive fusion algorithm marks the precise location of the Braille dot matrix using a binary mask, and generates a transition band on the adjacent side of the Braille area. The adjustable width of the transition band is a preset distance w. t This is used for smooth transitions. An improved Poisson fusion algorithm is employed, using a topographic map of concavity and convexity as a reference, to fuse multi-level depth maps into the transition zone and surrounding areas.

[0032] In the Braille region, the original gradient is preserved to ensure the integrity of the convex structure; within the transition zone, the gradients of the multi-layer depth maps are weighted and blended. ; Wherein, H1 is the fusion gradient field, which is the spatial gradient vector of the fused height field and is ultimately used for Poisson reconstruction; H2 is the Braille gradient field, which is the height gradient of the Braille embossed topographic map and can be obtained by calculating the gradient field of the embossed topographic map; H3 is the relief gradient field, which is the gradient of the relief multi-level depth map and can be obtained by calculating the gradient of the multi-level depth map; α is the hybrid weight coefficient, which refers to the proportion of the Braille gradient in the fusion, and its value ranges from [0, 1], decreasing linearly with distance from the Braille boundary.

[0033] Solving the Poisson equation This yields a globally continuous global height field H4.

[0034] Optionally, to ensure accurate correspondence between layers during multi-wavelength DLP exposure, the depth of each relief layer is linearly mapped to the actual physical height based on the total height range of the global height field.

[0035] Optionally, during the dual-channel depth map generation process, channel one is used to generate an absolute height map (for laser scanning), such as a single-channel grayscale image; to adapt to laser point-by-point scanning, the height values ​​of the blind spot area are converted into a scanning path sequence, including: center coordinates, scanning radius (determined by the diameter of the protrusion), and layered scanning height parameters.

[0036] Optionally, channel 2 is used to generate a layered exposure parameter map (for multi-wavelength DLP), i.e., a multi-channel (such as RGB) image, with each channel encoding different exposure parameters: R channel (records short wavelength exposure time to control shallow curing), G channel (records medium wavelength exposure time to control medium curing), and B channel (records long wavelength exposure time to control deep curing).

[0037] The parameter mapping rule calculates the RGB value of each pixel based on its height in the global height field and its layer. For example, if a pixel's height belongs to the second layer of the relief, its G channel value will be higher, while its R and B channel values ​​will be lower.

[0038] The final output is a dual-channel depth map containing two information channels as the printing pattern: Channel 1 (Braille channel) mainly carries the precise concave and convex information of the Braille dot matrix; Channel 2 (Embossing channel) carries the depth information of multi-level embossing.

[0039] As described in step S30, the 3D printing system includes a light field modulator (such as a spatial light modulator (SLM) such as a digital micromirror device (DMD) or liquid crystal on silicon (LCoS)), a laser (such as an ultraviolet laser), and a multi-wavelength DLP projection system (containing multiple light sources of different wavelengths, such as between 385nm and 520nm). The dual-channel depth map is loaded into the light field modulator, and the system allocates control signals to the laser and the DLP projection system according to the map data.

[0040] The light field modulator can function as a dynamic mask or an array of optical switches. Based on the data from the input image, it controls the "on" and "off" of the optical path pixel by pixel, or modulates the phase / amplitude of the light, thereby generating the desired light pattern on the resin surface. In other words, the light field modulator controls the on / off state of the corresponding micromirrors (for Braille) or modulates the intensity / mode of the light (for embossing) based on the pixel value corresponding to each printed pattern, thus "drawing" the corresponding cross-sectional pattern on the photosensitive resin.

[0041] The parsed and preprocessed dual-channel data is loaded into the frame buffer of the optical field modulator. Under system control, the optical field modulator can switch rapidly between two operating modes: a) Laser Modulation Mode: When the system needs to perform laser scanning in step S40, the modulator generates a dynamic mask corresponding to the Braille dot matrix based on the data from channel one. This mask determines that the laser beam can only pass through the micromirror area corresponding to the "dot matrix", thereby accurately scanning the concave and convex structure on the resin surface.

[0042] b) DLP Projection Modulation Mode: When the system needs to perform layered exposure, the modulator sequentially calls the depth data of different levels in Channel 2 and converts them into the projection pattern required by the multi-wavelength DLP projection system. Different patterns correspond to different wavelengths of exposure to achieve curing at different depths.

[0043] The light field modulator is located at the core of the entire optical path system. Light from the laser and multi-wavelength DLP light source is guided to the surface of the modulator through a specific optical system (such as a beam splitter and lens group). The light "encoded" by the modulator (i.e., the light carrying the printed pattern information) is then precisely focused or projected onto the resin layer surface of the printed substrate through a subsequent projection lens.

[0044] In addition, the 3D printing system also includes a resin coating system (which can spray or spread a photosensitive resin layer of a predetermined thickness) and a corresponding motion control platform.

[0045] As described in step S40, by coordinating two different additive manufacturing processes, laser scanning and multi-wavelength DLP projection, Braille embossed structures and multi-level relief shapes are simultaneously formed on the resin layer, achieving integrated manufacturing.

[0046] First, resin spraying and substrate preparation are carried out.

[0047] Optionally, the resin material can be a UV-curable resin (photopolymer), such as acrylate, epoxy resin, or a custom formulation. A liquid resin film of a predetermined thickness is formed on a printing substrate (such as a glass platform, the previous cured layer) using a precision spray head.

[0048] Optional Braille dot matrix area processing and laser scanning curing: The laser used can be an ultraviolet laser (such as 355nm or 405nm), which has high beam quality and a small focal point (down to the micrometer level).

[0049] The channel-one data (undulating topographic map) resolved by the optical field modulator is used by a galvanometer system to control the high-speed deflection of the laser beam in the XY plane, scanning point by point along a preset path. Based on the coordinate information of the undulating topographic map, the laser beam selectively scans the resin surface. During the scanning process, the laser energy is absorbed by the resin, triggering a photopolymerization reaction that instantly solidifies the liquid resin at the scanning point into a solid.

[0050] Optionally, by controlling the laser power, scanning speed, and dwell time, standard-compliant hemispherical bumps can be formed. In the areas not irradiated by the laser, the resin remains liquid, preparing for subsequent DLP exposure or the next layer printing.

[0051] Optional features include embossed area processing and multi-wavelength DLP layered exposure: The multi-wavelength DLP projection system integrates multiple LED or laser light sources of different wavelengths (e.g., 405nm violet light, 455nm blue light, 520nm green light); and has a built-in DLP chip (digital micromirror device) for generating two-dimensional light patterns.

