Three-dimensional model generation method and system based on lightweight glass containers

Abstract

The present invention relates to the technical field of model construction, and particularly relates to a three-dimensional model generation method and system based on a lightweight glass container. The method includes the following steps: using a preset multi-modal scanning sensor to collect information on the lightweight glass container, collecting multi-view visible light images, depth information, and structured light projection data, and constructing a standard multi-view point cloud data set; performing multi-modal fusion based on the standard multi-view point cloud data set to obtain multi-modal fusion data of the glass container; judging the thin wall of the multi-modal fusion data of the glass container and repairing the thin wall structure to obtain a three-dimensional point cloud model with high integrity; Therefore, the present invention solves the problems of low structural reconstruction accuracy, large transparent material error, and lack of process evaluation in traditional glass container modeling through multi-modal data fusion, geometric and optical joint optimization, and multi-scale path evaluation mechanism, and improves the geometric fidelity, material restoration degree, and controllability of the entire construction process of the three-dimensional model.

Description

Technical Field

[0001] The present invention relates to the technical field of model construction, and in particular, to a three-dimensional model generation method and system based on a lightweight glass container. Background Art

[0002] In the data acquisition stage, traditional three-dimensional modeling methods often rely on single-modal data sources (such as single-view images or basic depth information), and it is difficult to fully capture the complex curvature changes and thin-wall structures on the surface of glass containers. Especially in positions with obvious local geometric features such as bottlenecks, bottle shoulders, and bottle bottoms, where reflection and occlusion are likely to occur, the point cloud data often shows missing and degraded phenomena, resulting in limited subsequent model accuracy. Secondly, existing mesh reconstruction and optimization methods are mostly based on general algorithms, lacking a modeling mechanism for the regional geometric characteristics (such as non-uniform surface structures) and error accumulation characteristics of glass containers. It is often difficult to balance accuracy and complexity, and the generated mesh models have problems such as topological breaks, large edge folding errors, and frequent non-manifold structures, which limit the usability and stability of the models in industrial applications. In addition, in the current modeling process, the processing of model materials mostly stays at the level of unified texture mapping or rough assignment, and fails to express them specifically according to the optical properties of different structural regions, resulting in insufficient realism in the simulation, evaluation, or rendering links of the model. Especially in simulating the transparency, glossiness, and texture details of the glass surface, the accuracy is limited. Moreover, existing systems mostly lack a dynamic evaluation mechanism for the modeling process, and cannot effectively monitor and feedback problems such as point cloud missing, mesh degradation, and error diffusion during modeling, making it difficult to form a traceable and tunable closed-loop modeling process, thereby reducing the adaptability and robustness of the model under multi-scenario deployment. Summary of the Invention

[0003] Based on this, it is necessary to provide a three-dimensional model generation method and system based on a lightweight glass container to solve at least one of the above technical problems.

[0004] To achieve the above object, a three-dimensional model generation method based on a lightweight glass container, the method includes the following steps:

[0005] Step S1: Use a preset multi-modal scanning sensor to collect information of the lightweight glass container, collect multi-view visible light images, depth information, and structured light projection data, and construct a standard multi-view point cloud data set;

[0006] Step S2: Perform multi-modal fusion based on the standard multi-view point cloud data set to obtain multi-modal fusion data of the glass container; perform thin-wall judgment on the multi-modal fusion data of the glass container, and perform thin-wall structure repair to obtain a high-integrity three-dimensional point cloud model;

[0007] Step S3: Perform lightweight mesh optimization based on the high-integrity 3D point cloud model to obtain a 3D mesh model with low complexity; perform global error compensation for container casting and melting on the low-complexity 3D mesh model to obtain container casting and melting compensation data; optimize the glass contour material of the low-complexity 3D mesh model to obtain glass container curved surface block feature data; use the container casting and melting compensation data and the glass container curved surface block feature data to perform model lightweight optimization on the low-complexity 3D mesh model to obtain a high-precision and lightweight 3D model of the glass container;

[0008] Step S4: Perform glass surface texture evaluation based on the high-precision and lightweight 3D model of the glass container to obtain glass model texture evaluation data; perform model generation process evaluation based on the high-precision and lightweight 3D model of the glass container to obtain glass model generation process evaluation data; construct a full-process path for the glass model texture evaluation data and the glass model generation process evaluation data and mark weak points to obtain a 3D model construction report of the glass container.

[0009] The beneficial effects of the present invention are as follows. By deploying a multi-modal scanning system to collect multi-view visible light images, depth information, and structured light projection data, a standard multi-view point cloud dataset is constructed, enabling rich three-dimensional information to be obtained, which lays a foundation for subsequent data processing and model construction. Based on the standard multi-view point cloud dataset, multi-modal data fusion and thin-walled structure repair are carried out, which can effectively eliminate the differences and errors between data, reduce the data inconsistency problems caused by scanning accuracy limitations, and at the same time repair the thin-walled structure area, thereby improving the integrity of the point cloud model and providing more accurate and reliable basic data for the next step of mesh optimization and refinement processing. In step S3, by performing lightweight mesh optimization on the high-integrity three-dimensional point cloud model, a three-dimensional mesh model with low complexity is generated, effectively reducing the computational complexity of the model and improving the processing speed and efficiency of the model. Further global error compensation and material optimization make the finally generated three-dimensional model of the glass container have high precision, ensuring the accurate simulation of physical properties and appearance in practical applications. In step S4, by evaluating the surface texture and the model generation process of the high-precision and lightweight three-dimensional model of the glass container, the texture details of the model and potential errors in the generation process can be analyzed in detail in a data-driven manner, obtaining glass model texture evaluation data and generation process evaluation data. These evaluation data play an important role in the full-process path construction and weak point marking links, can locate the weak areas or errors in the model, and then generate a detailed three-dimensional model construction report of the glass container, providing strong data support for subsequent model optimization. Generally speaking, through the effective fusion and evaluation optimization of multi-modal data, this technical process improves the accuracy and reliability of the three-dimensional model, ensures the high-quality modeling and precise application of the glass container, and especially provides an important basis for the optimization of surface texture, geometric structure, and material properties. Therefore, the present invention solves the problems of low structural reconstruction accuracy, large errors in transparent materials, and lack of process evaluation in traditional glass container modeling through multi-modal data fusion, combined geometric and optical optimization, and multi-scale path evaluation mechanisms, improving the geometric fidelity, material restoration degree, and controllability of the entire construction process of the three-dimensional model.

[0010] Preferably, step S1 includes the following steps:

[0011] Step S11: Deploy a high-resolution visible light camera with a resolution greater than 20 million pixels and a frame rate of 30 fps and a structured light projector with a wavelength of 450 - 650 nm and a fringe density of 0.1 mm / line on the circular scanning track to collect multi-angle visible light image data and multi-angle structured light projection data at a spatial sampling interval of 0.5 mm;

[0012] Step S12: Deploy a ToF depth sensor with an accuracy of 0.1 mm and a sampling frequency of 20 Hz in the scanning system, collect the depth information of the glass container, and obtain the glass projection distortion parameters;

[0013] Step S13: Perform highlight suppression on the multi-angle visible light image data and multi-angle structured light projection data, and use the glass projection distortion parameters for refraction compensation to obtain a glass noise-reduced multi-source dataset;

[0014] Step S14: Align the glass noise-reduced multi-source dataset in space to generate a standard multi-viewpoint cloud dataset.

[0015] In the present invention, by deploying a high-resolution visible light camera and a structured light projector, and collecting multi-angle visible light images and structured light projection data with a small spatial sampling interval (0.5 mm), more detailed and high-quality 3D data can be obtained, ensuring the complementarity and comprehensiveness of the information obtained from different perspectives. These high-precision data provide a rich source of raw data for subsequent 3D modeling and repair. In step S12, the deployment of the ToF depth sensor ensures the high-precision acquisition of depth data, with an accuracy of 0.1 mm and a sampling frequency of 20 Hz, which provides a more accurate measurement of the depth information of the glass container, further improving the spatial resolution and geometric accuracy of the model. In addition, the glass projection distortion parameters provided by the depth sensor play an important role in subsequent refraction compensation, which can effectively reduce the errors caused by the refraction of the glass surface and ensure the accuracy of the point cloud data. In step S13, by performing highlight suppression on the multi-angle image data and structured light data, the noise caused by ambient light interference or surface reflection is further reduced, improving the data quality. At the same time, refraction compensation is performed with the help of the glass projection distortion parameters, effectively correcting the geometric distortion caused by optical refraction, making the multi-source dataset more real and accurate. Finally, in step S14, by aligning the noise-reduced dataset in space, a standard multi-viewpoint cloud dataset is generated, ensuring the spatial consistency and accurate alignment of the data, laying a solid foundation for subsequent 3D reconstruction and analysis. Generally speaking, this series of steps greatly improve the data quality through high-precision data acquisition, noise suppression, and distortion compensation, providing accurate and reliable basic data for subsequent 3D modeling, repair, and optimization, thus ensuring the efficiency and accuracy of the entire system in the glass container modeling process.

[0016] Preferably, the multi-modal data fusion based on the standard multi-viewpoint cloud dataset in step S2 includes:

[0017] Obtain a glass material parameter library;

[0018] Use the glass material parameter library to simulate the refraction, reflection, and scattering behaviors of light on the glass surface to generate an optical property mapping table;

[0019] Perform point binding between the optical property mapping table and the standard multi-view point cloud dataset to generate physically enhanced point cloud data;

[0020] Input the physically enhanced point cloud data into a differentiable renderer to calculate the L1 loss and SSIM loss between the rendered image and the real visible light image, and generate multi-modal fusion data for the glass container, where the weights of the L1 loss and the SSIM loss are 0.7 and 0.3 respectively.

