A three-dimensional inner envelope multi-view geometry consistent product appearance design method and system

Abstract

The application discloses a three-dimensional inner envelope multi-view geometric consistency product appearance design method and system, relates to the technical field of computer-aided design, computer vision and generative artificial intelligence, and comprises the following steps: obtaining a three-dimensional numerical model of the internal structure of a product, determining a three-dimensional inner envelope safety domain and projecting the three-dimensional inner envelope safety domain to a plurality of two-dimensional view planes to obtain space constraint conditions; inputting the space constraint conditions into a multi-view collaborative generation model, and based on the camera geometric relationship between view angles, limiting cross-view feature interaction through epipolar lines, homonym candidate regions or geometric consistency neighborhood to generate a plurality of two-dimensional appearance images of different view angles; reconstructing a three-dimensional appearance model of the product based on the images, and performing assembly interference detection and manufacturability detection; and mapping three-dimensional defect regions that do not meet a preset threshold to two-dimensional local feedback information to drive local regeneration, repair or deformation optimization. The application can improve the assembly feasibility, multi-view consistency and design iteration efficiency of an appearance scheme.

Description

Technical Field

[0001] This invention relates to the fields of computer-aided design, computer vision, and generative artificial intelligence. Specifically, it relates to a method and system for designing the appearance of a product with geometric consistency across three-dimensional inner envelope multi-view views. Background Technology

[0002] In industrial product design using computer-aided design, computer vision, and generative artificial intelligence technologies, the product's appearance directly impacts user perception, product recognizability, and market competitiveness. Traditional industrial design processes typically rely on designers drawing two-dimensional sketches or three-view diagrams based on their experience, followed by repeated modifications by structural engineers considering internal component space, assembly clearances, and manufacturing process requirements. This process suffers from long design cycles, high iteration costs, and low efficiency in connecting two-dimensional appearance designs with three-dimensional engineering structures.

[0003] In recent years, generative artificial intelligence technologies such as generative adversarial networks, diffusion models, flow matching models, and autoregressive image generation models have been increasingly applied to product appearance design, enabling the rapid generation of a large number of diverse conceptual images. However, existing generative appearance design methods still face the following problems when applied to actual industrial product development: First, existing image generation models primarily rely on two-dimensional image semantics and pixel distribution to generate appearance schemes, lacking constraints on the three-dimensional spatial occupancy of the product's internal structure. Consequently, the generated appearance contour may fail to cover internal components, or the inner boundary of the outer casing may encroach upon the safety space required by the internal structure, leading to a higher risk of subsequent assembly interference.

[0004] Second, existing multi-view generation methods typically generate multiple perspective images such as front view, side view, and top view. There is a lack of unified geometric consistency constraints between the views, which can easily lead to problems such as mismatched feature lines, drifting of component positions, misalignment of contours, or inconsistency in topology, affecting the stability of subsequent 3D appearance reconstruction.

[0005] Third, existing AI-generated appearance methods typically lack a feedback mechanism that integrates with assembly constraints and manufacturability evaluation. Even if the generated appearance design has a certain aesthetic appeal, it may still have problems such as insufficient draft angle, failure to meet minimum wall thickness requirements, local undercuts, and insufficient assembly clearance. These issues require multiple rounds of manual modification, resulting in high design iteration costs.

[0006] Therefore, there is an urgent need for a product appearance generation solution that can unify the three-dimensional constraints of the product's internal structure, the generation of geometric consistency across multiple views, the reconstruction of the three-dimensional appearance, and the feedback from assembly and manufacturability evaluation into the same technical process. Summary of the Invention

[0007] To address the aforementioned issues, the present invention aims to provide a three-dimensional inner envelope multi-view geometric consistency product appearance design method and system. This method involves converting a three-dimensional digital model of the product's internal structure into a three-dimensional inner envelope safety domain, projecting it to generate a multi-view constraint diagram, and then inputting this constraint along with multi-view geometric consistency constraints into the generated model. This ensures that the generated product appearance contour can encompass the three-dimensional inner envelope safety domain and improves the spatial projection consistency between multiple views. Simultaneously, through three-dimensional appearance reconstruction, assembly interference detection, and manufacturability testing, defective areas are back-projected into two-dimensional local feedback information to drive the generated model to perform local regeneration, local repair, or deformation optimization.