[0052] Optionally, an optical synthesis and projection system combines the light paths of different wavelengths of light sources and precisely projects them onto the resin surface. Different wavelengths of light have different penetration depths in the resin (Beer-Lambert law). Short-wavelength light is mainly absorbed by the surface layer, curing the shallow layer; long-wavelength light penetrates deeper, curing the lower layer.

[0053] Optionally, the light field modulator sequentially outputs two-dimensional patterns corresponding to different depth layers in the channel two data (multi-level depth map). The system performs exposures in order from shallow to deep: First layer (shallow texture): Using a short-wavelength light source, DLP projects the first layer pattern (such as fine lines and textures). The light is strongly absorbed on the resin surface, resulting in a shallow curing depth.

[0054] Second layer (intermediate relief): Switch to a medium wavelength light source, and DLP projects the second layer pattern (such as the main outline). The light penetrates the cured surface layer, initiating polymerization in the uncured resin below, forming a medium-depth structure.

[0055] The third layer and above (deep structure): using long-wavelength light sources to project deeper patterns, forming high-depth features such as the base or background of a relief sculpture.

[0056] Optionally, exposure of all wavelengths can be performed sequentially within the same resin layer, rather than through physical layer stacking. By controlling the exposure intensity and duration of each wavelength, a continuously varying curing gradient can be created along the depth direction (Z-axis) of the resin layer, thereby generating complex relief surfaces with smooth transitions in a single step.

[0057] Optionally, during DLP exposure, the Braille dot matrix area can be protected by a digital mask (controlled by a light field modulator) or physical shielding to prevent the relief exposure from affecting the solidified precise Braille dot structure. At the boundary between the Braille and the relief, DLP exposure can use a gradient grayscale pattern or a special wavelength combination to allow the relief structure to transition naturally to the base of the Braille dot matrix, ensuring tactile and visual continuity.

[0058] Optionally, laser scanning and DLP projection may be performed alternately in time or in separate spatial sections, coordinated by a unified control system to avoid mutual interference. For example, the entire layer may be laser-scanned and the Braille dots solidified first, followed by DLP layer-by-layer exposure within the same liquid resin area to form an embossed design.

[0059] As described in step S50, the aim is to thoroughly cure the photosensitive resin through global exposure with ultraviolet light, thereby enhancing the final mechanical properties of the printed part, completing the manufacturing process, and obtaining a ready-to-use integrated finished product.

[0060] In step S40, the resin is selectively cured in localized areas or at specific depths through laser scanning and multi-wavelength DLP layered exposure, forming a Braille dot matrix and embossed structure. Then, the entire printed part is uniformly exposed to ultraviolet light to ensure that any incompletely cured resin is thoroughly hardened, improving structural strength and stability.

[0061] The ultraviolet light source can be a high-power, broad-spectrum ultraviolet LED array or a mercury lamp. The wavelength range covers UV-A (315~400nm) and can be extended to UV~V (395-445nm) to ensure full activation of the residual photoinitiator in the resin.

[0062] In this way, the Braille dot matrix and 3D relief are formed in one piece, without any splicing marks. It not only meets the tactile reading needs of blind people, but also provides visual semantic supplement through the relief (such as patterns that can be recognized by sighted people).

[0063] In one embodiment, a seamless, high-precision, multi-layered 3D relief sculpture with associated semantics is manufactured in an automated 3D printing process. This not only improves production efficiency but also ensures the overall smoothness and consistency of the user's tactile exploration, enhancing the user experience. Furthermore, by generating multi-level depth maps through neural style transfer and using multi-wavelength light to control the curing depth, it can precisely reproduce the complex spatial layers and textures from shallow to deep undulations. This provides an efficient and precise production foundation for realizing personalized, highly expressive tactile content.

[0064] In one embodiment, based on the above embodiments, the Braille printing method combined with 3D relief further includes: When connecting and fusing the topographic maps and multi-level depth maps corresponding to each Braille dot matrix, the starting point is the end of the topographic map corresponding to each Braille dot matrix, and the corresponding multi-level depth maps are connected and fused at a preset distance. Specifically, a curvature algorithm is used to calculate the first curvature continuity between the final height of the convex-concave topographic map and the edge region layer height of the multi-level depth map, and the edge region layer height is adjusted based on the first curvature continuity with the goal of curvature smoothing.

[0065] In this embodiment, the starting point is the outer edge of the topographic map corresponding to each Braille dot matrix. The "outer edge" refers to the outer boundary of the Braille dot structure. For example, a Braille dot can be regarded as a tiny cylinder or dome, and its "outer edge" is the circular outline where the sidewall of the dot meets the base plane. This outline is defined as the spatial starting line of the fusion operation.

[0066] Starting from this initial line, define a transition zone of fixed width outwards (towards the relief area). The width of the transition zone is adjustable to a preset distance w. t The value can range from 0.5 to 2 cm. That is, at a preset distance from the end of the topographic map, the corresponding multi-level depth maps are connected.

[0067] Optionally, the highest point at the end of the topographic map (marked as point A) is connected to the highest point at the edge of the multi-level depth map at a preset distance (marked as point B), and the highest point at the edge of the multi-level depth map is connected to the highest point at the innermost edge of the edge region (marked as point C). Then, based on the height of each point of the continuous irregular line segment (ABC) formed by these points, parametric spline interpolation is used to calculate the curvature continuity, and it is marked as the first curve continuity.

[0068] Optionally, a cubic spline curve that guarantees the continuity of the second derivative is used to fit the three points. This process automatically generates a continuous mathematical curve that not only passes through the three points but is also smooth (without sharp corners) throughout its domain. Based on the smooth spline curve generated in the previous step, a series of points are sampled at high density along the curve length. For each sampled point, the precise curvature value at that point is calculated by substituting the first derivative (tangent direction) and second derivative (rate of change in the tangent direction) of the curve into the definition of curvature, thus obtaining a curvature distribution curve from point A to point C.

[0069] Optional, curvature continuity (C 2 Continuity refers to the smooth, gradual change in the "curvature" of a curve along its entire length, rather than abrupt jumps. This can be achieved by calculating the curvature values ​​at densely sampled points on the curve using algorithms and analyzing the smoothness of these curvature changes, quantifying the smoothness of this curvature change into a numerical index. Specifically, this is quantitatively assessed by examining whether the values ​​of the curvature distribution curve transition smoothly on both sides of key points (especially point B) and by calculating the smoothness of change along the entire curvature distribution curve (e.g., calculating the sum of squares of the curvature differences between adjacent sampled points).