[0021] Through physical enhancement and image calibration steps, the present invention significantly improves the accuracy and realism of 3D modeling of glass containers. In the process of obtaining the glass material parameter library and using it to simulate the refraction, reflection, and scattering behaviors of light on the glass surface, the optical properties of the glass material can be accurately captured, providing a sufficient physical basis for the subsequent generation of the optical property mapping table. The optical property mapping table provides important optical parameters for accurate 3D modeling by detailing the interaction between light and the glass surface, ensuring that the interaction between light and the glass surface can be fully simulated during the rendering process, reflecting the complex optical behaviors of the glass surface in the real world. When performing point binding between the optical property mapping table and the standard multi-view point cloud dataset, through this precise combination, a more abundant and physically based point cloud dataset can be obtained, making the generated point cloud data not only accurate in spatial structure but also more realistic in physical properties. This physically enhanced point cloud data is further input into a differentiable renderer to compare the rendered image with the real visible light image. By calculating the L1 loss and the SSIM loss, not only the structural and detailed similarity between the rendered image and the real image is ensured, but also the weights of the loss functions (0.7 and 0.3) are adjusted to balance different loss types, further optimizing the quality of the rendering result. The combination of the L1 loss and the SSIM loss ensures that the pixel differences and structural similarities of the image are balanced, enabling the finally generated multi-modal fusion data of the glass container to take into account both the detail restoration and structural stability of the image, significantly improving the accuracy and realism of the point cloud data. This process not only improves the influence of optical properties on the 3D model but also effectively enhances the data quality during the rendering process, thus providing higher-quality data support for subsequent 3D modeling and virtual reality applications.

[0022] Preferably, the thin-wall structure repair in step S2 includes:

[0023] Predict the thin-wall area less than 2 mm based on the multi-modal fusion data of the glass container to generate the predicted thickness distribution of the glass;

[0024] The predicted glass thickness distribution is input as a constraint condition into the Poisson surface reconstruction algorithm, where the depth of the Poisson surface reconstruction algorithm is 10 and the weight is 0.5, to obtain the glass curved surface grid vertex density data;

[0025] Iteratively Poisson reconstruct the parameters of the glass curved surface grid vertex density data and construct a three-dimensional grid model to obtain a high-integrity three-dimensional point cloud model.

[0026] In the present invention, by accurately predicting and optimizing the glass thickness distribution, combining the Poisson surface reconstruction algorithm and three-dimensional grid construction, the three-dimensional modeling accuracy and reliability of glass containers are significantly improved. First, through the prediction of the thin-walled area (less than 2 mm) based on the multi-modal fusion data of the glass container, the weak parts of the glass container can be accurately identified and quantified, and a detailed predicted glass thickness distribution is generated. This thickness distribution data not only provides an effective description of the thin-walled area, but also provides important physical constraint conditions for subsequent modeling and surface reconstruction. On this basis, inputting the predicted thickness distribution data as a constraint condition into the Poisson surface reconstruction algorithm can guide the algorithm to fully consider the actual thickness change when constructing the glass surface, thus avoiding modeling errors caused by missing surface details or unreasonable surface morphology. During the Poisson surface reconstruction process, setting the algorithm depth to 10 and the weight to 0.5 enables the algorithm to achieve a balance between details and global information, generating glass curved surface grid vertex density data that conforms to physical laws. Iteratively Poisson reconstructing the parameters further optimizes the grid construction process. By gradually adjusting the grid density and vertex position, the smoothness and continuity of the curved surface are ensured, and irregular shapes and surface inconsistencies during the modeling process are avoided. Finally, this series of optimization steps enables the three-dimensional grid model to have higher integrity and accuracy, generating a high-quality three-dimensional point cloud model that not only retains the fine structure and physical properties of the glass container, but also provides a solid foundation for subsequent model analysis, application, and precision machining. This process effectively improves the modeling accuracy and provides more realistic and reliable three-dimensional data for the design and analysis of glass containers in practical applications.

[0027] Preferably, the lightweight grid optimization based on the high-integrity three-dimensional point cloud model in step S3 includes:

[0028] Perform Gaussian curvature threshold judgment on the high-integrity three-dimensional point cloud model and conduct sensitivity analysis to obtain the sensitivity analysis data of the three-dimensional point cloud model, where the Gaussian curvature threshold is less than 0.01 / ; Use the sensitivity analysis data of the three-dimensional point cloud model to identify the notch-seam area and mark it as physical property exemption data;

[0029] Based on the physical property exemption data, perform grid deletion on the high-integrity three-dimensional point cloud model to obtain a grid area deletion model, where the deletion rate of grid deletion is 20% - 50%;

[0030] Perform mesh topology optimization on the mesh region deletion model to obtain a three-dimensional mesh model with low complexity.

[0031] The present invention effectively improves the processing efficiency and accuracy of the three-dimensional point cloud model, while reducing the consumption of computing and storage resources and optimizing the subsequent modeling and analysis processes through Gaussian curvature analysis, sensitivity analysis, notch-seam region identification, mesh deletion, and topology optimization. First, by performing Gaussian curvature threshold judgment on the high-integrity three-dimensional point cloud model, regions with small surface curvature changes can be effectively identified. These regions usually correspond to flat or nearly flat parts of the geometric structure. Setting a Gaussian curvature threshold less than 0.01 / helps to distinguish regions with significant detail changes from relatively flat regions, thereby reducing excessive attention to unimportant regions in the subsequent processing. Sensitivity analysis further refines the understanding of the performance of the three-dimensional point cloud model under different conditions, helping to identify parts of the model that are more sensitive to changes in physical properties. The generated sensitivity analysis data provides a clear basis for subsequent model optimization. On this basis, by utilizing the sensitivity analysis data, notch-seam regions can be identified. These regions are marked as physical property exemption data, meaning they do not need to be further processed or optimized. This helps to reduce unnecessary calculations and optimizations and improve the overall processing efficiency. Next, based on the physical property exemption data, the three-dimensional point cloud model is subjected to mesh deletion. When the deletion rate is controlled within the range of 20%-50%, the complexity of the model can be significantly reduced without significantly losing the key features of the model. Finally, after mesh deletion, mesh topology optimization is continued. By adjusting the connectivity and stability of the mesh structure, the processing efficiency and computing performance of the model are further improved. The finally obtained three-dimensional mesh model with low complexity retains the key geometric information of the model and reduces the computing and storage requirements, which is of great significance for subsequent visualization, analysis, and applications. Through this series of optimization steps, the accuracy, efficiency, and operability of the three-dimensional point cloud model in practical applications are ultimately improved, ensuring stability and effectiveness in complex data processing.

[0032] Preferably, the mesh topology optimization of the mesh region deletion model in step S3 includes:

[0033] Perform edge collapse error optimization on the mesh region deletion model to obtain edge collapse optimization data;

[0034] Perform mesh manifold integrity verification based on the edge collapse optimization data to obtain glass manifold integrity data;

[0035] Utilize the glass manifold integrity data to construct a mesh topology structure, and verify non-manifold edges and isolated vertices using Euler's formula to obtain a glass container mesh topology structure;

[0036] The mesh topology structure and mesh area deletion model of the glass container are optimized with low complexity to obtain a low-complexity three-dimensional mesh model.

[0037] The present invention improves the geometric topological stability and overall structural integrity of the three-dimensional mesh model through edge folding error optimization and manifold verification mechanisms, and at the same time significantly reduces the number of facets and data complexity of the model through low-complexity optimization means. At the data level, the mesh area deletion model is first subjected to edge folding error optimization operations, that is, while keeping the geometric shape basically unchanged, the edge pairs with the smallest error are selected for folding processing, thereby reducing redundant edges and faces and improving mesh simplification efficiency. Each foldable edge is evaluated by an error function, the folding priority is selected, and edge folding optimization data is generated, which contains local topological change information corresponding to the edge folding and its corresponding error value. Next, the mesh manifold integrity is verified based on the edge folding optimization data to ensure that the edge folding will not destroy the manifold structure of the original mesh, thereby generating glass manifold integrity data, which is used to record the integrity status of the mesh in terms of topological structure continuity, normal consistency, and edge-face-vertex relationship. On the basis of verifying the manifold, the integrity data is used to construct the mesh topology structure, and the Euler formula V−E+F=2 is introduced to screen and correct non-manifold edges and isolated vertices, further ensuring that the final generated glass container mesh topology structure meets the geometric modeling constraint requirements in terms of topological closure and structural connectivity. Subsequently, the deleted mesh region model is combined with the above mesh topology structure for low-complexity optimization. Facets are merged, vertices are relocated, and triangles are reconstructed while maintaining the main geometric features of the model, significantly reducing the total number of vertices, edges, and faces, and generating a low-complexity 3D mesh model with small data volume, low redundancy, and clear structure. From error data extraction to topology verification to low-complexity generation, this process realizes the effective integration of 3D geometric model structure optimization, data compression, and precision control, providing a high-precision, lightweight data foundation for subsequent rendering, simulation, or analysis.

[0038] Preferably, step S3 of performing global error compensation for container casting and melting on the low-complexity three-dimensional mesh model comprises:

[0039] Obtaining actual glass container measurement data;

[0040] Extracting geometric nodes of glass container based on low-complexity 3D mesh model;

[0041] Node error analysis is performed using the glass container geometric node extraction and measured glass container data to obtain glass container node comparison analysis data;

[0042] Deviation compensation is performed based on the comparative analysis data of the glass container nodes to obtain the container casting and melting compensation data.

[0043] Use the container casting and melting compensation data to optimize the material of a low-complexity three-dimensional mesh model, and obtain a high-precision and lightweight three-dimensional model of a glass container.

