Concrete frame structure drawing automatic identification and integrated modeling method and system
By using multimodal data processing and deep learning model training, the problems of low efficiency and accuracy in the conversion of two-dimensional concrete frame drawings into three-dimensional BIM models have been solved, achieving efficient and accurate automatic identification and integrated modeling to meet the needs of diverse engineering scenarios.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-01-08
- Publication Date
- 2026-04-03
AI Technical Summary
Existing technologies are inefficient, costly, and prone to human error in converting two-dimensional concrete frame drawings into three-dimensional BIM models. Furthermore, existing automated methods have poor generalization capabilities, lack deep semantic understanding, and cannot effectively associate graphic and textual information, resulting in models that lack engineering attributes.
By employing multimodal data processing, deep learning model training, component semantic parsing, spatial location positioning, 2D detail generation, and logical consistency verification, a fully automated and high-precision conversion from 2D drawings to 3D BIM models is achieved. This includes multimodal drawing data preprocessing, deep learning-based component matching and text semantic parsing, component spatial location positioning, attribute association, detail generation, and logical consistency verification.
It significantly improves the efficiency of component recognition and semantic parsing, ensures the accuracy of component recognition and attribute association, adapts to diverse engineering scenarios, meets different application needs, and realizes fully automatic and high-precision conversion from two-dimensional drawings to three-dimensional BIM models.
Smart Images
Figure CN121786937A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of intelligent processing of engineering design drawings and building information modeling (BIM) technology, and particularly relates to a method and system for automatic recognition and integrated modeling of concrete frame structure drawings. Background Technology
[0002] In the field of architectural engineering, the conversion of two-dimensional concrete frame drawings into three-dimensional Building Information Models (BIM) is a crucial step in achieving the industry's digital transformation. However, the current conversion process mainly relies on manual modeling, which suffers from low efficiency, high costs, and susceptibility to human error, making it difficult to guarantee the integrity and accuracy of non-geometric information in the two-dimensional drawings in the three-dimensional model. To improve efficiency, some existing automated methods attempt to use traditional image processing techniques or rule-based recognition methods for component analysis, but these methods have poor generalization ability, heavily rely on strict drawing specifications, and are insufficiently adaptable to diverse drawing styles. More importantly, these methods lack deep semantic understanding capabilities and cannot effectively associate scattered graphic, text, and annotation information, resulting in models that only possess geometric shapes but lack necessary engineering attributes, making it difficult to meet the needs of subsequent analysis and applications. Although the introduction of deep learning technology has brought new possibilities to this field, existing methods are still mostly limited to the recognition of simple components, and still have significant shortcomings in multi-scale target detection of complex drawings, cross-modal association of graphics and text, and automatic reasoning of topological relationships between components. These problems make it difficult for existing technologies to achieve fully automated, high-precision conversion from two-dimensional drawings to structured information models, becoming a technical bottleneck restricting the development of engineering applications. Therefore, developing a method that can efficiently and accurately achieve automatic recognition and integrated modeling of two-dimensional drawings has become an urgent technical challenge to be solved in this field. Summary of the Invention
[0003] To address the aforementioned technical problems, this invention provides a method and system for automatic identification and integrated modeling of concrete frame structure drawings. Through multimodal data processing, deep learning model training, component semantic parsing, spatial location positioning, two-dimensional detail generation, three-dimensional model construction, and logical consistency verification, it achieves fully automatic and high-precision conversion from two-dimensional drawings to three-dimensional BIM models.
[0004] To achieve the above-mentioned objectives, the first objective of this invention is to provide a method for automatic recognition and integrated modeling of concrete frame structure drawings, comprising: S1: Multimodal drawing data preprocessing; S2: Component matching and text semantic parsing based on deep learning; S3: Spatial location and attribute association of components; S4: Generation and standardized redrawing of 2D detail maps based on information completion; S5: Integrated construction and logical consistency verification of structural information model; S6: Model output and multi-scenario interactive applications.