[0070] After obtaining the continuity of the first curve, the layer height of the relief edge region is mapped to the height of the geometric control points. Specifically, let the height HB of point B correspond to the layer height of the outermost edge of the relief; let the height HC of point C correspond to the layer height of the first control point on the inner side of the relief edge. The parameter vector is denoted as P = [HB, HC].

[0071] Set initial state and constraints: The initial value P0 can be set to the default height extracted from the original depth map.

[0072] Given a set of parameters P i = [HB i HC i Update the 3D coordinates of points B and C (the X-coordinate can usually be fixed or adjusted accordingly). Use this to ensure C... 2 The continuous spline curve algorithm refits the smooth curve L passing through points A, B, and C. i For curve L i Dense sampling is performed to calculate its curvature distribution, and its first curvature continuity parameter value E is obtained. i (The smaller the value, the smoother the curvature).

[0073] Among them, E i The calculation can be the integral of the rate of change of curvature or the sum of squared discrete differences.

[0074] Optionally, an optimization algorithm can be used to systematically adjust P to minimize the objective function E: if the gradient of E with respect to HB and HC can be obtained (or approximated by the finite difference method), the parameters can be adjusted in the opposite direction of the gradient.

[0075] Example of a single iteration process: (1) Current state: parameter P i The corresponding continuity error E i .

[0076] (2) Trial adjustment: Generate a new set of candidate parameters P according to the optimization algorithm rules. i '.

[0077] (3) Evaluate candidate points: Perform step (2) and calculate P. i 'Corresponding new error E i '.

[0078] (4) Decision-making and updating: Comparison E i With E i ', if E i ' < E i If so, then accept this adjustment, and let P i+1 = P i Otherwise, reject or process according to the algorithm rules.

[0079] (5) Termination conditions: Optimization stops when any of the following conditions are met: the improvement of E is less than the preset threshold after multiple consecutive iterations; the preset maximum number of iterations is reached; or the value of E drops below the acceptable smoothness threshold.

[0080] After optimization, the parameter P' = [HB', HC'] that minimizes E is output.

[0081] Optionally, based on HB' and HC', update the layer height values ​​at corresponding positions in the multi-level depth map of the relief edge region. The layer heights at each position within the edge region can be adjusted according to the ratio of the difference between the old and new layer heights.

[0082] In this embodiment, by fine-tuning the edge layer height, the geometry at the junction is dynamically reshaped, ultimately optimizing the curvature continuity parameters calculated by a precise algorithm. This results in a seamless tactile experience when the fingertip glides across the manufactured physical model, improving the user's touch experience and making it particularly user-friendly for visually impaired users.

[0083] In one embodiment, based on the above embodiments, the Braille printing method combined with 3D relief further includes: After adjusting the edge region layer height based on the first curvature continuity, the curvature algorithm is used to calculate the second curvature continuity between the edge region layer height and the core region layer height of the multi-level depth map, and the core region layer height is adjusted based on the second curvature continuity with curvature smoothing as the goal.

[0084] In this embodiment, based on the fixed optimized edge region, the second curvature continuity between the edge region and the relief core region is optimized. Point D (or multiple points) is defined to represent typical high points or feature points of the relief core region, such as the center point. At this time, the innermost high point (point C) of the edge region and point D are connected so that the tactile profile line extends from BCD.

[0085] Then, referring to the calculation process of the first curvature continuity corresponding to line segment ABC, calculate the second curvature continuity corresponding to line segment BCD.

[0086] Then, referring to the above process of adjusting the floor height of the edge area, with the goal of smooth curvature, the floor height of the core area is adjusted according to the continuity of the second curvature.

[0087] In this way, the local transition optimization (first continuity) is sequentially coupled with the overall form coordination (second continuity), ensuring that the curvature of the complete tactile path from the Braille dots to the relief core is seamlessly transitioned, reducing the "discontinuity" or "abrupt steps" in tactile perception, and further improving the user experience.

[0088] In one embodiment, based on the above embodiments, the step of generating a multi-level depth map from the semantic graph associated with the Braille dot matrix through neural style transfer includes: The semantic graphics associated with Braille dots are input into a pre-trained model so that the pre-trained model can generate multi-level depth maps through neural style transfer; wherein, the pre-trained model learns and trains the association relationship between semantic graphics and multi-level depth maps in advance.

[0089] In this embodiment, the pre-trained model can intelligently transform any input semantic graph (two-dimensional image) into a multi-level depth map with a clear, discrete physical depth hierarchy.

[0090] Optionally, the model input and output are defined as follows: The input (semantic graph) is a standard two-dimensional RGB or grayscale image; the output is a multi-channel (e.g., 16-channel) two-dimensional image, or a single-channel grayscale image containing discrete height values ​​(where each grayscale value corresponds to a specific physical layer height). The value of each pixel in the image no longer represents color, but rather the relative height of that point on the final relief or the physical layer to which it belongs. For example, a value "0" represents the base (lowest point), and a value "15" represents the highest point of the relief.

[0091] Semantic consistency of the generated depth map: The depth distribution must conform to the physical structure of the graphic content (e.g., the foreground / main part of an object should be higher than the background; the core outline should have higher prominence).

[0092] While generating depth information, the model has performed neural style transfer processing on the input graphics, making its lines, textures, and light and shadow relationships present the target art style (such as the rough knife marks of wood carving and the mottled texture of stone carving), and this stylization effect is directly reflected in the pattern of depth changes.

[0093] Optionally, the depth levels are discrete and limited (e.g., 16 layers) to accommodate the process limitations of 3D printing; the transitions between different depth regions are smooth, avoiding steep cliffs or isolated slender columns that cannot be printed.

[0094] Optionally, the core architecture of the pre-trained model is an end-to-end cascaded deep learning architecture, which integrates two core modules: a neural style transfer and feature enhancement module, and a single-image depth estimation and hierarchical discretization module.

[0095] Optionally, a neural style transfer and feature enhancement module is used to receive the original semantic graphics and perform artistic transformation and structural enhancement on them. This module can use a style transfer network (such as AdaIN) based on feature extractors like VGGNet. The model has pre-learned the essential features of various art styles (wood carving, stone carving, line carving, etc.) and can fuse the content of the input graphics with the texture and brushstroke features of the target style according to a preset or specified style.

[0096] Optionally, after style transfer, an intermediate image with a stylized and structurally clear design can be generated by enhancing the subject outline and weakening irrelevant details using an edge detection algorithm or a lightweight CNN.