[0044] Through the quantitative comparative analysis of geometric node errors and the deviation compensation mechanism, the present invention improves the performance of the three-dimensional mesh model in terms of geometric accuracy and material mapping consistency, and finally generates a three-dimensional model of a glass container with high reduction degree and low complexity. Specifically, first, the measured data of the actual glass container is generally obtained from laser scanning, structured light or coordinate measuring equipment, and its data format exists in the form of a point coordinate set or a boundary feature set, serving as a real physical benchmark. Subsequently, by extracting geometric nodes from the low-complexity three-dimensional mesh model, the topological positions and spatial relationship information of the key structural points of the model are obtained. By performing point-by-point matching and spatial difference operations on this geometric node data and the measured data, a set of node error vectors can be formed, constituting the node comparative analysis data of the glass container. This data includes error indicators such as the Euclidean distance, normal offset angle, and adjacent side length deviation between each key node and the measured point in three-dimensional space. According to the error vector, perform the inverse transformation and interpolation adjustment of the spatial position to construct a compensation offset matrix, thereby completing the generation of the container casting and melting compensation data. The error equalization data not only includes geometric correction vectors, but also realizes the local limitation of the error propagation range and the boundary reconstruction of the material mapping area by integrating with the original mesh structure. On this basis, further use the error equalization data to optimize the material of the low-complexity three-dimensional mesh model, and assign different material attribute parameters according to the error residual intensity of each region, so as to avoid the inconsistency of lighting, reflection or transparency caused by geometric deviation. This optimization process comprehensively considers the structure matching accuracy, material parameter distribution and visual authenticity, so that the finally constructed high-precision and lightweight three-dimensional model of the glass container has good spatial reconstruction accuracy and material performance consistency while retaining the low-face-number structure, significantly enhancing the engineering adaptability and subsequent visualization ability of the model.

[0045] Preferably, the material optimization of the glass contour of the low-complexity three-dimensional mesh model includes:

[0046] Use the container casting and melting compensation data to perform surface block division on the bottleneck area, bottle shoulder area, bottle body area and bottle bottom area to obtain the surface block division characteristic data of the glass container; perform mesh model alignment on the surface block division characteristic data of the glass container to obtain a high-precision and lightweight three-dimensional model of the glass container, where the surface block division characteristic data of the glass container includes bottleneck-shoulder material optimization data, bottle body material optimization data and bottle bottom material optimization data;

[0047] Perform texture material mapping on the bottleneck area and the bottle shoulder area to obtain the bottleneck-shoulder material optimization data;

[0048] Perform glossiness material mapping on the bottle body area to obtain bottle body material optimization data;

[0049] Perform transparency material mapping on the bottle bottom area to obtain bottle bottom material optimization data.

[0050] In the present invention, the overall model is subjected to regional surface partitioning through container casting and melting compensation data, enabling heterogeneous material processing based on the degree of local geometric perturbation, and providing fine structural boundary conditions for subsequent multi-dimensional material assignment. Key parameters such as geometric continuity, curvature change rate, and normal direction discreteness extracted during the surface partitioning process are uniformly encoded as glass container surface partitioning feature data, and spatially aligned with the original mesh structure, effectively avoiding local misalignment or reconstruction errors in the model topology due to material mapping. Secondly, based on the feature partitioning data, texture mapping in the bottleneck-shoulder area, glossiness mapping in the bottle body area, and transparency modeling in the bottle bottom area are implemented. Region-specific material weight matrices reflecting micro-surface roughness, surface reflection coefficient, and incident light scattering intensity can be constructed respectively at the data level. These matrices are bound to the partitioning grid nodes of the corresponding regions to generate bottleneck-shoulder material optimization data, bottle body material optimization data, and bottle bottom material optimization data, each having the local expression ability for visual texture, light reflection, and light transmission performance. In the data integration stage, the material optimization data of each region undergoes a mesh model alignment operation to achieve boundary stitching and error smoothing compensation between different material mappings, while ensuring the consistency of material texture directionality and mesh manifold topology. The finally constructed high-precision and lightweight three-dimensional model of the glass container presents a realistic material response under various observation conditions while retaining the low-complexity characteristics of the original mesh, significantly enhancing the data adaptation ability and expression accuracy of the model in structural light simulation, optical tracking analysis, and virtual-real hybrid visualization applications.

[0051] Preferably, step S4 includes the following steps:

[0052] Step S41: Evaluate the glass surface texture based on the high-precision and lightweight three-dimensional model of the glass container to obtain glass model texture evaluation data;

[0053] Step S42: Evaluate the model generation process based on the high-precision and lightweight three-dimensional model of the glass container to obtain glass model generation process evaluation data;

[0054] Step S43: Perform two-dimensional curve evaluation and fitting analysis on the glass model texture evaluation data and the glass model generation process evaluation data to obtain glass model construction fitting data; Construct the path of the glass model construction fitting data and mark the weak points to obtain the three-dimensional model construction report of the glass container.

[0055] The present invention evaluates the glass surface texture based on a high-precision and lightweight three-dimensional model of a glass container, and can extract texture features such as surface roughness distribution, normal fluctuation frequency, and optical reflection non-uniformity from the mesh patch unit, and map these features into a high-dimensional texture vector space, thereby generating glass model texture evaluation data to reflect the coverage of the actual surface microstructure expression capability during the modeling process. At the same time, the model generation process evaluation quantitatively analyzes the error transmission path in the model building process by tracking the parameter trajectory formed by each key data node in the model generation chain (such as point cloud fusion density, curvature fitting residual, topology simplification intensity, etc.), and finally generates glass model generation process evaluation data. The above two types of heterogeneous evaluation data are fused through two-dimensional curve evaluation matching analysis, which can establish a cross-mapping between texture quality and generation process in the data space, and extract the section with local mutation or consistency deviation in the matching function curve, that is, the glass model construction matching data, which reflects the quantitative consistency between geometric expression capability and modeling reliability. The path is further constructed based on the matching data, and the geometric partition information of the model and the parameter sensitivity results are combined to identify the areas with drastic texture changes or obvious error amplification, automatically mark them as weak points, and finally output the glass container 3D model construction report. This report not only provides a traceable modeling process indicator sequence at the data level, but also embeds the analysis results of the coupling of texture expression and construction logic, providing a highly reliable data basis for subsequent structural defect prediction, virtual-real mapping rendering, and reverse correction of manufacturing parameters.

[0056] The multimodal sensing and acquisition module is used to collect information about lightweight glass containers using a preset multimodal scanning sensor, collect multi-view visible light images, depth information, and structured light projection data, and construct a standard multi-view point cloud data set;

[0057] The 3D point cloud fusion and repair module is used to perform multimodal fusion based on the standard multi-view point cloud data set to obtain multimodal fusion data of the glass container; perform thin-wall judgment on the multimodal fusion data of the glass container and perform thin-wall structure repair to obtain a high-integrity 3D point cloud model;

[0058] The model lightweight optimization and material enhancement module is used to perform lightweight mesh optimization based on a high-integrity 3D point cloud model to obtain a low-complexity 3D mesh model; perform global error compensation for container casting and melting on the low-complexity 3D mesh model to obtain container casting and melting compensation data; optimize the glass contour material of the low-complexity 3D mesh model to obtain glass container surface block feature data; use the container casting and melting compensation data and glass container surface block feature data to perform model lightweight optimization on the low-complexity 3D mesh model to obtain a high-precision and lightweight 3D model of the glass container;

[0059] The evaluation analysis and report generation module is used to evaluate the glass surface texture based on the high-precision and lightweight three-dimensional model of the glass container to obtain the texture evaluation data of the glass model; evaluate the model generation process based on the high-precision and lightweight three-dimensional model of the glass container to obtain the evaluation data of the glass model generation process; construct a full-process path for the texture evaluation data of the glass model and the evaluation data of the glass model generation process, and mark the weak points to obtain the three-dimensional model construction report of the glass container.

[0060] Through the joint acquisition of multi-view visible light images, depth information and structured light projection data by the multi-modal perception acquisition module of this system, the spatial coverage rate and geometric reduction accuracy of the point cloud data are significantly improved. Especially in dealing with the problem of surface information loss caused by factors such as reflection, light transmission, and occlusion of glass containers, a data system with complementary structures can be formed to construct a more complete and detailed multi-view point cloud data set. Secondly, after the three-dimensional point cloud fusion and repair module fuses different modal data, it judges the geometric consistency based on the common thin-wall regions in the glass material structure and executes the repair algorithm, effectively filling the sparse fragments caused by edges, depressions and transparent regions at the point cloud level, making the generated high-integrity three-dimensional point cloud model have stronger morphological continuity and data connectivity, providing a solid foundation for subsequent mesh construction. On this basis, the model lightweight optimization and material enhancement module constructs a three-dimensional mesh model with low complexity through topology reconstruction, patch simplification and regional partitioning methods, and superimposes the container casting and melting error compensation mechanism, significantly reducing the geometric deviation caused by casting errors or scanning perturbations. At the same time, combined with the material feature differences in key regions such as the bottleneck, bottle shoulder, bottle body and bottle bottom, local surface optimization and attribute enhancement are carried out to generate surface partition feature data including local texture, glossiness, transparency, etc., and finally output a high-precision and lightweight three-dimensional model with both visual authenticity and geometric accuracy. The evaluation analysis and report generation module further conducts a closed-loop verification of the model quality at the data level, constructs an evaluation path through indicators such as texture contrast, reflectivity distribution and error suppression efficiency, and locates the weak point regions in the construction process to form a three-dimensional model construction report of the glass container with multi-dimensional evaluation indicators, providing visual data basis and accurate feedback for subsequent manufacturing, simulation or quality control. Brief Description of the Drawings

[0061] Figure 1 It is a schematic diagram of the step flow of a three-dimensional model generation method based on a lightweight glass container;

[0062] Figure 2 It is Figure 1 a schematic diagram of the detailed implementation step flow of step S4 in

[0063] The realization, functional features and advantages of the object of the present invention will be further described in conjunction with the embodiments with reference to the accompanying drawings. Detailed Embodiments

[0064] The technical method of the present invention patent will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments in the present invention without creative efforts belong to the scope of protection of the present invention.