[0005] Preferably, S1 includes: S11: Build a dataset of concrete frame drawings that supports PDF, DWG, PNG and JPG formats. The drawing types include beam reinforcement drawings, column reinforcement drawings, slab reinforcement drawings and foundation drawings. S12: Convert non-image format drawings into high-resolution images and perform noise reduction filtering, geometric distortion correction, contrast enhancement and scale normalization operations. S13: Based on image content feature analysis and view boundary detection algorithms, the drawing is cropped and segmented using independent views as the basic unit, and named according to the standard format of "project number_drawing type_view number_timestamp"; S14: Perform quality verification and sample screening on the preprocessed view data, and remove duplicate, blurry, or substandard samples.
[0006] Preferably, S2 includes: S21: It adopts a two-stage target detection architecture, deploying a relationship detection model and a component detection model respectively, using an end-to-end optical character recognition model to extract text information, and optimizing for engineering drawing scenarios to expand the recognition capability of special symbols for steel bars; S22: Use the component detection model to annotate the basic elements and professional components of the drawings, and use the relationship detection model to annotate the relationships; S23, based on the semantic association of dimension annotation components in spatial relationships, by analyzing the spatial overlap features and relative position features of the dimension annotation relationship bounding box and the dimension annotation value and dimension annotation anchor point bounding box, the matching relationship between the dimension annotation component and the corresponding dimension annotation value and dimension annotation anchor point is determined; S24, Semantic Association of Beam Components Based on Spatial Relationships: By analyzing the spatial overlap and relative position characteristics of the beam annotation relationship bounding box and the beam and reinforcement annotation bounding box, the matching relationship between beam components and related annotations is constructed. S25, Semantic Association of Column Components Based on Spatial Relationships: By analyzing the spatial overlap and relative position characteristics of the column annotation relationship bounding box and the column and reinforcement annotation bounding box, the matching relationship between the column component and the corresponding annotation is determined. S26, Semantic association of reinforced concrete members based on spatial relationships: By analyzing the spatial overlap features and contextual position information of the boundary boxes of reinforced concrete relationships and the boundary boxes of reinforced concrete symbols and reinforcement annotations, the matching relationship between reinforced concrete members and their corresponding annotations is determined. S27, Enhanced Special Symbol Recognition: Based on the existing optical character recognition model, a special symbol dataset for rebar is introduced for incremental training, expanding the model's ability to recognize rebar symbols.
[0007] Preferably, S3 includes: S31. Determine the horizontal and vertical grid lines according to the grid naming rules and spatial relationships; S32, project the center points of all horizontal grid boundary frames onto the horizontal line, and project the dimension anchor points of all horizontal dimension components onto the horizontal line. If the anchor points on the left and right sides of the dimension component coincide with the projection points of the adjacent grid within a certain error, then define the dimension value of the dimension component as the distance between the two grids. S33, project the center points of all vertical grid boundary frames onto the vertical lines, and project the dimension anchor points of all vertical dimension components onto the vertical lines. If the anchor points on the left and right sides of the dimension component coincide with the projection points of the adjacent grid within a certain error, then define the dimension value of the dimension component as the distance between the two grids. S34 uses the horizontal and vertical grid lines of the component drawing as the grid line, and takes the intersection of the horizontal and vertical grid lines at the bottom left corner as the origin of the drawing coordinates. S35, determine the position of the component based on the geometry and orientation of the component boundary frame combined with the axis grid; S35, determine the component attribute information based on the text content of the component and the corresponding annotation.
[0008] Preferably, S4 includes: S41, merge the grid lines of all drawings to determine the project grid lines of the entire set of drawings, and offset the spatial positions of all components relative to the project grid lines; S42, determine the uniqueness of a component across different drawings based on its spatial position and geometric information in the project grid, and merge the component's attribute information; S43, Based on the spatial location and attribute information of the column components, draw the column layout details, column section details, and column longitudinal section details; S44. Based on the spatial location and attribute information of the beam members, draw detailed plan layout drawings, cross-sectional reinforcement drawings, and section reinforcement drawings of the beam in different planes. S45. Determine the area of the slab based on the spatial location of the beam members, and draw the reinforcement details of the slab.