[0097] Optionally, a single-image depth estimation and hierarchical discretization module is provided to predict the depth value of each pixel from the stylized image and quantize it to a finite number of printing levels. This module employs an encoder-decoder structured depth estimation network (such as variants of MiDaS or Depth Anything). The encoder part of the network extracts high-level semantic and geometric features from the stylized image, while the decoder part progressively upsamples to predict a continuous, relative depth map. This network has been pre-trained on massive monocular image-depth data pairs and has learned to infer universal rules of 3D geometry from 2D visual cues (such as perspective, occlusion, and texture gradation).

[0098] Optionally, the predicted continuous depth values ​​are mapped to N predefined discrete levels (e.g., 0-15) through a learnable quantization layer. The quantized depth map is then fine-tuned using a conditional random field or a lightweight convolutional network to eliminate noise, fill small holes, and ensure that depth boundaries are aligned with stylized contours in the image.

[0099] The joint training objective of the entire pre-trained model is to perform end-to-end supervised learning using massive amounts of semantic graphs, artistic style labels, and ideal multi-level deep graph triples.

[0100] The loss functions include content loss (ensuring the structure of the output depth map is consistent with the content of the input graphic), style loss (ensuring the pattern of depth changes (such as which parts should be convex and which parts should be concave) reflects the characteristics of the target artistic style), depth regression loss (the error between predicted depth and true depth (such as L1 loss)), hierarchy consistency loss (encouraging depth values ​​to clearly cluster at discrete hierarchy centers), and smoothness loss (maintaining smooth depth changes in non-edge regions).

[0101] Through this training, the model does not simply "draw" depth, but truly learns the complex mapping relationship between artistic style, visual semantics, and three-dimensional relief geometry.

[0102] The pre-trained model, once trained, is used to receive semantic graphics provided by upstream modules (either manually designed or semantic understanding modules). Through this pre-trained model, two key tasks—"artistic stylization" and "depth information generation"—are completed in one step, outputting multi-level depth maps that can be directly used for 3D printing data fusion.

[0103] In this way, the model learns in advance the mapping relationship between semantic graphics and multi-level depth maps, enabling it to generate complex, layered relief patterns.

[0104] In an alternative embodiment, the system can also integrate an end-to-end semantic understanding module, capable of directly inputting Braille codes or corresponding text semantics, and automatically completing the entire conversion from semantics to graphics, and then to depth maps. This could be achieved by integrating a text-to-image generation model based on large-scale image-text pair pre-training (such as a lightweight and controllable version of Stable Diffusion, or a dedicated icon / sketching generation model).

[0105] Thus, the system workflow example is as follows: Input the Braille text (e.g., "elephant"). Based on the text description, the model generates a concise and clear semantic graphic (e.g., a silhouette of an elephant). Constraints can be added to this process to ensure the graphic style is simple and the outline is clear, suitable for subsequent embossing conversion. This generated semantic graphic is then directly used as input to the pre-trained model (neural style transfer generates multi-level depth maps).

[0106] Alternatively, the semantic graph generator can be further integrated with the pre-trained model to form a larger end-to-end network.

[0107] In this way, by integrating a text / semantic understanding front-end, the entire process from Braille meaning to relief blueprint is fully automated, greatly improving the system's ease of use and intelligence, and making it suitable for large-scale, standardized, or rapid prototyping applications.

[0108] In another alternative embodiment, based on the pre-trained model described above, a three-layer encoder architecture of "semantic-structural-haptic" can also be constructed: (1) Semantic encoder: used to understand the abstract concepts and component composition (torso, nose, ears) of the input image (e.g., "elephant"), extract high-level semantic segmentation maps (high-dimensional feature tensors that retain contextual information of different semantic regions (e.g., "sky", "trees", "animal body") and attention heatmaps (generated by a lightweight attention branch that identifies which regions in the image are most critical for tactile recognition (e.g., facial features, object contours), and these regions should be prioritized and kept high-fidelity when generating depth). It can be constructed using visual backbones (e.g., CLIP visual encoder, ResNet, ViT) pre-trained on large datasets (e.g., ImageNet, COCO).

[0109] (2) Structural Encoder: Learns general geometric priors (such as how to represent "thickness" and "hair") from a large amount of 3D model data, understands concepts such as "depth", "occlusion", and "surface continuity", and outputs geometric feature maps (feature tensors that encode the rough 3D layout of the scene (foreground / background hierarchy, general surface direction)) and surface normal estimates (which can provide preliminary surface direction information as a reference for the decoder). Large-scale monocular depth estimation datasets and 3D model datasets can be used. On datasets such as NYU Depth V2, a depth estimation network can be trained as the initialization of the structural encoder, which gives it the basic ability to "guess the depth from the image". From 3D model libraries such as ShapeNet, multiple viewpoint images of the same object and their corresponding depth maps and normal maps are rendered. The training goal is to encode consistent structural features of the same object from different viewpoints in the latent space, which strengthens its understanding of the essence of 3D shape, rather than the appearance of a single viewpoint.

[0110] (3) Tactile Encoder: From real relief data scanned by tactile sensors, learn the mapping relationship between surface texture (rough, smooth) and micro-geometry, and output tactile feature vectors / feature maps (an abstract representation that encodes "tactile texture". For example, a feature vector may point to "fine grain", "directional stripes" or "warm and smooth"). A 3D convolutional neural network or point cloud network can be used to directly process 3D mesh or voxel data, allowing the model to learn to distinguish between texture pairs that "feel similar" and "feel different". For example, two scans of tree bark from different angles are "similar", while scans of tree bark and smooth marble are "different". Also, predict the physical descriptors of the texture (such as roughness, hardness score, and friction coefficient estimate).

[0111] The outputs of the three encoders interact in a cross-attention fusion module, ultimately generating a multi-level depth map by the decoder that simultaneously satisfies semantic correctness, structural rationality, and tactile sensitivity. For example, the "semantic encoder" emphasizes that "the elephant's trunk should be prominent," the "structural encoder" provides representations of cylinders and wrinkles, and the "tactile encoder" suggests adding micro-textures to the surface to simulate skin texture. The workflow is as follows: (a) Query, Key, Value: The output of the semantic encoder is used as the dominant "query" because it represents "what we want to generate". The outputs of the structural encoder and the haptic encoder are used as the "key" and "value", respectively.