[0008] To achieve the above technical objectives, this application provides a three-dimensional inner envelope multi-view geometrically consistent product appearance design method, comprising the following steps: Obtain a 3D digital model of the product's internal structure, determine the 3D inner envelope safety domain based on the 3D digital model, and project the 3D inner envelope safety domain onto the corresponding 2D view plane to obtain the spatial constraints corresponding to multiple viewpoints; Spatial constraints are input into the multi-view collaborative generation model. Based on the camera geometric relationship between at least two viewpoints, during the generation process of the multi-view collaborative generation model, cross-view feature interaction is restricted by epipolar lines, candidate regions of corresponding points, or geometrically consistent neighborhoods. This generates two-dimensional appearance images of the product from multiple viewpoints, so that the product appearance contours corresponding to the two-dimensional appearance images from multiple viewpoints cover the three-dimensional inner envelope safety domain, and the corresponding product appearance inner boundary does not intrude into the three-dimensional inner envelope safety domain. A 3D appearance model of the product is obtained by reconstructing the 3D appearance image based on multiple 2D appearance images from different perspectives. Perform assembly interference detection and manufacturability testing on the product's 3D appearance model; When a 3D defect region that does not meet the preset threshold is detected, the 3D defect region is mapped to a local penalty mask, local redrawing region or local condition control chart in the corresponding 2D view, and fed back to the multi-view collaborative generation model for local regeneration, local denoising, local repair or parametric deformation optimization.

[0009] Preferably, when obtaining spatial constraints, a three-dimensional digital model of the product's internal structure is obtained; The three-dimensional distance field is based on the calculation of the space occupied by the internal structure and the safety margin on the outside of the three-dimensional digital model. Based on preset safety distance thresholds, minimum wall thickness thresholds, heat dissipation gap thresholds, and / or assembly gap thresholds, the three-dimensional distance field is expanded to obtain a three-dimensional inner envelope safety domain. Based on the camera projection matrix of at least two preset reference viewpoints, the three-dimensional inner envelope safety domain is projected onto the corresponding two-dimensional view plane to generate a two-dimensional constraint graph for each view. The two-dimensional constraint graph includes at least one of the following: inner envelope occupancy mask, boundary distance transformation graph, and inner envelope constraint penalty weight graph. The two-dimensional constraint graph is used as a spatial constraint input to the image generation model, so that the product appearance contour generated by the image generation model covers the three-dimensional inner envelope safety domain, and the inner boundary of the generated product appearance does not intrude into the three-dimensional inner envelope safety domain.

[0010] Preferably, when obtaining spatial constraints, the three-dimensional distance field includes at least one of a signed distance field, an unsigned distance field, a voxel distance field, a point cloud distance field, or a neural implicit distance field; the three-dimensional inner envelope security domain is obtained by performing distance threshold-based dilation, morphological dilation, bias surface generation, or implicit isosurface extraction on the surface of the internal structure.

[0011] Preferably, when obtaining spatial constraints, the spatial constraints are used as at least one of the following input image generation models: binary mask, continuous distance map, alpha channel, conditional control map, attention bias map, loss function weight map, or sampling guide map.

[0012] Preferably, when acquiring two-dimensional appearance images from multiple perspectives, a multi-view collaborative generation model is constructed to jointly generate at least two two-dimensional appearance images of the same product from different perspectives; Obtain the camera projection matrix, camera extrinsic matrix, camera intrinsic matrix, fundamental matrix, or affine epipolar constraint between at least two different viewpoints; In the process of generating target view feature map by multi-view collaborative generation model, for query point in target view feature map, epiline, candidate region of corresponding point or geometrically consistent neighborhood corresponding to query point is determined in reference view feature map. By using attention masks, attention bias terms, or geometric consistency loss functions, the cross-view feature interaction range between the query point and the reference view feature map is limited, so that the generated two-dimensional appearance images from multiple perspectives meet the preset geometric consistency threshold at appearance boundaries, feature lines, and / or local component locations.

[0013] Preferably, when acquiring two-dimensional appearance images from multiple perspectives, the multi-view collaborative generation model includes a conditional diffusion model, a diffusion Transformer model, a flow matching generation model, an autoregressive image generation model, a generative adversarial network, or a combination thereof.

[0014] Preferably, when acquiring two-dimensional appearance images from multiple perspectives, the geometric consistency neighborhood is the strip region on both sides of the epipolar line; the width of the strip region adopts at least one of the following: a fixed pixel width, an annealing width that decreases with the generation time step, an adaptive width determined according to the depth estimation uncertainty, or a pyramid width that is scaled layer by layer according to the feature map resolution.

[0015] Preferably, when performing 3D appearance reconstruction, multi-view stereo vision, differentiable rendering, neural radiation field, 3D Gaussian representation, neural implicit symbolic distance field, parametric surface fitting, or a combination thereof are used; wherein, the 3D appearance model of the product includes at least one of triangular mesh, parametric CAD surface, implicit field, point cloud, or explicit Gaussian representation.