[0065] In addition, the accompanying drawings are only schematic diagrams of the present invention and are not necessarily drawn to scale. The same reference numerals in the drawings represent the same or similar parts, and thus their repeated description will be omitted. Some of the block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0066] It should be understood that although terms such as "first" and "second" may be used here to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, the first unit may be called the second unit, and similarly the second unit may be called the first unit. The term "and / or" used here includes any and all combinations of one or more of the listed associated items.

[0067] To achieve the above object, please refer to Figures 1 to 2 , a three-dimensional model generation method based on a lightweight glass container, the method comprising the following steps:

[0068] Step S1: Use a preset multi-modal scanning sensor to collect information on the lightweight glass container, collect multi-view visible light images, depth information, and structured light projection data, and construct a standard multi-view point cloud data set;

[0069] Step S2: Perform multi-modal fusion based on the standard multi-view point cloud data set to obtain multi-modal fusion data of the glass container; perform a thin-wall judgment on the multi-modal fusion data of the glass container and perform thin-wall structure repair to obtain a high-integrity three-dimensional point cloud model;

[0070] Step S3: Perform lightweight mesh optimization based on the high-integrity 3D point cloud model to obtain a 3D mesh model with low complexity; perform global error compensation for container casting and melting on the 3D mesh model with low complexity to obtain container casting and melting compensation data; optimize the glass contour material of the 3D mesh model with low complexity to obtain glass container curved surface block feature data; use the container casting and melting compensation data and the glass container curved surface block feature data to perform model lightweight optimization on the 3D mesh model with low complexity to obtain a high-precision and lightweight 3D model of the glass container;

[0071] Step S4: Perform glass surface texture evaluation based on the high-precision and lightweight 3D model of the glass container to obtain glass model texture evaluation data; perform model generation process evaluation based on the high-precision and lightweight 3D model of the glass container to obtain glass model generation process evaluation data; construct a full-process path for the glass model texture evaluation data and the glass model generation process evaluation data, and perform weak point marking to obtain a glass container 3D model construction report.

[0072] In the embodiment of the present invention, refer to Figure 1 As shown, it is a schematic diagram of the step flow of a method for constructing an online industrial analysis big data platform of the present invention. In this example, the method for constructing the online industrial analysis big data platform includes the following steps:

[0073] Step S1: Use a preset multi-modal scanning sensor to collect information on a lightweight glass container, collect multi-view visible light images, depth information, and structured light projection data, and construct a standard multi-view point cloud data set;

[0074] In the embodiments of the present invention, the core technology involved lies in multi-modal three-dimensional perception and dataset construction. The specific process first depends on high-precision and multi-view data collection of the surface of lightweight glass containers. To achieve this goal, the system deploys multi-modal scanning devices, including high-resolution visible light cameras, active depth cameras, and structured light projection devices. By setting a multi-view rotation platform or a multi-camera synchronous acquisition architecture, the system can synchronously obtain the response information of the container surface under RGB images, depth maps, and structured light interference fringe maps from different directions. RGB images are used to obtain visual features such as surface color and texture, depth maps capture surface spatial topography information, and structured light fringe maps are used to improve the stability and robustness of depth data extraction on smooth, transparent, or reflective material surfaces. The collected raw data needs to be multi-view aligned through a spatial registration algorithm. Typical methods include solving internal and external parameters based on a calibration board, feature point matching such as SIFT / SURF, and fine registration based on the Iterative Closest Point (ICP) algorithm. Subsequently, the data from all views is reconstructed into a three-dimensional point cloud in a unified coordinate system. To ensure the point cloud density and coverage, reasonable angular intervals and overlap rates need to be set during the acquisition stage, and operations such as noise filtering, edge correction, and redundant point removal are performed on the acquired data, thereby generating a standard multi-view point cloud dataset with a complete structure, uniform density, and consistent coordinates.

[0075] Step S2: Perform multi-modal fusion based on the standard multi-view point cloud dataset to obtain the multi-modal fusion data of the glass container; judge the thin wall of the multi-modal fusion data of the glass container, and perform thin wall structure repair to obtain a three-dimensional point cloud model with high integrity;

[0076] In the embodiments of the present invention, the key technology involved is to perform multi-modal data fusion and fine repair of thin-walled structures based on a standard multi-viewpoint cloud data set. The core task is to improve the structural integrity and geometric continuity of the 3D point cloud model. First, in the multi-modal fusion stage, it is necessary to perform pixel-by-pixel or point-by-point feature mapping of the texture information contained in the RGB image, the distance data in the depth map, and the optical coding results obtained by structured light projection with the aligned point cloud data. In specific operations, a projection-backprojection model is used to transform the image information of each view into the 3D space coordinate system through the camera intrinsics and pose matrix to achieve the fusion of color-geometry-photometric features. After the preliminary fusion, redundant points, outliers, and void regions caused by optical interference in the fusion results are processed. Commonly used algorithms include Statistical Outlier Removal and Voxel Grid Filtering, etc., to improve data consistency and uniformity. For the thin-walled structure part, point cloud missing usually occurs due to light transmittance, drastic curvature changes, or structural occlusion, and surface reconstruction technology needs to be used for complementation. During the reconstruction process, boundary matching, topological relationship reconstruction, and surface normal estimation between multi-viewpoint clouds are combined to construct a continuous surface; at the same time, interpolation methods such as Poisson Surface Reconstruction or Moving Least Squares (MLS) are introduced to perform high-fitting repair on the missing regions to ensure the complete expression of the reconstructed surface in terms of boundary continuity, normal consistency, and geometric stability.

[0077] Step S3: Perform lightweight grid optimization based on the high-integrity 3D point cloud model to obtain a low-complexity 3D grid model; perform global error compensation for container casting and melting on the low-complexity 3D grid model to obtain container casting and melting compensation data; optimize the glass contour material of the low-complexity 3D grid model to obtain glass container surface block feature data; use the container casting and melting compensation data and the glass container surface block feature data to perform model lightweight optimization on the low-complexity 3D grid model to obtain a high-precision-lightweight 3D model of the glass container;

[0078] In the embodiments of the present invention, the dense point cloud data output in the previous stage is input into a triangular mesh reconstruction algorithm. Common methods include Poisson Surface Reconstruction or Delaunay triangulation technology. By estimating the point cloud normal vector field and constructing a closed surface, the conversion of the original point cloud into a structured triangular mesh is realized. Subsequently, to reduce redundant patches and topological complexity, mesh simplification processing is required, that is, on the basis of maintaining the boundary shape, key feature lines, and local curvature changes, the number of patches is reduced. Common simplification methods include the Quadric Error Metrics (QEM) algorithm and the Edge Collapse strategy. The simplification process controls the patch merging order by defining an error cost function to minimize the global complexity. After completing the mesh simplification, a global error analysis of the geometric deviation introduced by the simplification is performed. The point-to-plane distance or Hausdorff distance between the source point cloud and the mesh model is used as an evaluation index, and error compensation is achieved through surface fine-tuning and local optimization (such as Laplacian smoothing or edge-preserving smoothing algorithms). In the material optimization stage, based on the initial RGB image data and the camera projection matrix, color textures are mapped onto the surface of the triangular mesh by methods such as Projective Texture Mapping or Multi-view Texture Blending to solve problems such as uneven illumination and projection distortion, and enhance the continuity and realism of the surface material. The finally output model has low complexity in the spatial structure and high resolution and consistency in the material expression, constituting a high-precision and lightweight three-dimensional representation of the glass container.

[0079] Step S4: Perform glass surface texture evaluation based on the high-precision and lightweight three-dimensional model of the glass container to obtain glass model texture evaluation data; perform model generation process evaluation based on the high-precision and lightweight three-dimensional model of the glass container to obtain glass model generation process evaluation data; construct a full-process path for the glass model texture evaluation data and the glass model generation process evaluation data, and mark weak points to obtain a glass container three-dimensional model construction report.

[0080] In the embodiments of the present invention, a data-driven texture quality analysis and modeling process traceability are performed on the generated high-precision and lightweight three-dimensional model of the glass container, and a full-process path and structural weak point marking mechanism is constructed based on the evaluation results. First, in the texture evaluation stage, taking the texture image mapped on the surface of the mesh model as the basic data, texture quality indicators in multiple dimensions such as color consistency, texture continuity, illumination balance, and local noise level are extracted. The specific methods include using the gray-level co-occurrence matrix (GLCM) to extract texture statistical features such as contrast, energy, correlation, and homogeneity, and at the same time combining a multi-scale edge detection algorithm to detect texture breakage and discontinuous regions, forming a texture defect annotation map, and mapping and comparing it with the original RGB image to extract the image-model difference quantization index and generate glass surface texture evaluation data. Secondly, in the model generation process evaluation link, the geometric deviation propagation in each stage of the entire three-dimensional modeling process is traced and analyzed. The inputs include the initial point cloud data, simplified mesh data, and the final three-dimensional model. Through multi-stage registration error analysis and Hausdorff distance evaluation, the position and magnitude of error accumulation in the modeling process are determined. At the same time, the process log and metadata are used to trace the change of point cloud density, the response of the simplification threshold, and the change of texture mapping accuracy, and the modeling process evaluation data is output. After obtaining the two types of evaluation data, by constructing a process path map based on the two-way indexing of the time axis and spatial coordinates, the data processing link is integrated in time sequence, and combined with the spatial error heat map analysis, the high-error regions and texture abnormal aggregation regions are marked on the surface of the three-dimensional model to realize the spatial positioning of weak points. Finally, the evaluation data, path map, and annotation information are integrated into a structured model construction report, and various quality indicators and traceability analysis results are output in a standardized format.