[0112] (b) Interaction process: The semantic query will "ask" the structure key: "In order to realize the semantics of this 'elephant trunk', what kind of spatial volume and surface structure should I have?" The structure value provides geometric suggestions accordingly. At the same time, the semantic query will also "ask" the tactile key: "What kind of microscopic texture should the semantics of this 'elephant skin' correspond to?" The tactile value provides texture suggestions accordingly.

[0113] (c) Fusion output: The outputs of the two attention mechanisms are concatenated or added together to form a unified conditional feature. This feature includes the intention of "making an elephant trunk", the geometric knowledge that "the nose should be cylindrical and wrinkled", and the tactile knowledge that "the skin should have a grainy texture".

[0114] Optionally, the output of the three-layer encoder is trained using a conditional decoder. The decoder can be a denoising network of a U-Net or a diffusion-like model, whose input is the conditional features output by the fusion module, and whose task is to progressively generate high-quality multi-level depth maps.

[0115] First, the semantic encoder is pre-trained on image segmentation tasks, the structural encoder on depth estimation and multi-view tasks, and the haptic encoder on haptic texture classification / contrast tasks. Then, end-to-end fine-tuning is performed using the final target dataset—high-quality paired data (images and haptic relief depth maps). The three encoders are frozen or slightly fine-tuned, and the fusion module and decoder are trained primarily.

[0116] The loss function is composite, including: L1 / L2 Loss (ensuring that the depth map is close to the real value at the pixel level), perceptual loss (using a pre-trained VGG network to compare differences at the feature level to ensure semantic consistency), adversarial loss (introducing a discriminator to make the generated depth map more closely resemble the real data in distribution, making it look more "natural" and "reasonable"), and physical constraint loss (differentiably calculating the printability (e.g., checking for steep slopes) and tactile safety (e.g., curvature smoothness) of the generated depth map as a regularization term).

[0117] Based on this three-layer encoder architecture, the generated results can be indirectly controlled by adjusting the input (e.g., emphasizing a specific region in the semantic segmentation map) or by intervening in the outputs of different encoders. Prior knowledge from the three domains of visual semantics, 3D geometry, and tactile physics is explicitly injected into the model through pre-training, transforming it from a black box into an interpretable and guideable expert system. The resulting relief depth map is semantically faithful to the original image, structurally reasonable and conforms to 3D common sense, and optimized for tactile feedback, possessing good recognizability and comfort.

[0118] In one embodiment, based on the above embodiments, the Braille printing method combined with 3D relief further includes: After adjusting the layer height of the multi-level depth map, iterative samples are generated based on the multi-level depth maps before and after the layer height adjustment and the corresponding semantic graphs. The iterative samples are then input into the pre-trained model for iterative training to update the pre-trained model.

[0119] In this embodiment, a self-optimizing closed-loop learning mechanism is introduced into the original Braille-embossing integrated printing process. That is, the system can automatically use the optimization results generated during its own processing to continuously train and improve its core intelligent model, thereby achieving iterative improvement in output quality.

[0120] Optionally, this process is initiated after the system completes the optimization and adjustment of the multi-level depth map. The system will capture and record the following key data: the multi-level depth map generated by the original model before optimization and adjustment, the optimized and adjusted higher-quality multi-level depth map, and the original semantic graph corresponding to the aforementioned depth map. These three are collectively encapsulated into a learning reference sample, which serves as the iterative sample for subsequent model iteration training.

[0121] The system continuously collects such iterative samples during operation, and initiates the learning process periodically or after reaching a certain number. The collected sample set is then input into the pre-trained model originally used to generate the depth map for updated training. After training, a new version of the model is obtained. This new model can incorporate prior knowledge gained from the optimization process of multi-level depth maps.

[0122] This updated model significantly improves the quality of its initial generation results when faced with new semantic graphics, bringing them closer to the state optimized by complex algorithms. This raises the starting point of the system and may simplify the subsequent work of fusing and generating printable patterns, especially the tuning of multi-level depth maps during the fusion process.

[0123] In one embodiment, based on the above embodiments, the Braille printing method combined with 3D relief further includes: In the process of forming a relief model using a multi-wavelength DLP projection system, a corresponding set of exposure parameters is dynamically generated for each relief level of the multi-level depth map; the set of exposure parameters includes projection wavelength, light intensity and exposure time. The multi-wavelength DLP projection system performs layered exposure according to the exposure parameter set corresponding to each relief level to form a relief structure on the resin layer.

[0124] In this embodiment, the exposure parameter set mainly includes three key variables: Projection wavelength: determines the penetration ability of light through the photosensitive resin (curing depth). This is the basis for achieving layering; Light intensity: determines the photon energy delivered to a unit volume of resin per unit time, directly affecting the curing speed and the clarity of the cured area boundary; Exposure time: determines the total amount of energy accumulated, and together with light intensity, determines the final form and degree of curing of a single layer.

[0125] Assuming the relief is pre-processed into three depth levels (shallow, medium, and dark), corresponding to short, medium, and long wavelengths, the system control flow is as follows: (1) Data preparation and parameter calculation After receiving the "multi-level depth map," the system separates it into three independent single-layer mask images (LayerMask), representing the top (shallow), middle, and bottom (deep) layers of the relief. For each layer's mask image, the system analyzes its graphic features in conjunction with 3D printed slice data and can dynamically generate a unique set of exposure parameters by calling a preset model. For example: Top layer (shallow layer, short wavelength): This layer requires high resolution and fine detail. Parameter settings could be: wavelength = 405nm (short-wave ultraviolet), light intensity = medium-high, exposure time = short. The goal is to achieve rapid and precise curing of the surface layer, resulting in clear relief details.

[0126] Middle layer (medium wave): The structure requires a certain strength and depth. Parameter settings could be: wavelength = 455nm (medium wave blue light), light intensity = medium, exposure time = medium. The purpose is to perform stable and controllable secondary curing beneath the already cured surface layer, enhancing the three-dimensional effect of the relief.

[0127] The bottom layer (deep, long wavelength): requires deeper penetration to form a supporting base. Parameter settings could be: wavelength = 520nm (long-wave visible light), light intensity = lower, exposure time = longer. The aim is to utilize the penetrating power of long wavelengths to cure the resin at a deeper depth with gentle but sustained energy, forming a deep, relief-like base while preventing the upper layers from deforming due to overexposure.