[0016] Preferably, when performing the tests, the manufacturability tests include at least one or more of the following: draft angle test, minimum wall thickness test, radius of curvature test, undercut area test, parting surface feasibility test, stiffener thickness test, and hole safety distance test.

[0017] Preferably, during the inspection process, the inspection results of assembly interference inspection and manufacturability inspection are fused into a multi-objective evaluation function, which includes an inner envelope term, an assembly interference term, and multiple viewpoints. Figure 1 At least two of the following: consistency, draft angle, wall thickness, and local smoothness.

[0018] Preferably, during the inspection, the three-dimensional defect region is mapped to a local penalty mask, local redraw area, or local condition control map in the corresponding two-dimensional view through the camera projection matrix, UV mapping relationship, differentiable rendering inverse mapping relationship, or nearest surface point mapping relationship.

[0019] Preferably, when acquiring creative design material images, a cross-modal feature coding model, a visual fundamental model, or an image segmentation model is used to extract global contour features, local component features, material features, or style features of the creative design material images; The coverage fit between the creative design material image and the three-dimensional inner envelope safety domain is calculated based on spatial constraints, and creative design material images that do not meet the coverage fit threshold are filtered out.

[0020] Preferably, when acquiring creative design material images, the cross-modal feature encoding model includes CLIP, SigLIP, DINO, DINOv2, SAM, visual Transformer encoder or a combination thereof; the coverage fit is calculated by at least one of cross-union ratio, chamfer distance, boundary distance transformation, contour coverage ratio or inner envelope constraint penalty value.

[0021] Based on the same inventive concept, this application also provides an intelligent design system for the appearance of industrial products, comprising: The module includes several functionalities: an inner envelope constraint modeling module for acquiring a 3D digital model of the product's internal structure, calculating the 3D distance field, and generating a 3D inner envelope safety domain; a multi-view constraint projection module for projecting the 3D inner envelope safety domain onto multiple 2D view planes, generating an inner envelope occupancy mask, a boundary distance transformation map, and / or an inner envelope constraint penalty weight map; a multi-view collaborative generation module for generating multiple perspective 2D appearance images based on the inner envelope occupancy mask, boundary distance transformation map, and / or inner envelope constraint penalty weight map, and restricting cross-view feature interactions based on geometric consistency constraints between views; a 3D appearance reconstruction module for reconstructing the product's 3D appearance model based on the multiple perspective 2D appearance images; and a manufacturability evaluation and feedback module for performing assembly interference detection and manufacturability detection on the product's 3D appearance model, mapping the detected 3D defect areas to 2D local feedback information to trigger the multi-view collaborative generation module to perform local regeneration, local repair, or deformation optimization.

[0022] Based on the same inventive concept, this application also provides a computer device, including at least one processor and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, and when the computer program is executed by the at least one processor, it implements the method described in any of the above-mentioned embodiments.

[0023] Based on the same inventive concept, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in any of the above-mentioned embodiments.

[0024] The present invention discloses the following technical effects: This invention improves the probability that the AI-generated appearance scheme meets the assembly space constraints by introducing a three-dimensional inner envelope security domain, an inner envelope occupancy mask, a boundary distance transformation map, and / or an inner envelope constraint penalty weight map into the image generation process. Furthermore, it allows for the quantitative control of assembly risks through preset safety thresholds.

[0025] This invention improves the stability of subsequent 3D appearance reconstruction by introducing epipolar neighborhood constraints, cross-view attention masks, attention bias terms, or geometric consistency loss, so that 2D images from different perspectives satisfy spatial projection correspondence within a preset error threshold.

[0026] This invention detects assembly interference, draft angle, minimum wall thickness, and undercut areas in a three-dimensional appearance model, and back-projects the defective areas into a two-dimensional local penalty mask, a local redrawing area, or a local condition control chart. This can drive the generated model to perform local regeneration or repair, reducing design iteration costs. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly described below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0028] Figure 1 This is a schematic diagram of the product appearance generation process based on three-dimensional inner envelope constraints and multi-view geometric consistency as described in this invention.

[0029] Figure 2 This is a schematic diagram illustrating the principle of generating a three-dimensional inner envelope safety domain from a three-dimensional distance field and projecting it to form a multi-view constraint diagram, as described in this invention.

[0030] Figure 3 This is a schematic diagram of the cross-view feature interaction mechanism in a multi-view collaborative generation model based on epipolar neighborhood constraints as described in this invention.