[0009] Preferably, S5 includes: S51, automatically constructs a 3D BIM model based on the type and spatial location of the components; S52, Assign attribute information to the 3D model based on the component's attribute information; S53 automatically completes the operation of component alignment and hierarchical assignment based on the component connection relationship, hierarchical structure and grid positioning; S54 performs spatial location conflict detection, component size design rules and constraints detection, attribute information integrity detection, and logical consistency verification on components.
[0010] Preferably, S6 includes: S61: Establish a two-way link between two-dimensional detailed drawings and three-dimensional models, and realize the linked query and dynamic retrieval of component information through a visual interface; S62: A custom JSON data structure stores the geometric and reinforcement properties of components and supports exporting to IFC, Revit, and Tekla formats.
[0011] The second objective of this invention is to provide an automatic recognition and integrated modeling system for concrete frame structure drawings, comprising: The preprocessing module preprocesses the multimodal drawing data; The matching and parsing module uses deep learning to perform component matching and text semantic parsing. The positioning and association module enables spatial location and attribute association of components. The completion specification module generates two-dimensional detailed drawings and standard redraws based on information completion; A verification module is built to integrate the construction of the structural information model with logical consistency verification. Interaction module, model output and multi-scenario interactive applications.
[0012] A third objective of this invention is to provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned method for automatic recognition and integrated modeling of concrete frame structure drawings.
[0013] A fourth objective of this invention is to provide a computer program product, including a computer program that, when executed by a processor, implements the aforementioned method for automatic recognition and integrated modeling of concrete frame structure drawings.
[0014] The advantages and positive effects of this application are: This invention significantly improves the efficiency of component recognition and semantic parsing through multimodal data processing and deep learning model training; ensures the accuracy of component recognition and attribute association through spatial relationship analysis and logical consistency verification; adapts to diverse engineering scenarios by supporting multiple drawing formats and types; and meets the needs of different application scenarios through custom JSON data structures and multi-format export functions.
[0015] In summary, this invention achieves fully automated and high-precision conversion from two-dimensional drawings to three-dimensional BIM models through multimodal data processing, deep learning model training, component semantic parsing, spatial location positioning, two-dimensional detail drawing generation, three-dimensional model construction, and logical consistency verification, and has significant engineering application value. Attached Figure Description
[0016] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Figure 1 A flowchart of a preferred embodiment of the present invention is shown; Figure 2 The original scanned drawing of the beam reinforcement diagram before standardization in a preferred embodiment of the present invention is shown. Obviously, the clarity of the original drawing is insufficient. Figure 3 This shows a schematic diagram of the standardized beam reinforcement diagram in a preferred embodiment of the present invention. Figure 4 This diagram illustrates the matching relationship between beam members and related annotations, and between dimensioned members and related annotations, in a preferred embodiment of the present invention. Figure 5 This shows a schematic diagram of the grid lines for the beam reinforcement diagram in a preferred embodiment of the present invention; Figure 6 This illustrates a perspective 3D model generated in Revit in a preferred embodiment of the present invention; Figure 7 A non-perspective 3D model generated in Revit is shown in a preferred embodiment of the present invention. Detailed Implementation
[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0018] Please see Figures 1 to 7 The first embodiment, a method for automatic recognition and integrated modeling of concrete frame structure drawings, mainly includes: S1: Multimodal drawing data preprocessing; S2: Component matching and text semantic parsing based on deep learning; S3: Spatial location and attribute association of components; S4: Generation and standardized redrawing of 2D detail maps based on information completion; S5: Integrated construction and logical consistency verification of structural information model; S6: Model output and multi-scenario interactive applications.