[0128] (2) Layered dynamic exposure execution The multi-wavelength DLP projection system strictly follows the parameter set generated above, performing exposures in a deep-to-shallow order (usually the bottom layer is exposed first to provide support for the upper layers): The system switches to a long-wavelength light source, adjusts the light intensity to the calculated value, and loads the mask image of the bottom layer. Exposure is initiated, with the duration precisely controlled to the exposure time calculated for the bottom layer. After completion, the bottom layer resin cures to the predetermined depth. The system then switches to a medium-wavelength light source, adjusts the light intensity, loads the middle layer mask image, and exposes it according to its corresponding exposure time.

[0129] Finally, the system switches to a shortwave light source to complete the top-level exposure using the top-level parameters.

[0130] In one embodiment, by dynamically allocating appropriate wavelengths, light intensities, and action times to each depth level of the relief, precise control over the solidified morphology is achieved, thereby creating a 3D relief with richer details, more reliable structure, and superior texture next to the Braille dot matrix, so that the final integrated "Braille-relief" product reaches a higher level of technology and artistry.

[0131] In one embodiment, based on the above embodiments, the exposure parameter set is generated using a reinforcement learning-based projection strategy optimization model; the training process of the projection strategy optimization model is as follows: We construct a Markov decision process that takes the current printing state as input, exposure parameter set as action, and a comprehensive score of printing quality and efficiency as reward, as the basic framework for the projection strategy optimization model. Using deep reinforcement learning algorithms, the projection strategy optimization model is trained in a resin curing simulation environment until iterative training converges to the optimal strategy.

[0132] In this embodiment, a projection strategy optimization model is introduced. This model uses deep reinforcement learning and its task is to automatically determine the optimal set of exposure parameters for any given relief layer data in order to achieve the best balance between printing quality and efficiency.

[0133] The theoretical basis of this model is the Markov decision process, which is used to formalize the process of selecting exposure parameters for each individual relief level as a step in a sequential decision problem.

[0134] The model's input is the current printed state, which is a multi-dimensional vector that may contain, but is not limited to, the following information: (1) Target level information: the geometric features of the current relief level to be exposed (such as area, perimeter, contour complexity, target depth / thickness).

[0135] (2) Material status information: current resin type, temperature, number of cured layers (historical cumulative exposure).

[0136] (3) Contextual information: the position of the level in the overall relief (top, middle, bottom) and its relative relationship with the adjacent Braille area.

[0137] (4) Equipment status: The current performance baseline of the projection system (such as the degree of light source attenuation).

[0138] (5) Action: The output of the model, i.e., a specific "set of exposure parameters". This corresponds to the agent's action in the Markov decision process. For example, an action can be represented as: wavelength = 455nm, light intensity = 85mW / cm 2 Exposure time = 2.3 seconds.

[0139] (6) Reward: After the action is executed, the system provides evaluation feedback to the model, namely, a comprehensive score of printing quality and efficiency. The reward function is a weighted sum of multiple sub-indicators: R = w1×quality score - w2×exposure time - w3×robustness penalty.

[0140] Where w1 is the weight corresponding to the quality score, which is calculated based on the curing simulation results. For example: the matching degree between the cured shape and the target shape (volume error, edge sharpness), the simulated value of the mechanical strength of the cured layer, and the simulated value of the interlayer bonding force.

[0141] Here, w2 is the weight corresponding to the exposure time, that is, the longer the exposure time, the greater the efficiency penalty.

[0142] Among them, w3 is the weight corresponding to the robustness penalty, which applies a huge negative reward to the combination of parameters that may lead to printing failure (such as layer peeling, over-curing leading to bubbles).

[0143] Optionally, the model can be pre-trained in a high-fidelity resin curing simulation environment, which is a computer-simulated virtual 3D printing system built on physical and chemical principles (such as light propagation models, resin photopolymerization kinetic models, and heat conduction models). It can take a "state" and "action" (exposure parameter set) as input and output the predicted curing result (3D voxel mesh or geometric model), thereby calculating the quality and efficiency indicators of the "print", i.e., the reward.

[0144] Optionally, the model employs deep reinforcement learning algorithms to construct the agent, such as deep deterministic policy gradient, proximal policy optimization, or Soft Actor-Critic. These algorithms can handle high-dimensional, continuous state and action spaces (wavelength, light intensity, and time are all continuous values).

[0145] In each training round, the agent observes the "state" given by the simulation environment and selects an "action" (exposure parameter) based on its current policy (network parameters). The simulation environment executes the action, calculates the fixed result and "reward," and updates to the new "state" (e.g., moving to the next simulation layer). This series of interactive data (state, action, reward, new state) is stored.

[0146] Optionally, the agent uses stored interaction data to continuously update its policy network parameters through deep reinforcement learning algorithms. The goal is to learn a mapping function (policy) that maximizes the expected value of long-term cumulative reward when faced with any input "state".

[0147] After multiple iterations of trial and error and parameter updates in the simulation environment, when the policy's performance (average cumulative reward) stabilizes at a high level and no longer improves significantly, the model training is considered to have converged to the optimal (or near-optimal) policy. At this point, the policy network has internalized the complex photopolymerization laws, material response characteristics, and the quality-efficiency trade-off.

[0148] In one embodiment, a projection strategy optimization model based on deep reinforcement learning is constructed and, guided by a comprehensive score of printing quality and efficiency, engages in large-scale self-play and learning within a virtual resin curing simulation environment. Ultimately, this model can dynamically generate robust, efficient, and high-quality exposure parameters for complex and varied 3D embossed Braille printing tasks.

[0149] In one embodiment, based on the above embodiments, the Braille printing method combined with 3D relief further includes: In the process of creating a textured structure using laser scanning, the absolute Z-axis coordinate of the laser focus at the current scanning point is dynamically calculated based on the deviation between the real-time height of the resin surface and the target height of the textured topographic map. Based on the absolute Z-axis coordinate, the Z-axis position of the laser focus is controlled by the galvanometer system and the dynamic focusing device, so that the laser spot is focused at a constant energy density on the resin liquid surface that needs to be cured or on the resin surface that needs to be etched.

[0150] In this embodiment, when the laser scans the resin surface along a predetermined path and constructs the microstructure corresponding to the convex and concave topographic map, the system will simultaneously measure the actual height of the resin surface where the scanning point is located in real time (e.g., through a confocal displacement sensor or a laser rangefinder).

[0151] The system compares this real-time measured height with the preset target height on the topographic map at that location, and instantly calculates the height deviation between the two. Based on the calculated real-time height deviation and the planned processing depth of the current scanning point (whether it is adding or removing material), the system dynamically calculates the absolute Z-axis coordinates of the laser focus (i.e., the focal point of the laser spot) at the current moment and the current scanning point through a control algorithm.

