Building model construction method based on AI drawing recognition

Abstract

The invention relates to the technical field of architectural drawing processing, in particular to a building model construction method based on AI drawing recognition, which comprises the following steps of: recognizing and classifying elements, such as walls, doors and windows, stairs and the like, in a plane drawing through an AI algorithm; a large amount of drawing data is trained by using a deep learning technology, so that a system accurately understands drawing symbols, lines and text information and converts the information into three-dimensional model elements. The spatial relationship between the elements is automatically established, and a foundation is laid for three-dimensional model construction. Training a plane drawing, fixing the weight of the multi-view geometric encoder as a reference model, and inputting multi-view data. During training, diffusion loss is concentrated in aligned distribution, a CAD compiler executes strict command rules, and sometimes, a curve does not form a closed surface, so that CAD codes cannot pass inspection. Under the LLM framework, the reward function design reflects human preferences, a reinforcement learning technology is adopted to align strategy models, a code checker is introduced to serve as an implicit reward model, and potential code denoising is achieved. And analyzing the multi-view data, merging different view angle data, automatically aligning and merging, and generating a three-dimensional model framework.

Description

Technical Field

[0001] The present invention relates to the technical field of architectural drawings, and in particular to a method for constructing a building model based on AI drawing recognition. Background Art

[0002] Existing building modeling methods primarily rely on manual interpretation of design drawings and modeling using software such as CAD. This approach is not only time-consuming and labor-intensive, but also prone to modeling errors due to human error. Furthermore, with the increasing complexity and diversity of architectural designs, traditional methods are no longer able to meet the demands for efficient and accurate modeling. Especially when dealing with large-scale, highly complex architectural designs, the efficiency and quality of manual modeling often become bottlenecks that hinder project progress. Therefore, it is crucial to develop a method that can automatically, quickly, and accurately identify and construct building models from design drawings. Summary of the Invention

[0003] The purpose of the present invention is to provide a building model construction method based on AI drawing recognition.

[0004] In order to achieve the above object, the present invention adopts the following technical solutions:

[0005] The building model construction method based on AI drawing recognition includes the following steps:

[0006] Extract elements from plan drawings, such as walls, doors, windows, and staircase building components, and intelligently identify and classify these elements through AI algorithms;

[0007] During the recognition process, deep learning technology is used to train a large amount of drawing data, enabling the system to accurately understand the symbols, lines, and text information in the drawings;

[0008] Convert it into corresponding elements in the 3D model, and automatically establish the spatial relationship between the elements based on the relative position and size information of the elements and the perspective of the plane drawing, providing a basis for subsequent 3D model construction;

[0009] Train the plane and fix the weights of the trained multi-view geometry encoder as the reference model, and then continue to input multi-view data;

[0010] During training, the diffusion loss focuses on aligning distributions without explicit geometric supervision. The CAD compiler enforces strict command rules, and in some cases, curves do not form closed surfaces, causing the generated CAD code to fail the compiler check.

[0011] In the LLM framework, the reward function is designed to reflect human preferences by comparing model output data. Reinforcement learning techniques are used to align the policy model. A code checker is introduced as an implicit reward model to achieve the denoising effect on the potential code.

[0012] Parse multi-view data, merge data from different perspectives, automatically align and merge them, and generate a 3D model framework.

[0013] Preferably, corresponding symbols are marked on the plan drawings for building components such as walls, doors, windows, and stairs for easy identification; at the same time, specific marks are added to the viewing angles of the drawings to facilitate determination of the direction and position of the drawings.

[0014] Preferably, during the multi-image merging process, the elements and perspectives in each drawing are quickly identified through the above-mentioned symbols and identifiers, thereby enabling rapid classification and information processing.

[0015] Preferably, three-dimensional reconstruction is performed. During the reconstruction process, the model automatically corrects and compensates for geometric inconsistencies caused by perspective differences and error factors, ensuring that the generated three-dimensional model has high accuracy and consistency. At the same time, deep learning algorithms are used to fuse and optimize multi-view data to further improve the detail performance and realism of the three-dimensional model.

[0016] Preferably, the CAD compiler is responsible for parsing and processing CAD drawing files. The compiler has built-in advanced AI algorithms that can intelligently identify various elements in the drawings, such as walls, doors, windows, and stairs, and automatically convert them into corresponding components in the 3D model. Furthermore, the CAD compiler can also recognize annotations and symbols in the drawings, such as dimension annotations and material annotations, ensuring that this critical information is accurately preserved during the 3D reconstruction process. Through the processing of the CAD compiler, the original CAD drawings can be efficiently converted into high-precision and highly consistent 3D building models, providing a solid foundation for subsequent design, construction, and analysis.

[0017] Preferably, the code checker automatically analyzes the generated building model code, identifies and corrects errors and non-compliances therein, and ensures the accuracy and reliability of the model code.

[0018] Preferably, the ambiguity and uncertainty in the drawing data are handled by introducing advanced deep learning technology and algorithm optimization.