[0031] Figure 4 This is a schematic diagram illustrating the generation of geometric consistency of VR headset appearance in multiple views, three-dimensional reconstruction, and inner envelope coverage effect in the embodiments described in this invention.

[0032] Figure 5 This is the overall flowchart of the product appearance generation system based on three-dimensional inner envelope constraints and multi-view geometric consistency as described in this invention.

[0033] in, Figure 2 In the diagram, 201 is the three-dimensional digital model of the internal structure; 202 is the three-dimensional inner envelope safety domain or inner envelope safety boundary; 203 is the distance field calculation space or voxelization space; 204 is the inner envelope occupancy mask of the first view or the multi-view constraint graph; 205 is the top view projection plane; 206 is the second view projection plane; 207 is the inner envelope occupancy mask of the second view; and 208 is the top view inner envelope occupancy mask, boundary distance transformation graph, and / or inner envelope constraint penalty weight graph.

[0034] Figure 3 In the diagram, 301 is the feature map of the target view; 302 is the feature map of the reference view; 303 is the location of the query feature in the target view; 304 is the cross-view feature interaction path defined by the geometric relationship between views; and 305 is the epipolar neighborhood or candidate region of the same name in the reference view corresponding to the location of the query feature.

[0035] Figure 4In the diagram, 401 represents the generated VR headset multi-view appearance outline; 402 represents the multi-view geometric consistency reference line or inner envelope projection reference line; 403 represents the 3D appearance model; 404 represents the appearance surface or local mesh area of ​​the 3D appearance model; 405 represents the appearance shell; 406 represents the 3D inner envelope safety boundary; 407 represents the safety clearance; and 408 represents a magnified view of the local cross-section. Detailed Implementation

[0036] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0037] like Figures 1 to 5 As shown, the present invention provides a product appearance design method, system, device and medium based on three-dimensional inner envelope constraints and multi-view geometric consistency, which are applied in the fields of computer-aided design, computer vision and generative artificial intelligence. It includes steps such as three-dimensional inner envelope security domain construction, multi-view constraint graph generation, multi-view geometric consistency collaborative generation, three-dimensional appearance reconstruction and assembly constraint and manufacturability evaluation feedback.

[0038] The following describes the invention using the intelligent generation of the appearance of a novel lightweight VR head-mounted display as a specific embodiment.

[0039] In one embodiment, the system first acquires a 3D CAD solid model of the internal hardware components of the VR headset. The 3D CAD solid model can be in STEP, IGES, OBJ, STL, or other data formats suitable for 3D geometric calculations. The internal hardware components may include a binocular optical lens module, a motherboard circuit, a front-mounted cooling fan, a rear-mounted curved battery compartment, and a connecting bracket. The system can also acquire manufacturing process constraints, such as the use of injection molding for the outer shell, the demolding direction vector, the minimum wall thickness threshold, the minimum draft angle threshold, the heat dissipation gap threshold, and the assembly gap threshold.

[0040] Step S1: Construction of 3D Inner Envelope Security Region and Generation of Multi-View Constraint Graph In one embodiment, step S1 is used to convert the three-dimensional digital model of the product's internal structure into a two-dimensional constraint graph that can be used to generate a model. The two-dimensional constraint graph includes at least one of an inner envelope placeholder mask, a boundary distance transformation graph, and an inner envelope constraint penalty weight graph.

[0041] For example, the system places the acquired three-dimensional digital model 201 of the internal structure into a distance field calculation space or a voxelization space 203, and performs voxelization processing on it. The system calculates the distance from the coordinates of any point in the calculation space to the surface of the internal structure, generating a three-dimensional distance field. The three-dimensional distance field can be a signed distance field (SDF), an unsigned distance field (UDF), a point cloud distance field, a voxel distance field, or a neural implicit distance field.

[0042] For example, when using a symbolic distance field, the system can specify that the distance value inside the internal structure surface is negative, and the distance value outside is positive. The system determines the three-dimensional inner envelope safety domain based on a preset safety threshold D_safe. The preset safety threshold D_safe can be determined by one or more of the following: minimum safety wall thickness, heat dissipation gap, assembly gap, or structural tolerance. For example, if the minimum safety wall thickness is 1.5 mm and the heat dissipation gap is 1.0 mm, then D_safe can be set to 2.5 mm. The system traverses the voxel space and determines the voxel region that satisfies SDF(x,y,z)≤D_safe as the three-dimensional inner envelope safety domain jointly formed by the internal structure and its outer safety margin, and can use the isosurface of SDF(x,y,z)=D_safe as the three-dimensional inner envelope safety boundary 202. The three-dimensional inner envelope safety domain is used to constrain the subsequently generated appearance contour to must cover the safety domain, and the inner boundary of the appearance must not intrude into the safety domain.