[0019] To better understand the technical concept of this invention, the following non-limiting description is provided: In the above embodiments, S1 specifically includes: S11, construct a standardized, multimodal concrete frame drawing dataset. Data formats include, but are not limited to, vector formats such as PDF and DWG, as well as common image formats such as PNG and JPG. Drawing types are not limited to beam reinforcement drawings, column reinforcement drawings, slab reinforcement drawings, and foundation drawings. This step constructed a standardized, multimodal dataset of concrete frame drawings. The data formats cover vector formats such as PDF, DWG, and DXF, as well as common raster image formats such as PNG, JPG, and TIFF, ensuring the integrity and reusability of the drawing information. The drawing types are extensive, including beam reinforcement drawings, column reinforcement drawings, slab reinforcement drawings, and foundation drawings, while also extending to detailed node drawings, sectional views, and structural layout drawings, striving to comprehensively cover all types of drawings involved in concrete frame design. The dataset will emphasize annotation quality and structural hierarchy, including multi-dimensional annotations such as component geometry information, material properties, and reinforcement details, to support applications in multiple fields such as deep learning, image recognition, and BIM modeling.
[0020] S12 converts non-image format drawings into high-resolution image formats using a specific conversion tool, and performs standardization operations on the converted images, including noise reduction filtering, geometric distortion correction, contrast enhancement, and scale normalization. This step converts non-image format drawings into high-resolution image formats (such as PNG, TIFF, or JPEG 2000) using professional file conversion tools, such as CAD export plugins or dedicated drawing processing software, while maintaining the detail accuracy and line clarity of the original drawings. Subsequently, a series of standardization operations are performed on the converted images, including but not limited to noise reduction using adaptive filters, elimination of geometric distortion through affine transformation or perspective correction techniques, enhancement of image contrast using histogram equalization or CLAHE methods, and size calibration and scale normalization of the images based on the scale markings in the original drawings. This ensures that all drawings achieve a unified and usable state in terms of visual presentation and measurement standards.
[0021] S13, based on image content feature analysis and view boundary detection algorithms, uses independent views as the basic unit to crop and segment drawings, and uniformly names them according to the standard format of "project number_drawing type_view sequence number_time stamp", thus establishing a standardized view file management system; This step, based on image content feature analysis and view boundary detection algorithms, automatically crops and segments drawings using independent views as the basic unit. This effectively separates various structural views, sectional views, and details. Each view file is uniformly named and serialized according to the standard format of "Project Number_Drawing Type_View Sequence Number_Timestamp," thereby establishing a standardized view file storage and retrieval system and improving the organization efficiency and reusability of drawing data.
[0022] S14 performs quality verification and sample screening on the preprocessed view data, removing duplicate, blurry, or substandard samples to ensure the quality and usability of the dataset.
[0023] This step involves systematically verifying the quality of the preprocessed view data and rigorously screening samples. It combines automated tools with manual review to identify and remove duplicate, ambiguous, incorrectly labeled, or substandard samples. At the same time, it checks the consistency and integrity of the data to ensure that the final dataset has high quality, high availability, and good generalization ability, providing a reliable data foundation for subsequent model training and evaluation.
[0024] In this embodiment, S2 specifically includes: S21, component recognition adopts a two-stage target detection architecture based on deep learning, deploying a relationship detection model and a component detection model respectively. Text parsing adopts an end-to-end optical character recognition (OCR) model that integrates text detection and recognition, and is optimized for engineering drawing scenarios; This embodiment employs a deep learning-based two-stage object detection architecture for component recognition. This architecture first analyzes the spatial and logical relationships between elements in the image using a relationship detection model, and then uses a component detection model to accurately locate and classify specific components, ensuring the accuracy and robustness of the recognition results. For text parsing, the system integrates an end-to-end optical character recognition (OCR) model, which simultaneously performs text detection and recognition tasks, significantly improving processing efficiency. Furthermore, considering the complex layouts, dense annotations, and professional symbols commonly found in engineering drawings, the OCR model has undergone several optimizations, including enhanced adaptability to distorted text, low-resolution characters, and background interference, thereby achieving higher-quality text parsing results in engineering drawing scenarios.