[0019] Preferably, when new drawing data is input, the model quickly processes it and updates the existing three-dimensional model based on the re-input plane drawing information, including the viewing angle and elements in the drawing.

[0020] The present invention has at least the following beneficial effects:

[0021] This design significantly improves the efficiency and accuracy of building model construction through automation and intelligent means, while reducing manual intervention and time costs. The use of deep learning technology and algorithm optimization effectively overcomes the ambiguity and uncertainty of drawing data, enhancing the model's ability to analyze and accurately interpret drawing data. The model can rapidly process and update existing 3D models and adapt to input from multiple perspectives and planar drawings, demonstrating exceptional flexibility and adaptability. By introducing reward functions and reinforcement learning techniques, this design guides the strategic model to generate building models that better align with human preferences and expectations, providing more reliable and convenient support for subsequent design, construction, and analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0023] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0025] Reference Figure 1 , a building model construction method based on AI drawing recognition, comprising the following steps:

[0026] First, elements are extracted from the floor plan, such as walls, doors, windows, stairs and other building components. Then, AI algorithms are used to intelligently identify and classify these elements.

[0027] During the recognition process, deep learning technology is used to train a large amount of drawing data, enabling the system to accurately interpret the symbols, lines, and text information in the drawings and perform 3D reconstruction. During the reconstruction process, the model automatically corrects and compensates for geometric inconsistencies caused by perspective differences and error factors, ensuring the generated 3D model is highly accurate and consistent. At the same time, deep learning algorithms are used to fuse and optimize multi-view data, further improving the detail and realism of the 3D model.

[0028] Then, the extracted elements are converted into corresponding elements in the 3D model. Based on the relative position and size information of the elements and the perspective of the plane drawing, the spatial relationship between the elements is automatically established, laying the foundation for the subsequent 3D model construction.

[0029] Train the plane drawings and fix the weights of the trained multi-view geometry encoder as the reference model, then continue to input multi-view data;

[0030] During the training process, diffusion loss is concentrated on aligning distributions without explicit geometric supervision. The CAD compiler enforces strict command rules, and sometimes curves do not form closed surfaces, causing the generated CAD code to fail the compiler check. The CAD compiler is responsible for parsing and processing drawing files in CAD format. It has built-in advanced AI algorithms that can intelligently identify various elements in the drawings, such as walls, doors, windows, stairs, etc., and automatically convert them into corresponding components in the three-dimensional model. In addition, the CAD compiler can also recognize annotations and symbols in the drawings, such as dimension annotations, material annotations, etc., to ensure that these key information is accurately retained during the three-dimensional reconstruction process. Through the processing of the CAD compiler, the original CAD drawings can be efficiently converted into high-precision and high-consistency three-dimensional building models, providing a solid foundation for subsequent design, construction and analysis work;

[0031] In the LLM framework, the reward function is designed to reflect human preferences by comparing model output data, and reinforcement learning techniques are used to align the policy model. A code checker is introduced as an implicit reward model to achieve a denoising effect on the underlying code.

[0032] Parse multi-view data, merge data from different perspectives, automatically align and merge, and generate a 3D model framework.

[0033] On the floor plan, mark the walls, doors, windows, stairs and other building components with corresponding symbols for easy identification; at the same time, add specific marks to the perspective of the drawing to facilitate the determination of the direction and position of the drawing.

[0034] In the process of merging multiple drawings, the above symbols and logos can be used to quickly identify the elements and perspectives in each drawing, thereby enabling rapid classification and information processing;

[0035] Once recognition is complete, the system reconstructs these elements in 3D, based on preset building model construction rules and in accordance with their actual proportions and positional relationships. During the reconstruction process, the algorithm considers various factors, such as wall thickness, door and window size and placement, and staircase structure, to ensure that the resulting building model is both accurate and realistic. Furthermore, the system analyzes the material and texture of the identified elements to more realistically reproduce the building's appearance in the 3D model.

[0036] The code checker automatically analyzes the generated building model code, identifying and correcting errors and non-compliances to ensure the accuracy and reliability of the model code. Through continuous iteration and optimization, the reward function and reinforcement learning techniques gradually guide the policy model to generate building models that better align with human preferences and expectations. Simultaneously, the implicit rewards provided by the code checker also drive the model code towards higher quality. This design not only improves the efficiency and accuracy of building model construction but also provides more reliable and convenient support for subsequent design, construction, and analysis.

[0037] By introducing advanced deep learning technology and algorithm optimization to deal with the ambiguity and uncertainty in drawing data, such as line intersections, occlusions, and perspective distortion under different viewing angles, we can effectively solve these problems and improve the model's ability to understand and accurately grasp drawing data.

[0038] When new drawing data is input, the model quickly processes it and updates the existing 3D model based on the newly input plan drawing information, including perspectives and elements, without having to rebuild it from scratch. This feature makes our model extremely flexible and adaptable in practical applications.

[0039] According to the above working process, the advantages of this design are

[0040] Efficiency: Through automation and intelligent methods, this design significantly improves the efficiency and accuracy of building model construction, reducing manual participation and time costs.