[0043] For example, the system projects the three-dimensional inner envelope safety domain onto the corresponding two-dimensional view plane based on the camera projection matrix of multiple preset reference viewpoints, generating a first view inner envelope occupancy mask or multi-view constraint map 204, a second view inner envelope occupancy mask 207, and a top view inner envelope occupancy mask, boundary distance transformation map, and / or inner envelope constraint penalty weight map 208.

[0044] For example, the area covered by the projection of the 3D inner envelope safety domain is marked as the inner envelope occupancy area, and its corresponding mask value can be set to 1; the uncovered area is marked as the outer candidate design area, and its corresponding mask value can be set to 0. The system further performs a distance transformation on the boundary of the inner envelope occupancy area to obtain a boundary distance transformation map. The boundary distance transformation map is used to represent the distance from each pixel position in the 2D view to the boundary of the 3D inner envelope safety domain projection.

[0045] For example, the system can also generate an inner envelope constraint penalty weight map based on the inner envelope occupancy mask and / or boundary distance transformation map. The inner envelope constraint penalty weight map is used to assign higher penalty weights to regions where the appearance contour intrudes into the three-dimensional inner envelope safety domain, approaches the three-dimensional inner envelope safety boundary, or does not meet a preset safety threshold. It can also serve as a conditional input for the generated model, loss function weights, attention bias terms, local redraw constraints, or post-processing evaluation criteria. The system supports outputting the two-dimensional constraint map as a visualization layer to the front-end interactive interface, using it as an enveloping constraint condition to constrain the appearance contour enveloping relationship in the two-dimensional generation space.

[0046] Step S2: Cross-modal feature decoupling and construction of a constrained creative asset library In one embodiment, step S2 is used to obtain creative design materials and filter out materials that do not meet the coverage and fit requirements based on the two-dimensional constraint diagram generated in step S1.

[0047] For example, the system extracts two-dimensional planar materials with streamlined contour features from a creative database. The system uses a cross-modal feature encoding model, a visual fundamental model, or an image segmentation model to extract features from the materials, encoding the local feature maps or segmented regions of the materials as at least one of local component features, material features, and style features, and encoding their global features as global contour features. The cross-modal feature encoding model may include CLIP, SigLIP, DINO, DINOv2, SAM, a visual Transformer encoder, or a combination thereof.

[0048] For example, the system binarizes the extracted appearance contour and performs matching calculations with the corresponding view inner envelope placeholder mask, boundary distance transformation map, and / or inner envelope constraint penalty weight map generated in step S1. The matching calculations include at least one of the following: intersection-union ratio, chamfer distance, boundary distance transformation, contour coverage rate, or inner envelope constraint penalty value. When it is detected that the appearance inner boundary corresponding to the material contour crosses the inner envelope placeholder area, or the material outer contour fails to cover the inner envelope placeholder area, the system determines that its coverage fit does not meet a preset threshold and filters out the material. The system stores the feature vectors that meet the preset coverage fit threshold in the constrained creative asset library.

[0049] Step S3: Collaborative generation based on multi-view geometric consistency constraints In one embodiment, step S3 is used to introduce epipolar neighborhood constraints, corresponding point candidate regions, attention masks, attention bias terms, or geometric consistency losses during the multi-view collaborative generation process to reduce the risks of contour misalignment, component drift, and topological inconsistency between different views.

[0050] For example, the system constructs a multi-view collaborative generation model. This model can employ a conditional diffusion model, a diffusion Transformer model, a flow matching generation model, an autoregressive image generation model, a generative adversarial network, or a combination thereof. The system uses at least one of the extracted global contour features, local component features, material features, and style features as generation conditions, and inputs the inner envelope placeholder mask, boundary distance transformation map, and / or inner envelope constraint penalty weight map as spatial constraints into the multi-view collaborative generation model. Combined with... Figure 3 As shown, during the multi-view collaborative generation process, the query feature position 303 in the target view feature map 301 can be mapped to the epipolar neighborhood or corresponding point candidate region 305 in the reference view feature map 302 based on the camera parameters, projection relationship, or estimated depth information between the target view and the reference view. The cross-view feature interaction path 304 is restricted to the epipolar neighborhood or corresponding point candidate region 305, or constrained by attention mask, attention bias term, cost volume constraint, or geometric consistency loss, thereby reducing the risk of contour misalignment, component drift, and topological inconsistency between different views.