[0025] S22, The labeling system of the component inspection model covers the basic elements of the drawings and professional components. The labels should include at least: title, scale, grid, dimension value, dimension anchor point, beam, column, slab, reinforcement, reinforcement label, etc.; The labeling labels of the relationship inspection model should include at least: dimension relationship, beam relationship, column relationship, reinforcement relationship, etc. In this step, the component inspection model's annotation system is comprehensive, covering basic elements and key professional components in the drawings. Its labels include at least: title, scale, grid, dimension values, dimension anchor points, beam, column, slab, reinforcement, and detailing annotations. Meanwhile, the relationship inspection model's annotation system focuses on the semantic and spatial relationships between elements, with labels covering at least multiple dimensions such as dimension annotation relationships, beam annotation relationships, column annotation relationships, and reinforcement annotation relationships. These two models together support the structured and semantic parsing of drawing information, providing a foundation for automated drawing recognition and intelligent review.
[0026] S23, based on the semantic association of dimension annotation components in spatial relationships, by analyzing the spatial overlap features and relative position features of the dimension annotation relationship bounding box and the dimension annotation value and dimension annotation anchor point bounding box, the matching relationship between the dimension annotation component and the corresponding dimension annotation value and dimension annotation anchor point is determined; This step is based on the semantic association of dimensioned components in spatial relationships. By analyzing the spatial overlap and relative position features between the dimensioned component boundary box and the dimensioned value and dimensioned anchor point boundary box, and combining contextual geometric constraints and layout logic, a multi-level correspondence between components and annotations is established. This accurately determines the semantic matching relationship between dimensioned components and their corresponding dimensioned values and dimensioned anchor points, further improving the automation and intelligence level of engineering drawing understanding.
[0027] S24, Semantic Association of Beam Components Based on Spatial Relationships: By analyzing the spatial overlap and relative position characteristics of the beam annotation relationship bounding box and the beam and reinforcement annotation bounding box, the matching relationship between beam components and related annotations is constructed. This step is based on the semantic association of beam components in spatial relationships: by analyzing the spatial overlap and relative position features between the beam annotation relationship bounding box and the beam and reinforcement annotation bounding boxes, and combining geometric attributes such as the intersection-union ratio, distance vector and orientation angle between the bounding boxes, a stable matching relationship between beam components and related annotations is constructed; further, spatial topological constraints and contextual semantic consistency verification are introduced to improve the accuracy and robustness of the association between components and annotations, and ensure the complete transmission of structural information and the efficiency of automated recognition.
[0028] S25, Semantic Association of Column Components Based on Spatial Relationships: By analyzing the spatial overlap and relative position characteristics of the column annotation relationship bounding box and the column and reinforcement annotation bounding box, the matching relationship between the column component and the corresponding annotation is determined. This step is based on the semantic association of column components according to spatial relationships: by analyzing the spatial overlap and relative position characteristics of the column annotation relationship boundary box and the column and reinforcement annotation boundary boxes, the matching relationship between column components and corresponding annotations is determined. Specifically, this method uses the degree of geometric overlap, relative distance, and orientation relationship between boundary boxes to identify annotation information belonging to the same component, thereby realizing the automatic association of column components with semantic elements such as reinforcement annotations and dimension annotations in structural drawings, effectively improving the accuracy and completeness of component information extraction in BIM models.
[0029] S26, Semantic association of reinforced concrete members based on spatial relationships: By analyzing the spatial overlap features and contextual position information of the boundary boxes of reinforced concrete relationships and the boundary boxes of reinforced concrete symbols and reinforcement annotations, the matching relationship between reinforced concrete members and their corresponding annotations is determined. This step is based on the semantic association of steel reinforcement components according to spatial relationships: The method first identifies and extracts the spatial geometric features of the boundary boxes of steel reinforcement relationships, steel reinforcement symbols, and reinforcement annotation boundary boxes. It then calculates the overlap, relative position, and directional relationships between the boundary boxes, and performs a comprehensive analysis in conjunction with contextual layout information. Specifically, the system detects whether boundary boxes overlap, contain, or are adjacent, and infers the semantic matching relationship between the steel reinforcement component and its corresponding annotation based on geometric indicators such as projection alignment, axial distance, and area intersection. This process fully integrates visual structure and semantic context, effectively improving the accuracy and robustness of steel reinforcement component identification and annotation association.
[0030] S27, Enhanced Special Symbol Recognition: Based on the existing optical character recognition model, a special symbol dataset for rebar is introduced for incremental training, expanding the model's ability to recognize rebar symbols.