[0152] This calculation ensures that, regardless of how the resin surface fluctuates due to curing shrinkage, flow, or pre-processing, the laser focus can be precisely positioned at the material interface where photocuring or photolithography actually needs to occur.

[0153] The system synchronously sends the calculated absolute Z-axis coordinate command to the dynamic focusing device (e.g., motorized focusing mirror or acousto-optic modulator) of the laser processing head. The galvanometer system is responsible for controlling the high-speed deflection of the laser beam in the XY plane to determine the scanning point. The dynamic focusing device works in conjunction with the galvanometer to adjust the focal length of the laser beam in real time and quickly, so that the laser focus falls precisely on the absolute Z-axis coordinate position calculated in the previous step.

[0154] Throughout the dynamic focusing process, the laser output power will be compensated accordingly or kept constant to ensure that when the laser focus is locked on the target Z-axis coordinate, the laser energy density of the laser spot focused on the resin (whether it is the surface of liquid resin or the surface of cured resin) remains constant.

[0155] Optionally, if the focus is located on the surface of the liquid resin, a constant energy density is used to achieve precise photocuring and additively construct bumps; if the focus is located on the surface of the cured resin, a constant energy density is used to achieve precise photolithography and subtractively sculpt concave points.

[0156] This overcomes the problems of inaccurate Braille dot height and shape distortion caused by uneven resin surface, curing shrinkage, or lamination errors, ensuring that each Braille dot reaches the designed terrain height. Furthermore, by ensuring the focal point is always on the optimal processing plane, energy diffusion caused by defocusing is avoided, resulting in sharper Braille dot edges, steeper sidewalls, and clearer, more distinct tactile feedback. Simultaneously, closed-loop control of the printing process is achieved, reducing sensitivity to fluctuations in the external environment (such as resin temperature and platform flatness), and improving the overall stability and repeatability of the process.

[0157] In one embodiment, based on the above embodiments, the 3D printing system further includes a focusing processing unit used in conjunction with a laser and / or a multi-wavelength DLP projection system; the Braille printing method combined with 3D relief also includes: During the fabrication of the concave-convex structure and / or the relief structure, the surface three-dimensional morphology of the solidified area is acquired in real time. Based on the real-time deviation between the surface three-dimensional morphology and the target model, and the predetermined action path of the focusing processing unit, the position of the focus of the focusing processing unit and / or the action angle are dynamically adjusted so that the focusing processing unit acts on a local area of ​​the resin surface to be processed. The focusing processing unit is a functional material spraying head.

[0158] In this embodiment, during the process of constructing concave-convex or relief structures using lasers or multi-wavelength DLP systems, the system integrates an online three-dimensional topography detection module (e.g., based on structured light, laser scanning, or confocal microscopy principles). This module acquires the surface three-dimensional topography data of the cured area in real time and non-contact after key processing stages (e.g., after single-layer exposure is completed, or after a processing sub-area is completed), constructing a "digital twin" surface model of the current printed part.

[0159] Optionally, the system will accurately compare the real-time measured 3D topography with the original design target digital model (convex-concave topographic map or multi-level depth map) in the same spatial coordinate system. The algorithm will calculate a real-time geometric deviation map between the two in terms of height, contour, local tilt, etc. This deviation map clearly indicates which areas have insufficient material (below the target), which areas have excessive material (above the target), and whether there are any unexpected surface defects (such as depressions or burrs).

[0160] Optionally, the control algorithm may perform dynamic programming based on the calculated real-time geometric deviation map and the subsequent predetermined action path of the focused processing unit (i.e., the functional material spraying head).

[0161] Optionally, based on the deviation between the current area to be processed and the target model (e.g., the need to fill low-lying areas with material), the precise positioning of the printhead nozzles on the X, Y, and Z axes can be dynamically adjusted to ensure that the material is accurately delivered to the local area that needs compensation, rather than simply being sprayed uniformly along a predetermined path.

[0162] Optionally, for the sides or non-horizontal surfaces of complex reliefs, the system may dynamically adjust the spray angle of the print head relative to the surface to be treated to optimize material adhesion, fill complex contours, or reduce shading effects.

[0163] Optionally, the material extrusion flow rate or droplet jet frequency of the spray head can also be dynamically controlled in conjunction with adjustments to the focus and angle. For areas requiring filling, the flow rate is increased; for areas where buildup needs to be avoided, the flow rate is reduced or even stopped.

[0164] Optionally, the functional material spraying head can perform additive compensation or subtractive pretreatment based on the original processing path according to the dynamically generated adjustment instructions mentioned above.

[0165] Optionally, for areas with insufficient material, the spray head will precisely apply functional materials (e.g., photocurable repair resins, or materials with specific textures / colors) directly to the defective areas at an appropriate angle and flow rate. Subsequently, the system may utilize a low-energy focused light source (such as an auxiliary UV LED) for immediate localized curing, fixing the compensating material in place.

[0166] Optionally, for areas with slight material excess, as an advanced treatment, the system can first spray a weak solvent or surface conditioner (as a type of functional material) to slightly soften or smooth the peaks of excess material, preparing for subsequent laser finishing or global curing, or directly adjust the processing parameters of subsequent layers to compensate.

[0167] In this way, for multi-level relief structures, it can dynamically compensate for sidewall tilt distortion or detail loss caused by deep exposure, significantly improving the forming fidelity of complex curved surfaces and minute features. At the same time, it can identify and fill surface holes and smooth step effects in real time, directly improving the tactile smoothness and visual coherence of Braille dots and relief surfaces.

[0168] Optionally, the functional material spray head can be equipped with different materials, such as resins with different hardness, color, conductivity or magnetic properties, to achieve differentiated design of material properties between the Braille area and the embossed area, such as making the Braille dots more wear-resistant, or giving the embossed area decorative colors.

[0169] In one embodiment, the deep integration of real-time morphological feedback and dynamic focusing material spraying provides online quality control and process compensation methods for high-precision, high-fidelity Braille-emboss integrated manufacturing, greatly enhancing the adaptability of the solution and the success rate of one-time molding of the finished product.

[0170] Furthermore, this application embodiment also provides a 3D printing system, the internal architecture of which can be as follows: Figure 2 As shown, the system includes a processor, memory, communication interface, and input interface connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores data called by the computer programs. The communication interface is used for data communication with external terminals. The input interface is used to receive signals from external devices. When the computer program is executed by the processor, it implements a Braille printing method combining 3D relief as described in the above embodiment.

[0171] Those skilled in the art will understand that Figure 2 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the 3D printing system to which the present application is applied.