[0081] Preferably, step S1 includes the following steps:

[0082] Step S11: Deploy a high-resolution visible light camera with a resolution greater than 20 million pixels and a frame rate of 30 fps and a structured light projector with a wavelength of 450 - 650 nm and a fringe density of 0.1 mm / line on the circular scanning orbit, and collect multi-angle visible light image data and multi-angle structured light projection data at a spatial sampling interval of 0.5 mm;

[0083] Step S12: Deploy a ToF depth sensor with an accuracy of 0.1 mm and a sampling frequency of 20 Hz in the scanning system, collect the depth information of the glass container, and obtain the glass projection distortion parameters;

[0084] Step S13: Perform highlight suppression on the multi-angle visible light image data and multi-angle structured light projection data, and use the glass projection distortion parameters for refraction compensation to obtain a glass noise reduction multi-source data set;

[0085] Step S14: Align the glass noise reduction multi-source dataset spatially to generate a standard multi-view point cloud dataset.

[0086] In the embodiments of the present invention, the high-resolution visible light image acquisition and structured light projection are collaboratively realized by a visible light camera (resolution not less than 20 million pixels, frame rate 30fps) and a structured light projector (wavelength range 450–650nm, stripe density 0.1mm / line) deployed on a circular scanning orbit. The sampling interval is set to 0.5mm to ensure complete image coverage and sufficient spatial detail resolution ability. The projection distortion of the structured light stripes formed on the glass surface will then be modeled and analyzed in the depth compensation stage. In S12, a ToF depth sensor (accuracy 0.1mm, sampling frequency 20Hz) is deployed to perform synchronous depth scanning on the target object, generating a sequence of multi-frame depth maps, extracting physical measurement data including changes in the refraction path, object surface deformation response, and glass medium transmission distortion, etc., and then constructing a refraction distortion parameter model of the glass material. In S13, for the strong specular reflection problem caused by the glass material in the visible light image and the structured light image, a specular highlight detection algorithm based on color space transformation (such as brightness threshold separation and morphological filtering in the HSV space) and an image restoration method (such as region interpolation-based or edge-preserving smoothing algorithm) are used for specular highlight suppression processing; subsequently, according to the refraction model constructed in S12, the multi-angle acquired images are refraction compensated, and methods based on ray-tracing or image re-mapping are used to correct the imaging distortion caused by the glass transparency, and a glass noise reduction multi-source dataset is output. In S14, a spatial registration technology is used to align the multi-source multi-view image data to a unified coordinate system. Usually, preliminary registration is first achieved by solving the external parameters of multiple cameras based on a calibration board, and then the Iterative Closest Point (ICP) or a point cloud registration algorithm based on feature matching is used to perform spatial fusion and error minimization processing on the depth map and the structured light contour map generated by image reconstruction, finally forming a standard multi-view point cloud dataset, laying a data foundation for subsequent three-dimensional model construction and structural analysis.

[0087] Preferably, the multi-modal data fusion based on the standard multi-view point cloud dataset in step S2 includes:

[0088] Obtain a glass material parameter library;

[0089] Use the glass material parameter library to simulate the refraction, reflection, and scattering behaviors of light on the glass surface to generate an optical property mapping table;

[0090] Perform point binding on the optical property mapping table and the standard multi-view point cloud dataset to generate physically enhanced point cloud data;

[0091] Input the physically enhanced point cloud data into a differentiable renderer to calculate the L1 loss and SSIM loss between the rendered image and the real visible light image, and generate multi-modal fusion data for the glass container. The weights of the L1 loss and SSIM loss are 0.7 and 0.3 respectively.

[0092] In the embodiments of the present invention, the technical means described in this section cover multiple key steps such as optical modeling of glass materials, binding of point clouds and physical properties, differentiable rendering, and point cloud calibration optimization based on image reconstruction errors. The core objective is to introduce real physical characteristics on the basis of a standard multi-view point cloud dataset to improve the optical consistency and spatial accuracy of 3D data. First, by obtaining a glass material parameter library, physical property data including refractive index, reflectivity, scattering coefficient, absorption coefficient, dispersion parameter, etc. are collected. These data can be obtained from actual measurements or optical simulation models. For example, based on the Fresnel equation and Rayleigh scattering theory, the propagation path and energy loss of light in different wavelength bands are modeled and analyzed. Subsequently, according to the various properties provided in the material parameter library, path tracing or Monte Carlo ray propagation simulation methods are used to simulate the refraction, reflection, and scattering behaviors of light on the glass interface, and an optical property mapping table is generated. Indexed by spatial coordinates, this mapping table records information such as the light propagation direction, intensity attenuation coefficient, and reflectivity distribution of each point under specific viewing angles and illumination conditions. Next, a point binding operation is performed between this optical mapping table and each spatial point in the standard multi-view point cloud dataset, that is, attaching its corresponding physical optical property vector to each point cloud coordinate point, thereby forming physically enhanced point cloud data. On this basis, a differentiable renderer is introduced. The physically enhanced point cloud is input into the renderer to synthesize a corresponding pseudo-image, which is compared with the actually captured visible light image at the pixel level, and the point cloud is optimized by constructing an image reconstruction loss function. The loss function consists of two parts: L1 loss (absolute error) and SSIM (structural similarity index), which measure the difference in image pixel intensity and the similarity of structural texture respectively. Their weighting ratios are set to 0.7 and 0.3, so as to retain structural consistency while ensuring brightness matching. Finally, the point cloud coordinates or additional property vectors are optimized using the backpropagation mechanism, and multi-modal fusion and calibration of the physically enhanced point cloud are realized under the differentiable renderer framework to obtain calibrated multi-modal fusion point cloud data that is more consistent with the real visual performance.

[0093] Preferably, the thin-walled structure repair in step S2 includes:

[0094] Predict the thin-walled area less than 2 mm based on the multi-modal fusion data of the glass container to generate the predicted thickness distribution of the glass.

[0095] The predicted thickness distribution of the glass is input as a constraint condition into the Poisson surface reconstruction algorithm, where the depth of the Poisson surface reconstruction algorithm is 10 and the weight is 0.5, to obtain the vertex density data of the glass curved surface grid;

[0096] Iterative Poisson reconstruction parameters are performed on the vertex density data of the glass curved surface grid, and a three-dimensional grid model is constructed to obtain a high-integrity three-dimensional point cloud model.

[0097] In the embodiment of the present invention, this technical means mainly focuses on the thickness prediction of the thin-walled region of the glass based on calibrated multi-modal fusion point cloud data, constraint-driven Poisson surface reconstruction, and iterative optimization of the grid construction process. The core lies in the integration of geometric features, physical properties, and surface reconstruction algorithms to achieve the construction of a three-dimensional point cloud model with high precision and good continuity. First, based on the calibrated fused point cloud data, the spatial thickness of the glass container surface needs to be predicted, and the region with a thickness less than 2 mm is defined as the thin-walled structure. This prediction process can be modeled through local curvature analysis, normal distribution change rate, and inter-point Euclidean distance evaluation methods, supplemented by a thickness estimation algorithm based on voxel projection or bilateral difference interpolation to form a high-resolution thickness distribution map. This predicted thickness distribution data of the glass is then used as the constraint condition input for the Poisson surface reconstruction process. Specifically, a weighted Poisson surface reconstruction algorithm (Poisson Surface Reconstruction) is adopted, with the depth parameter set to 10 to ensure that the reconstruction captures the level of detail, and the weight is set to 0.5 to achieve a balance between point cloud density and reconstruction smoothness. The algorithm constructs an octree voxel structure for solving the Poisson equation, converts the normal field of the point cloud into a scalar field gradient field, thereby inversely deducing the implicit surface function, and generating an initial surface grid from the isosurface of this function. After initially obtaining the vertex density data of the surface grid, the Poisson reconstruction parameters are further iteratively optimized, that is, on the premise of maintaining the continuity of the initial solution structure, the octree depth, interpolation radius, and regularization term weight are adjusted to minimize the geometric deviation between the actual point cloud and the reconstructed surface. The optimization process is usually guided by the local residual function convergence criterion and the grid reconstruction error feedback mechanism. Finally, through the grid connection topology reconstruction of the optimized density data, combined with the triangulation algorithm and the boundary reprojection strategy, the construction of the three-dimensional grid model is completed, forming a glass three-dimensional point cloud model with high structural integrity and good thickness continuity, providing data support for subsequent steps such as geometric compression, surface feature analysis, and material mapping.

[0098] Preferably, the lightweight grid optimization based on the high-integrity three-dimensional point cloud model in step S3 includes:

[0099] Perform Gaussian curvature threshold judgment on the high-integrity three-dimensional point cloud model, and conduct sensitivity analysis to obtain the sensitivity analysis data of the three-dimensional point cloud model, where the Gaussian curvature threshold is less than 0.01 / ; Use the sensitivity analysis data of the three-dimensional point cloud model to identify the notch-seam area and mark it as physical property exemption data;

[0100] Based on the physical property exemption data, perform mesh deletion on the high-integrity three-dimensional point cloud model to obtain a mesh area deletion model, where the deletion rate of mesh deletion is 20%-50%;

[0101] Perform mesh topology optimization on the mesh area deletion model to obtain a three-dimensional mesh model with low complexity.