[0031] Enhanced recognition of specialized symbols builds upon existing optical character recognition (OCR) models by incrementally training them using a high-quality dataset of specialized rebar symbols. This effectively expands the model's ability to recognize industry-specific symbols for rebar. This method not only maintains the model's recognition performance for general characters but also significantly improves the accuracy and robustness of specialized annotations such as rebar grades, markings, and diameter symbols. This better meets the needs of applications such as digitizing engineering drawings, tracing rebar quality, and intelligent construction management.
[0032] In this embodiment, S3 specifically includes: S31. Determine the horizontal and vertical grids according to the grid naming rules and spatial relationships. Specific steps include analyzing the frame plan layout, identifying the main axis directions, combining the arrangement of structural components, clarifying the positioning basis of the horizontal and vertical axes, and using a unified numbering system to systematically name the axes to ensure that the grid can clearly and accurately convey spatial positioning information during the design and construction phases.
[0033] S32, project the center points of all horizontal grid boundary frames onto the horizontal line, and project the dimension anchor points of all horizontal dimension components onto the horizontal line. If the anchor points on the left and right sides of the dimension component coincide with the projection points of the adjacent grid within a certain error, then define the dimension value of the dimension component as the distance between the two grids. S33, project the center points of all vertical grid boundary frames onto the vertical lines, and project the dimension anchor points of all vertical dimension components onto the vertical lines. If the anchor points on the left and right sides of the dimension component coincide with the projection points of the adjacent grid within a certain error, then define the dimension value of the dimension component as the distance between the two grids. S34 uses the horizontal and vertical grid lines of the component drawing as the grid line, and takes the intersection of the horizontal and vertical grid lines at the bottom left corner as the origin of the drawing coordinates. S35, determine the position of the component based on the geometry and orientation of the component boundary frame combined with the axis grid; S35, determine the name, size, reinforcement and other attribute information of the component based on the text content of the component and the corresponding annotation.
[0034] In this embodiment, the process of generating and normalizing the two-dimensional detail map based on information completion in step S4 includes: S41, merge the grid lines of all drawings to determine the project grid lines of the entire set of drawings, and offset the spatial positions of all components relative to the project grid lines; S42, determine the uniqueness of a component across different drawings based on its spatial position and geometric information in the project grid, and merge the component's attribute information; S43. Based on the spatial location and attribute information of the column components, draw the column layout details, column section details, and column longitudinal section details.
[0035] S44. Based on the spatial location and attribute information of the beam members, draw detailed plan layout drawings, cross-sectional reinforcement drawings, and section reinforcement drawings of the beam in different planes. S45. Determine the area of the slab based on the spatial location of the beam members, and draw the reinforcement details of the slab.
[0036] In this embodiment, S5 includes: S51, automatically constructs a 3D BIM model based on the type and spatial location of the components; More specifically, based on key parameters such as component type and spatial location, algorithms and rule engines are used to automatically construct refined 3D BIM models, realizing intelligent conversion from 2D information to 3D structure; S52, based on the component's attribute information, assign material, reinforcement, and construction attribute information to the 3D model; More specifically, based on the detailed attribute data of the components, such as material specifications, reinforcement parameters and construction requirements, the three-dimensional model is dynamically assigned corresponding physical and engineering attributes, enhancing the information dimension and engineering practicality of the model; S53 automatically completes operations such as component alignment and hierarchy assignment based on component connection relationships, hierarchical structure and grid positioning; More specifically, based on the connection relationships between components, their hierarchical structure, and the grid positioning information, the system automatically performs operations such as component alignment, spatial repositioning, and hierarchical structure optimization to ensure overall model coordination and logical consistency. S54 performs logical consistency checks on components, including spatial location conflict detection, component size design rules and constraints detection, and attribute information integrity detection.
[0037] More specifically, multi-dimensional logical consistency checks are performed on components, including spatial location conflict detection, whether component dimensions conform to design rules and constraints, and whether attribute information is complete and valid, in order to improve model quality and compliance.