[0051] For example, when the multi-view collaborative generation model generates a query point q_1 on the first view feature map 301, the system calculates the epipolar line or candidate matching region corresponding to the query point q_1 on the reference view feature map 302 based on the preset camera extrinsic matrix, camera intrinsic matrix, or fundamental matrix between the main view and the side view, and expands it to both sides based on a preset pixel width to form an epipolar neighborhood or corresponding point candidate region 305. The system can limit or adjust the feature interaction range between the query point q_1 and the corresponding candidate region in the reference view feature map through attention mask, attention bias term, or geometric consistency loss, so that the output multi-view image satisfies the three-dimensional spatial projection correspondence within a preset error threshold.

[0052] Step S4: 3D Appearance Reconstruction and Geometric Consistency Verification Combination Figure 4 The illustrated embodiment shows a schematic diagram of the VR headset appearance multi-view geometric consistency generation, 3D reconstruction, and inner envelope covering effect. This embodiment intuitively demonstrates the technical effects of the present invention in the multi-view collaborative generation, 3D appearance reconstruction, and inner envelope covering verification stages.

[0053] like Figure 4As shown on the left, in the two-dimensional joint generation stage, the system outputs a VR headset multi-view appearance contour 401 that satisfies the multi-view geometric consistency constraint. By introducing epipolar neighborhood constraints, corresponding point candidate regions, attention masks, attention bias terms, or geometric consistency loss, a geometric correspondence 402 is formed between the front view, top view, and / or side view, ensuring that the edge features, component positions, and appearance contours in each view satisfy spatial projection consistency within a preset error threshold. Based on the above multi-view appearance image, the system calls at least one method among multi-view geometric reconstruction, differentiable rendering optimization, neural implicit reconstruction, point cloud reconstruction, mesh reconstruction, 3D Gaussian reconstruction, or parametric surface fitting to generate, as shown in the image. Figure 4 The product's 3D appearance model 403 is shown on the right. Thanks to the aforementioned 3D inner envelope constraint and multi-view geometric consistency constraint, the appearance contour of the 3D appearance model 403 can cover the 3D inner envelope safety domain, and its inner boundary does not intrude into the 3D inner envelope safety domain, so that its appearance surface or local mesh region 404 presents a relatively continuous geometric shape.

[0054] like Figure 4 As shown in the enlarged cross-sectional view 408, a safety gap 407 is formed between the outer shell 405 and the three-dimensional inner envelope safety boundary 406. The three-dimensional inner envelope safety boundary 406 can be generated according to a preset safety threshold D_safe. The safety gap 407 is used to characterize the reserved distance between the inner boundary of the outer shell and the three-dimensional inner envelope safety boundary, thereby indicating that the inner boundary of the outer shell 405 does not intrude into the three-dimensional inner envelope safety domain. The output three-dimensional appearance model can be used for subsequent assembly interference detection, manufacturability evaluation, and engineering review reference.

[0055] Step S5: Assembly Constraints and Manufacturability Evaluation Feedback In one embodiment, step S5 is used to perform assembly interference detection and manufacturability detection on the reconstructed three-dimensional appearance model, and to feed back the detected three-dimensional defect areas to the generated model.

[0056] For example, the system traverses the facets or curved surfaces on the reconstructed 3D appearance model and calculates the angle between the surface normal vector N_i and the preset demolding direction vector V_mold for draft angle detection. If a facet has a draft angle less than a preset safe demolding angle, or a local thickness less than a preset minimum wall thickness threshold, the system marks it as a region that does not meet the preset threshold.

[0057] For example, the system integrates the aforementioned manufacturability indicators with assembly interference scores to construct a multi-objective evaluation function. When a snap-fit, thin-walled region, undercut region, assembly interference region, or other region that does not meet the preset threshold is detected, the system can generate a visual warning label containing defect location attributes in the front-end interface or interaction log; at the same time, based on the camera projection matrix, UV mapping relationship, or differentiable rendering reverse mapping relationship, the three-dimensional defect region is mapped back to the corresponding two-dimensional view to generate a two-dimensional local penalty mask, local redrawing region, or local condition control chart.

[0058] For example, the system feeds back the two-dimensional local penalty mask, the local redrawn region, or the local condition control chart to the multi-view collaborative generation model in step S3, triggering local regeneration, local denoising, local repair, or parametric deformation optimization of the defective region. After one or more iterations, when the multi-objective evaluation function meets the convergence condition or reaches the preset number of iterations, it outputs a VR head-mounted display appearance multi-view scheme and a three-dimensional appearance model that meet the preset assembly constraints and manufacturability evaluation threshold. It should be understood that although the steps in the flowchart of this application embodiment are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction for the execution of these steps, and they can be executed in other orders. Moreover, at least some of the steps in the figure may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps. Obviously, those skilled in the art can make various modifications and variations to this invention without departing from the spirit and scope of this invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention also intends to include these modifications and variations.