[0102] In the embodiment of the present invention, this section of technical means focuses on a series of continuous operations such as local curvature feature extraction, sensitivity area identification, exemption area marking, mesh deletion, and topological structure optimization of the high-integrity three-dimensional point cloud model from the data level. The purpose is to maintain geometric stability under the premise of simplifying complex structures. First, calculate the Gaussian curvature of the three-dimensional point cloud model, and use the principal curvature derivative formula K = , where and are the maximum and minimum principal curvatures at each point. In order to avoid the interference of abnormal noise on the curvature, a local surface approximation technology based on normal fitting is adopted. Based on the point-normal pair, the curvature value is stably solved through the weighted least squares surface fitting algorithm. All regions with Gaussian curvature less than 0.01 are determined as flat curvature regions, and their response capabilities in the structure degradation or perturbation scenarios are evaluated through sensitivity analysis. A sensitivity coefficient is assigned to these regions using an evaluation method based on the distribution of the response intensity of point position perturbation. Subsequently, combined with the geometric texture change information, regions with significant sensitivity characteristics and structural mutations (such as notches or seams) are located, and these regions are marked as "physical property exemption data", meaning that they need to be retained in the subsequent deletion to maintain structural identifiability. Then, without affecting the integrity of the exemption area, perform the mesh deletion operation. The deletion strategy is regulated according to the point density redundancy and the uniformity of normal distribution, and the deletion rate is controlled between 20% and 50% of the total number of the overall mesh. During the deletion process, a patch error control function and a triangle mesh simplification algorithm (such as the QEM algorithm) are used to maintain geometric approximation. Finally, based on the mesh area deletion model after deletion, perform mesh topology optimization. Eliminate topological redundancy and improve the connectivity structure quality through operations such as edge collapse, face merging, and point repositioning, ensuring that the simplified mesh reaches a better patch distribution balance and shape expression stability while maintaining the original geometric constraints, so as to obtain a three-dimensional mesh model with low complexity, laying a data foundation for the subsequent material mapping and accuracy compensation stages

[0103] Preferably, the mesh topology optimization of the mesh region deletion model in step S3 includes:

[0104] Performing edge collapse error optimization on the mesh region deletion model to obtain edge collapse optimization data;

[0105] Performing mesh manifold integrity verification based on the edge collapse optimization data to obtain glass manifold integrity data;

[0106] Using the glass manifold integrity data to construct a mesh topology structure, and using Euler's formula to verify non-manifold edges and isolated vertices, to obtain a glass container mesh topology structure;

[0107] Performing low-complexity optimization on the glass container mesh topology structure and the mesh region deletion model to obtain a low-complexity three-dimensional mesh model.

[0108] In the embodiments of the present invention, the technical means mainly involve the geometric error control and topological consistency construction of the deleted mesh model, and specifically are carried out from four data processing links: edge collapse error measurement, mesh manifold verification, topological structure reconstruction, and complexity control. First, for the mesh region deletion model, an edge collapse optimization method based on the principle of minimizing local geometric deviation is adopted. By calculating the error quadric matrix (Quadric Error Metrics, QEM) of each candidate collapse edge, the normal vector change, angle deviation, and position fitting error between adjacent faces after edge collapse are evaluated, so as to generate edge collapse optimization data, and record the numerical impact of each folding operation on the geometric reconstruction of the original mesh surface in the local coordinate system. Subsequently, based on the edge collapse optimization data, the integrity of the manifold of the deletion model is verified. A mesh topology graph is constructed using a face-edge-point three-dimensional adjacency list structure to identify all structural nodes with non-manifold edges (shared by three or more faces) and isolated vertices (not associated with any face). And accordingly generate glass manifold integrity data. Then, use the above integrity data to reconstruct the mesh topology structure. By using depth-first search (DFS) and boundary tracking algorithms to identify connected regions and closed boundaries, ensure that the adjacency of all faces satisfies topological consistency, and introduce Euler's formula for global structure verification, and calculate the number of vertices , the number of edges , the number of faces , and combine with the Euler characteristic number Verify whether the grid is a closed polyhedron structure, detect and locate topological anomalies from a topological perspective. Finally, combine the reconstructed grid topology with the original trimmed model, and perform low-complexity optimization processing by means of simplifying the patch layout, merging coplanar triangles, and sparsifying dense vertex regions, etc., to control the number of patches and grid density to achieve sparse representation while retaining the structural boundaries, and finally form a low-complexity three-dimensional grid model with both topological closure, manifold consistency, and data structure simplicity.

[0109] Preferably, the global error compensation and material optimization for the low-complexity three-dimensional grid model in step S3 include:

[0110] Obtain the measured data of the glass container;

[0111] Extract the geometric nodes of the glass container based on the low-complexity three-dimensional grid model;

[0112] Use the geometric node extraction of the glass container and the measured data of the glass container to perform node error analysis to obtain the node comparison analysis data of the glass container;

[0113] Perform deviation compensation according to the node comparison analysis data of the glass container to obtain the container casting and melting compensation data.

[0114] In the embodiments of the present invention, the technical means mainly cover steps such as comparative analysis of measured data and low-complexity three-dimensional grid models, error compensation, and material optimization, focusing on the whole process of geometric node extraction, error analysis, deviation correction, and model accuracy improvement. First, the measured glass container measurement data obtained usually includes specific geometric information such as size, shape, and position, and is usually obtained by high-precision three-dimensional scanning or other measurement technologies (such as laser scanning, structured light, ToF sensors, etc.). Based on the low-complexity three-dimensional grid model, geometric node extraction algorithms (such as vertex extraction, boundary tracking, etc.) are used to accurately locate and extract key structural units (such as nodes, edges, faces) in the grid model. Next, by comparing the extracted geometric nodes with the measured data, node error analysis is carried out, and geometric error measurement methods such as Euclidean distance and least squares fitting are used to calculate the deviation of each geometric node in dimensions such as position, shape, and normal direction. Through these analyses, glass container node comparison analysis data is generated, which records the error distribution between the model nodes and the measured data, reflecting the difference between the model accuracy and the actual object. Based on this error data, further deviation compensation is carried out. Common methods include node position adjustment, normal direction correction, and local geometry repair by the least squares method, etc., thereby generating container casting and melting compensation data. This process corrects the model in terms of geometric structure, reduces the deviation from the measured data, and optimizes the accuracy of the three-dimensional model. Finally, the low-complexity three-dimensional grid model is optimized for materials using the error equilibrium data. The goal of material optimization is to make the material properties of the model (such as density, stiffness, texture, etc.) more in line with the physical characteristics of the actual object. The optimization process is usually carried out by means of material property mapping, optical property adjustment, and alignment with the measured data, and finally a high-precision and lightweight three-dimensional model of the glass container is obtained. This optimization not only improves the geometric accuracy of the model but also ensures the similarity of the material properties to the actual glass container, making the final model meet high-precision standards in many aspects.

[0115] Preferably, using the container casting and melting compensation data to optimize the low-complexity three-dimensional grid model includes:

[0116] Using the container casting and melting compensation data to perform surface block division on the bottleneck area, bottle shoulder area, bottle body area, and bottle bottom area to obtain glass container surface block division characteristic data; aligning the grid model with the glass container surface block division characteristic data to obtain a high-precision and lightweight three-dimensional model of the glass container, where the glass container surface block division characteristic data includes bottleneck-shoulder material optimization data, bottle body material optimization data, and bottle bottom material optimization data;

[0117] Performing texture material mapping on the bottleneck area and the bottle shoulder area to obtain bottleneck-shoulder material optimization data;

[0118] Perform glossiness material mapping on the bottle body area to obtain bottle body material optimization data;

[0119] Perform transparency material mapping on the bottle bottom area to obtain bottle bottom material optimization data.

[0120] In the embodiments of the present invention, the focus is on the collaborative integration of error equalization applications, curved surface segmentation processing, grid alignment mechanisms, and regional material optimization techniques in different functional regions of glass containers, aiming to achieve geometric accuracy control while taking into account the visual authenticity and data load controllability of the model, thereby generating a three-dimensional model with high precision, high fidelity, and lightweight characteristics. First, the system introduces error equalization data as the basis for geometric correction. By performing statistical regression and local compensation on the node deviations in the original three-dimensional model, the surface coherence and structural stability are effectively improved. On this basis, for the four key regions of the glass container, namely the bottleneck, shoulder, body, and bottom, a differential curved surface segmentation strategy is implemented. This segmentation process integrates local Gaussian curvature analysis, principal direction normal vector distribution analysis, and curved surface slope gradient identification methods to systematically identify and classify each region in terms of geometric shape, curvature change, and optical characteristics, realizing the extraction of segmentation features driven by physical characteristics. Each curved surface segment not only contains its corresponding geometric information (such as curvature, thickness distribution, connection boundary conditions, etc.), but also carries local surface texture parameters, material response characteristics, and regional topological structure constraints. These multi-dimensional feature data serve as alignment inputs and play a key role in the subsequent grid model alignment stage. The grid alignment process combines local feature point matching algorithms with global rigid body transformation optimization to ensure seamless transitions in vertex connection, boundary smoothness, and normal continuity between different regions, thereby constructing a geometrically accurate and topologically reasonable basic three-dimensional framework. After completing the construction of the structural framework, the system further performs refined material texture mapping optimization on each functional region. Since the bottleneck and shoulder regions have drastic structural transition changes and usually carry external identification marks, a texture mapping enhancement method is adopted. Based on the image texture mapping algorithm and transparency / refractive index adjustment function, high-detail texture reproduction is performed on this region, enabling the surface texture to truly reflect the microscopic characteristics and processing marks of the glass; for the body region, a high-dynamic glossiness mapping technique is used to simulate the specular reflection and diffuse reflection ratios at different lighting angles, enhancing its smoothness performance and visual response consistency; for the bottom region, a transparency control mapping technique is used to accurately simulate the light transmittance, light intensity attenuation caused by thickness, and internal reflection characteristics, thereby enhancing its optical authenticity. Throughout the material mapping process, parameter configuration is based on the region-specific feature database to ensure that each region can reflect the same material perception and visual effect as the actual object during rendering. In summary, the present invention works together at multiple levels of error compensation, structural segmentation, geometric alignment, and regional material mapping. The generated high-precision and lightweight three-dimensional glass container model not only has micron-level accuracy in geometric restoration but also can realistically restore the reflection, light transmission, and texture detail characteristics of the actual glass in visual reconstruction, greatly improving the application adaptability and authenticity of the model in multiple scenarios such as virtual display, industrial design, and digital manufacturing.