[0038] In this embodiment, S6 includes: S61 establishes a two-way link between two-dimensional detailed drawings and three-dimensional models, and realizes the linkage query, dynamic retrieval and design modification of component information through a visual interface, ensuring the consistency of design data in multi-dimensional expression and improving the efficiency of design coordination and information interaction. S62 allows for customized JSON data structures based on component type, uniformly storing component geometry, reinforcement parameters, and other key attribute information. It also supports exporting to various mainstream professional formats such as IFC, Revit, and Tekla, enabling efficient data exchange and collaboration across platforms and systems.
[0039] An automatic recognition and integrated modeling system for concrete frame structure drawings is provided to implement the automatic recognition and integrated modeling method for concrete frame structure drawings in the above embodiments. The system includes: The preprocessing module preprocesses the multimodal drawing data; The matching and parsing module uses deep learning to perform component matching and text semantic parsing. The positioning and association module enables spatial location and attribute association of components. The completion specification module generates two-dimensional detailed drawings and standard redraws based on information completion; A verification module is built to integrate the construction of the structural information model with logical consistency verification. Interaction module, model output and multi-scenario interactive applications.
[0040] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned method for automatic recognition and integrated modeling of concrete frame structure drawings.
[0041] A computer program product includes a computer program that, when executed by a processor, implements the aforementioned method for automatic recognition and integrated modeling of concrete frame structure drawings.
[0042] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented, in whole or in part, as a computer program product, the computer program product includes one or more computer instructions. When the computer program instructions are loaded or executed on a computer, the processes or functions described in the embodiments of the present invention are generated, in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)).
[0043] The above description is only a preferred embodiment of the present invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.
Claims
1. A method for automatic recognition and integrated modeling of concrete frame structure drawings, characterized in that, include: S1: Multimodal drawing data preprocessing; S2: Component matching and text semantic parsing based on deep learning; S3: Spatial location and attribute association of components; S4: Generation and standardized redrawing of 2D detail maps based on information completion; S5: Integrated construction and logical consistency verification of structural information model; S6: Model output and multi-scenario interactive applications.
2. The method for automatic recognition and integrated modeling of concrete frame structure drawings according to claim 1, characterized in that, S1 includes: S11: Build a dataset of concrete frame drawings that supports PDF, DWG, PNG and JPG formats. The drawing types include beam reinforcement drawings, column reinforcement drawings, slab reinforcement drawings and foundation drawings. S12: Convert non-image format drawings into high-resolution images and perform noise reduction filtering, geometric distortion correction, contrast enhancement and scale normalization operations. S13: Based on image content feature analysis and view boundary detection algorithms, the drawing is cropped and segmented using independent views as the basic unit, and named according to the standard format of "project number_drawing type_view number_timestamp"; S14: Perform quality verification and sample screening on the preprocessed view data, and remove duplicate, blurry, or substandard samples.
3. The method for automatic recognition and integrated modeling of concrete frame structure drawings according to claim 1, characterized in that, S2 include: S21: It adopts a two-stage target detection architecture, deploying a relationship detection model and a component detection model respectively, using an end-to-end optical character recognition model to extract text information, and optimizing for engineering drawing scenarios to expand the recognition capability of special symbols for steel bars; S22: Use the component detection model to annotate the basic elements and professional components of the drawings, and use the relationship detection model to annotate the relationships; S23, based on the semantic association of dimension annotation components in spatial relationships, by analyzing the spatial overlap features and relative position features of the dimension annotation relationship bounding box and the dimension annotation value and dimension annotation anchor point bounding box, the matching relationship between the dimension annotation component and the corresponding dimension annotation value and dimension annotation anchor point is determined; S24, Semantic Association of Beam Components Based on Spatial Relationships: By analyzing the spatial overlap and relative position characteristics of the beam annotation relationship bounding box and the beam and reinforcement annotation bounding box, the matching relationship between beam components and related annotations is constructed. S25, Semantic Association of Column Components Based on Spatial Relationships: By analyzing the spatial overlap and relative position characteristics of the column annotation relationship bounding box and the column and reinforcement annotation bounding box, the matching relationship between the column component and the corresponding annotation is determined. S26, Semantic association of reinforced concrete members based on spatial relationships: By analyzing the spatial overlap features and contextual position information of the boundary boxes of reinforced concrete relationships and the boundary boxes of reinforced concrete symbols and reinforcement annotations, the matching relationship between reinforced concrete members and their corresponding annotations is determined. S27, Enhanced Special Symbol Recognition: Based on the existing optical character recognition model, a special symbol dataset for rebar is introduced for incremental training, expanding the model's ability to recognize rebar symbols.