Claims

1. A method for designing the appearance of a product with geometrically consistent three-dimensional inner envelope multi-view view, characterized in that, Includes the following steps: A three-dimensional digital model of the internal structure of the product is obtained. Based on the three-dimensional digital model, a three-dimensional distance field representing the space occupied by the internal structure and its outer safety margin is calculated. The three-dimensional distance field is expanded according to a preset safety distance threshold, minimum wall thickness threshold, heat dissipation gap threshold and / or assembly gap threshold to obtain a three-dimensional inner envelope safety domain. The three-dimensional inner envelope safety domain is then projected onto the corresponding two-dimensional view plane to obtain spatial constraints corresponding to multiple viewpoints. The spatial constraints are input into the multi-view collaborative generation model. Based on the camera geometric relationship between at least two viewpoints, during the generation process of the multi-view collaborative generation model, cross-view feature interactions are restricted by epipolar lines, candidate regions of corresponding points, or geometrically consistent neighborhoods. This generates two-dimensional appearance images of the product from multiple viewpoints, such that the product appearance contours corresponding to the two-dimensional appearance images from multiple viewpoints cover the three-dimensional inner envelope safety domain, and the corresponding inner boundary of the product appearance does not intrude into the three-dimensional inner envelope safety domain. Based on the two-dimensional appearance images from multiple perspectives, a three-dimensional appearance reconstruction is performed to obtain a three-dimensional appearance model of the product. Assembly interference detection and manufacturability detection are performed on the three-dimensional appearance model of the product; When a 3D defect region that does not meet the preset threshold is detected, the 3D defect region is mapped to a local penalty mask, local redrawing region or local condition control map in the corresponding 2D view, and fed back to the multi-view collaborative generation model for local regeneration, local denoising, local repair or parametric deformation optimization.

2. The product appearance design method according to claim 1, characterized in that, When obtaining spatial constraints, obtain a three-dimensional digital model of the product's internal structure; The three-dimensional distance field is calculated based on the three-dimensional digital model to represent the space occupied by the internal structure and the safety margin on its outer side. The three-dimensional distance field is expanded based on the preset safety distance threshold, minimum wall thickness threshold, heat dissipation gap threshold and / or assembly gap threshold to obtain the three-dimensional inner envelope safety domain. Based on the camera projection matrix of at least two preset reference viewpoints, the three-dimensional inner envelope safety domain is projected onto the corresponding two-dimensional view plane to generate a two-dimensional constraint graph for each view. The two-dimensional constraint graph includes at least one of the following: inner envelope occupancy mask, boundary distance transformation graph, and inner envelope constraint penalty weight graph. The two-dimensional constraint diagram is input as a spatial constraint condition into the multi-view collaborative generation model, so that the product appearance contour generated by the multi-view collaborative generation model covers the three-dimensional inner envelope safety domain, and the inner boundary of the generated product appearance does not intrude into the three-dimensional inner envelope safety domain.

3. The product appearance design method according to claim 2, characterized in that: When acquiring spatial constraints, the three-dimensional distance field includes at least one of a signed distance field, an unsigned distance field, a voxel distance field, a point cloud distance field, or a neural implicit distance field; the three-dimensional inner envelope security domain is obtained by performing distance threshold-based dilation, morphological dilation, bias surface generation, or implicit isosurface extraction on the internal structure surface.

4. The product appearance design method according to claim 2, characterized in that: When obtaining spatial constraints, the spatial constraints are input into the multi-view collaborative generation model as at least one of the following: a binary mask, a continuous distance map, an alpha channel, a conditional control map, an attention bias map, a loss function weight map, or a sampling guidance map.

5. The product appearance design method according to claim 1, characterized in that, When acquiring two-dimensional appearance images from multiple perspectives, a multi-view collaborative generation model is constructed to jointly generate at least two two-dimensional appearance images of the same product from different perspectives. The multi-view collaborative generation model is also used to obtain feedback results and perform optimization. Obtain the camera projection matrix, camera extrinsic matrix, camera intrinsic matrix, fundamental matrix, or affine epipolar constraint between at least two different viewpoints; During the process of generating the target view feature map by the multi-view collaborative generation model, for the query point in the target view feature map, the epiline, candidate region of the same name point or geometrically consistent neighborhood corresponding to the query point is determined in the reference view feature map. By using attention masks, attention bias terms, or geometric consistency loss functions, the cross-view feature interaction range between the query point and the reference view feature map is limited, so that the generated two-dimensional appearance images from multiple perspectives meet the preset geometric consistency threshold at appearance boundaries, feature lines, and / or local component positions.