[0121] As an example of the present invention, refer to Figure 2 As shown, in this example, step S4 includes:

[0122] Step S41: Conduct a glass surface texture evaluation based on the high-precision and lightweight three-dimensional model of the glass container to obtain glass model texture evaluation data;

[0123] Step S42: Conduct a model generation process evaluation based on the high-precision and lightweight three-dimensional model of the glass container to obtain glass model generation process evaluation data;

[0124] Step S43: Conduct a two-dimensional curve evaluation and fitting analysis on the glass model texture evaluation data and the glass model generation process evaluation data to obtain glass model construction fitting data; construct the path of the glass model construction fitting data and mark the weak points to obtain a three-dimensional model construction report of the glass container.

[0125] In the embodiment of the present invention, it involves surface texture evaluation based on a high-precision and lightweight three-dimensional model, model generation process evaluation, and generating a complete three-dimensional model construction report of the glass container through data analysis and matching. First, conduct a texture evaluation based on the high-precision three-dimensional model of the glass container. This process evaluates the authenticity and accuracy of surface details by analyzing the surface features of the model in detail. The texture evaluation data usually includes the measurement and verification of attributes such as surface gloss, transparency, and roughness. This evaluation data can reflect the quality of the glass container surface, and thus provide a basis for subsequent material optimization and rendering. Next, according to this three-dimensional model, an evaluation of the model generation process is also required. The purpose of this evaluation is to detect errors, defects, or deficiencies that occur during the three-dimensional modeling process. The generation process evaluation data mainly includes the accuracy analysis of different algorithm applications during the modeling process, the evaluation of surface smoothness, and error tracking during the geometric construction process. All these data reflect the stability and accuracy of model generation. Conducting a two-dimensional curve evaluation and fitting analysis on the texture evaluation data and the generation process evaluation data is a process of integrating and comparing the evaluation data. Through two-dimensional curve evaluation, the degree of fit of the model surface in different evaluation dimensions can be determined, that is, the geometric accuracy and texture matching degree of the model, so as to reveal whether there are deviations in the actual construction process of the model. Combining these data, generate glass model construction fitting data to further optimize the accuracy and quality of model generation. On this basis, conduct path construction and weak point marking, and identify potential problem areas of the model through technical means, such as local geometric defects, texture mismatches, or error areas during the construction process. Finally, form a three-dimensional model construction report of the glass container. This report not only details the key data and problem areas during the model construction process but also provides data support and a basis for subsequent improvement and optimization. The entire process uses data-driven analysis and evaluation to continuously refine and optimize the model, ensuring that the finally generated three-dimensional model of the glass container has high precision and high consistency in all aspects.

[0126] Particularly importantly, step S41 includes the following steps:

[0127] Step S411: Extract the texture image according to the UV map of the high-precision and lightweight three-dimensional model of the glass container, and calculate the texture contrast using the gray-level co-occurrence matrix to obtain the glass model texture contrast data;

[0128] Step S412: Perform illumination angle simulation based on the glass model texture contrast data to obtain the simulated reflectivity distribution map;

[0129] Step S413: Calculate the root mean square of the measured reflectivity based on the simulated reflectivity distribution map to obtain the glass model texture evaluation data.

[0130] In the embodiment of the present invention, by introducing the analysis of the UV map texture image based on the high-precision and lightweight three-dimensional model, a multi-level data-driven method for visual quality evaluation of glass containers is constructed. By extracting the UV map of the model and constructing the corresponding two-dimensional texture image, and then calculating its texture contrast based on the gray-level co-occurrence matrix (GLCM), not only the quantitative modeling of the surface texture distribution law is realized, but also the original subjective visual attributes are transformed into quantifiable statistical feature data, thereby obtaining the glass model texture contrast data, providing a basis for subsequent illumination simulation. Using this texture contrast data to drive the illumination angle simulation process, by setting different incident angles and observation angles, calculate the change of the reflectivity of the glass surface under various illumination conditions, and generate the simulated reflectivity distribution map. This map not only reflects the optical response of the texture details in the visual performance, but also reveals the reflection behavior characteristics of the model under different material distributions. By performing root mean square error analysis on the simulated reflectivity distribution map and the reflectivity data of the measured glass sample, the glass model texture evaluation data is obtained. This data has strong accuracy discrimination ability and texture simulation verification value, and can effectively evaluate the accuracy of the current model in terms of apparent detail expression and optical consistency. The whole process starts from the two-dimensional features of the texture image, through illumination simulation and error evaluation, forming a complete data chain, so that the surface texture of the glass container three-dimensional model not only remains consistent geometrically, but also has a high reduction ability to the real optical characteristics at the visual performance level, thereby providing solid data support and evaluation basis for the model in application scenarios such as digital rendering, industrial inspection, and virtual display.

[0131] Particularly importantly, step S42 includes the following steps:

[0132] Step S421: Calculate the missing rate of the initial point cloud in the thin-walled area according to the high-precision and lightweight three-dimensional model of the glass container, and mark the unrepaired coordinates to obtain the thin-walled unrepaired coordinate data;

[0133] Step S422: Performing mesh optimization phase evaluation on the thin-wall unrepaired coordinate data to obtain mesh optimization phase data;

[0134] Step S423: Perform global error suppression performance analysis on the data in the mesh optimization phase and evaluate the model generation process to obtain glass model generation process evaluation data.

[0135] In the embodiment of the present invention, steps S421 to S423 in this embodiment construct a set of evaluation paths for the three-dimensional modeling process of glass containers based on point cloud missing monitoring and error suppression feedback mechanism, focusing on solving the structural integrity and error diffusion problems of thin-walled areas during the modeling process. Specifically, the initial point cloud missing rate of typical thin-walled areas is calculated using the high-precision-lightweight three-dimensional model of the glass container, the point cloud data loss problem caused by scanning occlusion, curvature mutation or reflectivity change is identified, and the unrepaired coordinate positions are spatially annotated to generate thin-walled unrepaired coordinate data, thereby constructing an initial defect positioning mechanism for modeling weak areas at the point cloud level. Based on the unrepaired coordinate data, step S422 further enters the mesh optimization stage evaluation, and by comparing the geometric continuity before and after optimization, mesh density changes, topological consistency indicators and other data dimensions, the structural recovery of the thin-walled area is quantitatively analyzed to obtain mesh optimization stage data. Such data not only reflects the repair ability of mesh reconstruction for missing point compensation, but also lays an analytical foundation for subsequent error control. Next, the data from the mesh optimization stage is used as input, and the error suppression performance analysis is carried out in combination with the error accumulation in the global modeling path. An evaluation system is constructed using multi-dimensional indicators such as root mean square deviation, normal offset statistics, and mesh volume error ratio, and finally the evaluation data of the glass model generation process is formed. This data can be used to systematically reveal the impact of different links in the modeling process on the recovery of thin-walled structures and the overall geometric accuracy, and has the functional value of finely controlling the modeling quality and tracing the source of errors. Therefore, this technical path has opened up the evaluation loop from micro data defects to macro modeling quality through key steps such as point cloud missing location, mesh optimization evaluation, and error suppression analysis, and improved the process interpretability and modeling reliability of the glass container 3D modeling system.

[0136] In this specification, a three-dimensional model generation system based on a lightweight glass container is provided, which is used to execute the above-mentioned three-dimensional model generation method based on a lightweight glass container. The three-dimensional model generation system based on a lightweight glass container includes:

[0137] The multimodal sensing and acquisition module is used to collect information about lightweight glass containers using a preset multimodal scanning sensor, collect multi-view visible light images, depth information, and structured light projection data, and construct a standard multi-view point cloud data set;

[0138] The three-dimensional point cloud fusion and repair module is used to perform multi-modal fusion based on a standard multi-view point cloud dataset to obtain multi-modal fusion data of a glass container; judge the thin wall of the multi-modal fusion data of the glass container and repair the thin wall structure to obtain a three-dimensional point cloud model with high integrity;

[0139] The model lightweight optimization and material enhancement module is used to perform lightweight mesh optimization based on the three-dimensional point cloud model with high integrity to obtain a three-dimensional mesh model with low complexity; perform global error compensation for container casting and melting on the three-dimensional mesh model with low complexity to obtain container casting and melting compensation data; optimize the glass contour material of the three-dimensional mesh model with low complexity to obtain glass container surface segmentation feature data; use the container casting and melting compensation data and the glass container surface segmentation feature data to perform model lightweight optimization on the three-dimensional mesh model with low complexity to obtain a high-precision and lightweight three-dimensional model of the glass container;

[0140] The evaluation analysis and report generation module is used to perform glass surface texture evaluation based on the high-precision and lightweight three-dimensional model of the glass container to obtain glass model texture evaluation data; perform model generation process evaluation based on the high-precision and lightweight three-dimensional model of the glass container to obtain glass model generation process evaluation data; construct a full-process path for the glass model texture evaluation data and the glass model generation process evaluation data and mark weak points to obtain a three-dimensional model construction report of the glass container.