4. The method for automatic recognition and integrated modeling of concrete frame structure drawings according to claim 1, characterized in that, S3 include: S31. Determine the horizontal and vertical grid lines according to the grid naming rules and spatial relationships; S32, project the center points of all horizontal grid boundary frames onto the horizontal line, and project the dimension anchor points of all horizontal dimension components onto the horizontal line. If the anchor points on the left and right sides of the dimension component coincide with the projection points of the adjacent grid within a certain error, then define the dimension value of the dimension component as the distance between the two grids. S33, project the center points of all vertical grid boundary frames onto the vertical lines, and project the dimension anchor points of all vertical dimension components onto the vertical lines. If the anchor points on the left and right sides of the dimension component coincide with the projection points of the adjacent grid within a certain error, then define the dimension value of the dimension component as the distance between the two grids. S34 uses the horizontal and vertical grid lines of the component drawing as the grid line, and takes the intersection of the horizontal and vertical grid lines at the bottom left corner as the origin of the drawing coordinates. S35, determine the position of the component based on the geometry and orientation of the component boundary frame combined with the axis grid; S35, determine the component attribute information based on the text content of the component and the corresponding annotation.
5. The method for automatic recognition and integrated modeling of concrete frame structure drawings according to claim 1, characterized in that, S4 includes: S41, merge the grid lines of all drawings to determine the project grid lines of the entire set of drawings, and offset the spatial positions of all components relative to the project grid lines; S42, determine the uniqueness of a component across different drawings based on its spatial position and geometric information in the project grid, and merge the component's attribute information; S43, Based on the spatial location and attribute information of the column components, draw the column layout details, column section details, and column longitudinal section details; S44. Based on the spatial location and attribute information of the beam members, draw detailed plan layout drawings, cross-sectional reinforcement drawings, and section reinforcement drawings of the beam in different planes. S45. Determine the area of the slab based on the spatial location of the beam members, and draw the reinforcement details of the slab.
6. The method for automatic recognition and integrated modeling of concrete frame structure drawings according to claim 1, characterized in that, S5 include: S51, automatically constructs a 3D BIM model based on the type and spatial location of the components; S52, Assign attribute information to the 3D model based on the component's attribute information; S53 automatically completes the operation of component alignment and hierarchical assignment based on the component connection relationship, hierarchical structure and grid positioning; S54 performs spatial location conflict detection, component size design rules and constraints detection, attribute information integrity detection, and logical consistency verification on components.
7. The method for automatic recognition and integrated modeling of concrete frame structure drawings according to claim 1, characterized in that, S6 include: S61: Establish a two-way link between two-dimensional detailed drawings and three-dimensional models, and realize the linked query and dynamic retrieval of component information through a visual interface; S62: A custom JSON data structure stores the geometric and reinforcement properties of components and supports exporting to IFC, Revit, and Tekla formats.
8. A system for automatic recognition and integrated modeling of concrete frame structure drawings, characterized in that, include: The preprocessing module preprocesses the multimodal drawing data; The matching and parsing module uses deep learning to perform component matching and text semantic parsing. The positioning and association module enables spatial location and attribute association of components. The completion specification module generates two-dimensional detailed drawings and standard redraws based on information completion; A verification module is built to integrate the construction of the structural information model with logical consistency verification. Interaction module, model output and multi-scenario interactive applications.
9. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the automatic recognition and integrated modeling method for concrete frame structure drawings as described in any one of claims 1-7.
10. A computer-readable storage medium storing a computer program, characterized in that, When executed by the processor, the program implements the automatic recognition and integrated modeling method for concrete frame structure drawings as described in any one of claims 1-7.