6. The product appearance design method according to claim 5, characterized in that: When acquiring two-dimensional appearance images from multiple perspectives, the multi-view collaborative generation model includes a conditional diffusion model, a diffusion Transformer model, a flow matching generation model, an autoregressive image generation model, a generative adversarial network, or a combination thereof.

7. The product appearance design method according to claim 5, characterized in that: When acquiring two-dimensional appearance images from multiple perspectives, the geometric consistency neighborhood is the strip region on both sides of the epipolar line; the width of the strip region adopts at least one of the following: a fixed pixel width, an annealing width that decreases with the generation time step, an adaptive width determined according to the uncertainty of depth estimation, or a pyramid width that is scaled layer by layer according to the feature map resolution.

8. The product appearance design method according to claim 1, characterized in that: When performing 3D appearance reconstruction, multi-view stereo vision, differentiable rendering, neural radiation field, 3D Gaussian representation, neural implicit symbolic distance field, parametric surface fitting or a combination thereof are used; wherein, the 3D appearance model of the product includes at least one of triangular mesh, parametric CAD surface, implicit field, point cloud or explicit Gaussian representation.

9. The product appearance design method according to claim 1, characterized in that: When performing the tests, the manufacturability tests include at least one or more of the following: draft angle test, minimum wall thickness test, radius of curvature test, undercut area test, parting surface feasibility test, stiffener thickness test, and hole safety distance test.

10. The product appearance design method according to claim 1, characterized in that: During the inspection, the results of the assembly interference inspection and manufacturability inspection are fused into a multi-objective evaluation function, which includes at least two of the following: inner envelope coverage term, assembly interference term, multi-view consistency term, draft angle term, wall thickness term, and local smoothness term.

11. The product appearance design method according to claim 1, characterized in that: During the detection process, the three-dimensional defect region is mapped to a local penalty mask, local redrawing region, or local condition control map in the corresponding two-dimensional view through the camera projection matrix, UV mapping relationship, differentiable rendering inverse mapping relationship, or nearest surface point mapping relationship.

12. The product appearance design method according to claim 1, characterized in that: When acquiring creative design material images, the global contour features, local component features, material features, or style features of the creative design material images are extracted using cross-modal feature coding models, visual basic models, or image segmentation models. The coverage fit between the creative design material image and the three-dimensional inner envelope safety domain is calculated based on the spatial constraints, and creative design material images that do not meet the coverage fit threshold are filtered out.

13. The product appearance design method according to claim 12, characterized in that: When acquiring creative design material images, the cross-modal feature encoding model includes CLIP, SigLIP, DINO, DINOv2, SAM, visual Transformer encoder or a combination thereof; the envelopment fit is calculated by at least one of cross-union ratio, chamfer distance, boundary distance transformation, contour coverage ratio or inner envelope constraint penalty value.

14. A three-dimensional inner envelope multi-view geometrically consistent product appearance design system, used to execute the product appearance design method as described in claim 1, characterized in that, include: The inner envelope constraint modeling module is used to obtain a three-dimensional digital model of the internal structure of the product, calculate a three-dimensional distance field representing the space occupied by the internal structure and its outer safety margin, and expand the three-dimensional distance field according to a preset safety distance threshold, minimum wall thickness threshold, heat dissipation gap threshold and / or assembly gap threshold to obtain a three-dimensional inner envelope safety domain. The multi-view constraint projection module is used to project the three-dimensional inner envelope safety domain onto multiple two-dimensional view planes to generate an inner envelope occupancy mask, a boundary distance transformation map, and / or an inner envelope constraint penalty weight map. The multi-view collaborative generation module is used to generate two-dimensional appearance images from multiple perspectives based on the inner envelope occupancy mask, boundary distance transformation map and / or inner envelope constraint penalty weight map, and to restrict cross-view feature interaction based on geometric consistency constraints between views. The 3D appearance reconstruction module is used to reconstruct a 3D appearance model of the product based on the 2D appearance images from the multiple perspectives. The manufacturability evaluation and feedback module is used to perform assembly interference detection and manufacturability detection on the three-dimensional appearance model of the product, and to map the detected three-dimensional defect areas into two-dimensional local feedback information to trigger the multi-view collaborative generation module to perform local regeneration, local repair or deformation optimization.

15. A computer device, characterized in that, include: At least one processor; And a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, which, when executed by the at least one processor, implements the method as described in any one of claims 1 to 13.

16. 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 method as described in any one of claims 1 to 13.

Images