[0141] Therefore, from any perspective, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the application documents are intended to be encompassed within the present invention.

[0142] The above are only specific embodiments of the present invention, enabling those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for generating a three-dimensional model based on a lightweight glass container, characterized in that Including the following steps: Step S1: Use a preset multi-modal scanning sensor to collect information about the lightweight glass container, collect multi-view visible light images, depth information, and structured light projection data, and construct a standard multi-view point cloud dataset; Step S2: Perform multi-modal fusion based on the standard multi-view point cloud dataset to obtain multi-modal fusion data of the glass container; judge the thin wall of the multi-modal fusion data of the glass container, and perform thin wall structure repair to obtain a high-integrity three-dimensional point cloud model; Step S3: Perform lightweight grid optimization based on the high-integrity three-dimensional point cloud model to obtain a low-complexity three-dimensional grid model; perform global error compensation for container casting and melting on the low-complexity three-dimensional grid model to obtain container casting and melting compensation data; optimize the glass contour material of the low-complexity three-dimensional grid model to obtain glass container surface block feature data; use the container casting and melting compensation data and the glass container surface block feature data to perform model lightweight optimization on the low-complexity three-dimensional grid model to obtain a high-precision and lightweight three-dimensional model of the glass container; Step S4: Perform glass surface texture evaluation based on the high-precision and lightweight three-dimensional model of the glass container to obtain glass model texture evaluation data; perform model generation process evaluation based on the high-precision and lightweight three-dimensional model of the glass container to obtain glass model generation process evaluation data; construct a full-process path for the glass model texture evaluation data and the glass model generation process evaluation data, and perform weak point marking to obtain a glass container three-dimensional model construction report.

2. The three-dimensional model generation method based on a lightweight glass container according to claim 1, wherein Step S1 includes the following steps: Step S11: Deploy a high-resolution visible light camera with a resolution greater than 20 million pixels and a frame rate of 30fps and a structured light projector with a wavelength of 450 - 650nm and a stripe density of 0.1mm / line on the circular scanning orbit to collect multi-angle visible light image data and multi-angle structured light projection data at a spatial sampling interval of 0.5mm; Step S12: Deploy a ToF depth sensor with an accuracy of 0.1mm and a sampling frequency of 20Hz in the scanning system to collect the depth information of the glass container and obtain the glass projection distortion parameters; Step S13: Perform highlight suppression on the multi-angle visible light image data and the multi-angle structured light projection data, and use the glass projection distortion parameters for refraction compensation to obtain a glass noise-reduced multi-source dataset; Step S14: Perform spatial alignment on the glass noise-reduced multi-source dataset to generate a standard multi-view point cloud dataset.

3. The three-dimensional model generation method based on a lightweight glass container according to claim 1, wherein The multi-modal data fusion based on the standard multi-view point cloud dataset in Step S2 includes: Obtain a glass material parameter library; Use the glass material parameter library to simulate the refraction, reflection, and scattering behaviors of light on the glass surface to generate an optical property mapping table; Perform point binding on the optical property mapping table and the standard multi-view point cloud dataset to generate physically enhanced point cloud data; Input the physically enhanced point cloud data into a differentiable renderer to calculate the L1 loss and SSIM loss between the rendered image and the real visible light image, and generate multi-modal fusion data of the glass container, where the weights of the L1 loss and the SSIM loss are 0.7 and 0.3 respectively.

4. The method for generating a three-dimensional model based on a lightweight glass container according to claim 1, wherein The thin wall structure repair in Step S2 includes: Predict the thin-walled area less than 2 mm based on the multi-modal fusion data of the glass container, and generate the predicted glass thickness distribution; Input the predicted glass thickness distribution as a constraint condition into the Poisson surface reconstruction algorithm, where the depth of the Poisson surface reconstruction algorithm is 10 and the weight is 0.5, to obtain the vertex density data of the glass surface mesh; Iteratively Poisson reconstruct the parameters of the vertex density data of the glass surface mesh and construct a three-dimensional mesh model to obtain a high-integrity three-dimensional point cloud model.

5. The method for generating a three-dimensional model based on a lightweight glass container according to claim 1, wherein The lightweight mesh optimization based on the high-integrity three-dimensional point cloud model in step S3 includes: Perform Gaussian curvature threshold judgment on the high-integrity three-dimensional point cloud model and conduct sensitivity analysis to obtain the sensitivity analysis data of the three-dimensional point cloud model, where the Gaussian curvature threshold is less than 0.01 / ; Use the sensitivity analysis data of the three-dimensional point cloud model to identify the notch-seam area and mark it as physical property exemption data; Delete the mesh of the high-integrity three-dimensional point cloud model based on the physical property exemption data to obtain the mesh area deletion model, where the deletion rate of the mesh deletion is 20%-50%; Optimize the mesh topology of the mesh area deletion model to obtain a three-dimensional mesh model with low complexity.

6. The three-dimensional model generation method based on a lightweight glass container according to claim 5, characterized in that The mesh topology optimization of the mesh area deletion model in step S3 includes: Optimize the edge collapse error of the mesh area deletion model to obtain the edge collapse optimization data; Verify the integrity of the mesh manifold according to the edge collapse optimization data to obtain the glass manifold integrity data; Use the glass manifold integrity data to construct the mesh topology structure, and verify the non-manifold edges and isolated vertices with Euler's formula to obtain the glass container mesh topology structure; Optimize the glass container mesh topology structure and the mesh area deletion model for low complexity to obtain a three-dimensional mesh model with low complexity.

7. The three-dimensional model generation method based on a lightweight glass container according to claim 1, wherein The global error compensation for container casting and melting of the three-dimensional mesh model with low complexity in step S3 includes: Obtain the measured data of the glass container; Extract the geometric nodes of the glass container based on the three-dimensional mesh model with low complexity; Analyze the node errors by using the geometric node extraction of the glass container and the measured data of the glass container to obtain the comparative analysis data of the glass container nodes; Perform deviation compensation according to the comparative analysis data of the glass container nodes to obtain the compensation data for container casting and melting.

8. The three-dimensional model generation method based on a lightweight glass container according to claim 1, wherein, The optimization of the glass contour material of the three-dimensional mesh model with low complexity includes: Use the compensation data for container casting and melting to perform surface block division on the bottleneck area, bottle shoulder area, bottle body area, and bottle bottom area to obtain the surface block division characteristic data of the glass container; align the mesh model with the surface block division characteristic data of the glass container to obtain a high-precision and lightweight three-dimensional model of the glass container, where the surface block division characteristic data of the glass container includes the bottleneck-shoulder material optimization data, bottle body material optimization data, and bottle bottom material optimization data; Perform texture material mapping on the bottleneck area and the bottle shoulder area to obtain the bottleneck-shoulder material optimization data; Perform specular material mapping on the bottle body area to obtain the bottle body material optimization data; Perform transparency material mapping on the bottle bottom area to obtain the bottle bottom material optimization data.

9. The three-dimensional model generation method based on a lightweight glass container according to claim 1, characterized in that, Step S4 includes the following steps: Step S41: Evaluate the glass surface texture according to the high-precision and lightweight three-dimensional model of the glass container to obtain the glass model texture evaluation data; Step S42: Evaluate the model generation process according to the high-precision and lightweight three-dimensional model of the glass container to obtain the glass model generation process evaluation data; Step S43: Perform two-dimensional curve evaluation and fitting analysis on the glass model texture evaluation data and the glass model generation process evaluation data to obtain the glass model construction fitting data; construct the path of the glass model construction fitting data and perform weak point marking to obtain the three-dimensional model construction report of the glass container.

10. A three-dimensional model generation system based on lightweight glass containers, characterized in that, A three-dimensional model generation system for a lightweight glass container for performing the method for generating a three-dimensional model of a lightweight glass container as described in claim 1, the three-dimensional model generation system for a lightweight glass container comprising: A multimodal perception acquisition module for using a preset multimodal scanning sensor to collect information on the lightweight glass container, collecting multi-view visible light images, depth information, and structured light projection data, and constructing a standard multi-view point cloud data set; A three-dimensional point cloud fusion and repair module for performing multimodal fusion based on the standard multi-view point cloud data set to obtain multimodal fusion data of the glass container; performing a thin-wall judgment on the multimodal fusion data of the glass container and performing thin-wall structure repair to obtain a high-integrity three-dimensional point cloud model; A model lightweight optimization and material enhancement module for performing lightweight grid optimization based on the high-integrity three-dimensional point cloud model to obtain a low-complexity three-dimensional grid model; performing global error compensation for container casting and melting on the low-complexity three-dimensional grid model to obtain container casting and melting compensation data; optimizing the glass profile material of the low-complexity three-dimensional grid model to obtain glass container surface patch feature data; using the container casting and melting compensation data and the glass container surface patch feature data to perform model lightweight optimization on the low-complexity three-dimensional grid model to obtain a high-precision - lightweight three-dimensional model of the glass container; An evaluation analysis and report generation module for performing glass surface texture evaluation based on the high-precision - lightweight three-dimensional model of the glass container to obtain glass model texture evaluation data; performing model generation process evaluation based on the high-precision - lightweight three-dimensional model of the glass container to obtain glass model generation process evaluation data; constructing a full-process path for the glass model texture evaluation data and the glass model generation process evaluation data and performing weak point marking to obtain a three-dimensional model construction report of the glass container.

Images