[0172] Furthermore, this application also proposes a computer-readable storage medium comprising a computer program that, when executed by a processor, implements the steps of the Braille printing method incorporating 3D relief as described in the above embodiments. It is understood that the computer-readable storage medium in this embodiment can be either a volatile or non-volatile readable storage medium.

[0173] In summary, the Braille printing method, 3D printing system, and computer-readable storage medium combined with 3D relief provided in this application embodiment enable the production of seamless, high-precision, multi-layered 3D relief integrated products with associated semantics in an automated 3D printing process. This not only improves production efficiency but also ensures the overall smoothness and consistency of the user's tactile exploration, enhancing the user experience. Furthermore, by generating multi-level depth maps through neural style transfer and using multi-wavelength light to control the curing depth, it can precisely reproduce the complex spatial layers and textures from shallow textures to deep undulations. This provides an efficient and precise production foundation for realizing personalized, highly expressive tactile content.

[0174] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media provided in this application and in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual-speed SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0175] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0176] The above description is only a preferred embodiment of this application and does not limit the patent scope of this application. Any equivalent structural or procedural changes made based on the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A Braille printing method combining 3D relief, characterized in that, include: The Braille dot matrix is ​​converted into a concave-convex topographic map, and the semantic graphics associated with the Braille dot matrix are used to generate a multi-level depth map through neural style transfer. Using an adaptive fusion algorithm, the concave and convex topographic maps and multi-level depth maps corresponding to each Braille dot matrix are connected and fused to generate a dual-channel depth map as a printing pattern; The printed pattern is input into the light field modulator of the 3D printing system; the 3D printing system also includes a laser and a multi-wavelength DLP projection system; The 3D printing system controls the printing pattern read by the light field modulator, sprays resin onto the printing substrate, and uses a laser to scan the resin surface for the Braille dot matrix area to form a concave-convex structure corresponding to the concave-convex topographic map; and, on the side adjacent to the Braille dot matrix area, a multi-wavelength DLP projection system is used to perform layered exposure to project the relief structure represented by multi-level depth maps onto the resin layer to form a relief shape; wherein, the multi-wavelength DLP projection system uses light of different wavelengths to control the curing depth of the resin for the same relief structure. The printed parts are globally cured using ultraviolet light, forming a single finished product that integrates Braille and 3D relief.

2. The Braille printing method combining 3D relief as described in claim 1, characterized in that, The Braille printing method combining 3D relief also includes: When connecting and fusing the topographic maps and multi-level depth maps corresponding to each Braille dot matrix, the starting point is the end of the topographic map corresponding to each Braille dot matrix, and the corresponding multi-level depth maps are connected and fused at a preset distance. Specifically, a curvature algorithm is used to calculate the first curvature continuity between the final height of the convex-concave topographic map and the edge region layer height of the multi-level depth map, and the edge region layer height is adjusted based on the first curvature continuity with the goal of curvature smoothing.

3. The Braille printing method combining 3D relief as described in claim 2, characterized in that, The Braille printing method combining 3D relief also includes: After adjusting the edge region layer height based on the first curvature continuity, the curvature algorithm is used to calculate the second curvature continuity between the edge region layer height and the core region layer height of the multi-level depth map, and the core region layer height is adjusted based on the second curvature continuity with curvature smoothing as the goal.

4. The Braille printing method combining 3D relief as described in claim 2 or 3, characterized in that, The step of generating a multi-level depth map from the semantic graph associated with the Braille dot matrix through neural style transfer includes: The semantic graphics associated with Braille dots are input into a pre-trained model so that the pre-trained model can generate multi-level depth maps through neural style transfer; wherein, the pre-trained model learns and trains the association relationship between semantic graphics and multi-level depth maps in advance. The Braille printing method combining 3D relief also includes: After adjusting the layer height of the multi-level depth map, iterative samples are generated based on the multi-level depth maps before and after the layer height adjustment and the corresponding semantic graphs. The iterative samples are then input into the pre-trained model for iterative training to update the pre-trained model.

5. The Braille printing method combining 3D relief as described in claim 1, characterized in that, The Braille printing method combining 3D relief also includes: In the process of forming a relief model using a multi-wavelength DLP projection system, a corresponding set of exposure parameters is dynamically generated for each relief level of the multi-level depth map; the set of exposure parameters includes projection wavelength, light intensity and exposure time. The multi-wavelength DLP projection system performs layered exposure according to the exposure parameter set corresponding to each relief level to form a relief structure on the resin layer.

6. The Braille printing method combining 3D relief as described in claim 5, characterized in that, The exposure parameter set is generated using a reinforcement learning-based projection strategy optimization model; the training process of the projection strategy optimization model is as follows: We construct a Markov decision process that takes the current printing state as input, exposure parameter set as action, and a comprehensive score of printing quality and efficiency as reward, as the basic framework for the projection strategy optimization model. Using deep reinforcement learning algorithms, the projection strategy optimization model is trained in a resin curing simulation environment until iterative training converges to the optimal strategy.

7. The Braille printing method combining 3D relief as described in claim 1, characterized in that, The Braille printing method combining 3D relief also includes: In the process of creating a textured structure using laser scanning, the absolute Z-axis coordinate of the laser focus at the current scanning point is dynamically calculated based on the deviation between the real-time height of the resin surface and the target height of the textured topographic map. Based on the absolute Z-axis coordinate, the Z-axis position of the laser focus is controlled by the galvanometer system and the dynamic focusing device, so that the laser spot is focused at a constant energy density on the resin liquid surface that needs to be cured or on the resin surface that needs to be etched.

8. The Braille printing method combining 3D relief as described in claim 1, 5, 6, or 7, characterized in that, The 3D printing system also includes a focusing processing unit used in conjunction with a laser and / or a multi-wavelength DLP projection system; the Braille printing method combined with 3D relief also includes: During the fabrication of the concave-convex structure and / or the relief structure, the surface three-dimensional morphology of the solidified area is acquired in real time. Based on the real-time deviation between the surface three-dimensional morphology and the target model, and the predetermined action path of the focusing processing unit, the position of the focus of the focusing processing unit and / or the action angle are dynamically adjusted so that the focusing processing unit acts on a local area of ​​the resin surface to be processed. The focusing processing unit is a functional material spraying head.

9. A 3D printing system, characterized in that, The main control device of the 3D printing system includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the steps of the Braille printing method combined with 3D relief as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the Braille printing method incorporating 3D relief as described in any one of claims 1 to 8.

Images