[0041] Accuracy: By leveraging deep learning technology and algorithm optimization, this design effectively addresses the ambiguity and uncertainty issues in drawing data, improving the model’s ability to understand and accurately interpret drawing data.

[0042] Flexibility: The model can quickly process and update existing 3D models, adapt to the input of different perspectives and plane drawing information, and has extremely high flexibility and adaptability.

[0043] Reliability: This design introduces reward functions and reinforcement learning techniques to guide the policy model to generate building models that are more in line with human preferences and expectations, providing more reliable and convenient support for subsequent design, construction, and analysis.

[0044] The core of this model lies in how to effectively integrate drawing data from different viewpoints to ensure that the resulting 3D model is both accurate and complete. To achieve this, we first pre-process the drawing data from each viewpoint using a deep learning algorithm to extract key symbols, lines, and text. This information is then input into a specially designed multi-view geometry encoder, which can understand and parse the data from these different viewpoints, automatically aligning and merging them to produce a unified 3D model framework.

[0045] During this process, we focused on addressing the ambiguity and uncertainty in the drawing data, such as line intersections, occlusions, and perspective distortion under different viewing angles. By introducing advanced deep learning techniques and algorithm optimization, we were able to effectively address these issues and improve the model's understanding and accuracy of the drawing data.

[0046] We've also designed a flexible and efficient data update mechanism for the model. When new drawing data is input, the model can quickly process it and update the existing 3D model without having to rebuild it from scratch. This feature makes our model extremely flexible and adaptable in practical applications.

[0047] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions merely illustrate the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A building model construction method based on AI drawing recognition, characterized in that: The steps include: Extract elements from plan drawings, such as walls, doors, windows, and staircase building components, and intelligently identify and classify these elements through AI algorithms; During the recognition process, deep learning technology is used to train a large amount of drawing data, enabling the system to accurately understand the symbols, lines, and text information in the drawings; Convert it into corresponding elements in the 3D model, and automatically establish the spatial relationship between the elements based on the relative position and size information of the elements and the perspective of the plane drawing, providing a basis for subsequent 3D model construction; Train the plane and fix the weights of the trained multi-view geometry encoder as the reference model, and then continue to input multi-view data; During training, the diffusion loss focuses on aligning the distribution without explicit geometric supervision. The CAD compiler enforces strict command rules, and in some cases the curves do not form a closed surface, causing the generated CAD code to fail the compiler check. In the LLM framework, the reward function is designed to reflect human preferences by comparing model output data. Reinforcement learning techniques are used to align the policy model. A code checker is introduced as an implicit reward model to achieve the denoising effect on the potential code. Parse multi-view data, merge data from different perspectives, automatically align and merge them, and generate a 3D model framework.

2. The building model construction method based on AI drawing recognition according to claim 1 is characterized in that: On the floor plan, mark the walls, doors, windows, stairs and other building components with corresponding symbols for easy identification; at the same time, add specific marks to the perspective of the drawing to facilitate the determination of the direction and position of the drawing.

3. The building model construction method based on AI drawing recognition according to claim 2 is characterized in that: In the process of merging multiple drawings, the above symbols and logos can be used to quickly identify the elements and perspectives in each drawing, thereby enabling rapid classification and information processing.

4. The building model construction method based on AI drawing recognition according to claim 1 is characterized in that: During the 3D reconstruction process, the model automatically corrects and compensates for geometric inconsistencies caused by perspective differences and error factors, ensuring that the generated 3D model has high accuracy and consistency. At the same time, deep learning algorithms are used to fuse and optimize multi-view data to further enhance the detail and realism of the 3D model.

5. The building model construction method based on AI drawing recognition according to claim 1 is characterized in that: The CAD compiler is responsible for parsing and processing CAD drawing files. Equipped with advanced AI algorithms, the compiler intelligently identifies various elements in the drawings, such as walls, doors, windows, and stairs, and automatically converts them into corresponding components in the 3D model. Furthermore, the CAD compiler recognizes annotations and symbols in the drawings, such as dimensions and material annotations, ensuring that this critical information is accurately preserved during the 3D reconstruction process. Through the CAD compiler's processing, original CAD drawings can be efficiently converted into highly accurate and consistent 3D building models, providing a solid foundation for subsequent design, construction, and analysis.

6. The building model construction method based on AI drawing recognition according to claim 1 is characterized in that: The code checker automatically analyzes the generated building model code, identifies and corrects errors and non-compliances, and ensures the accuracy and reliability of the model code.

7. The building model construction method based on AI drawing recognition according to claim 1 is characterized in that: By introducing advanced deep learning technology and algorithm optimization, the ambiguity and uncertainty in drawing data are handled.

8. The building model construction method based on AI drawing recognition according to claim 1 is characterized in that: When new drawing data is input, the model quickly processes it and updates the existing 3D model based on the re-input plane drawing information, including the perspective and elements in the drawing.

Images