Method for generating building information modeling by using 2d drawing

Abstract

The present invention relates to a building information modeling (BIM) generation method using a 2D drawing, which generates BIM by learning a 2D drawing of a building. The method includes the steps of: inputting an image of a 2D drawing and drawing information of a building to a main server through a user terminal; preprocessing the image of the 2D drawing of the building; learning metadata in which object information and attribute information are mapped by recognizing an object in the drawing by using an already trained deep learning model from the preprocessed image of the 2D drawing; applying a semantic segmentation model of segmenting components of each object at a pixel level by using a semantic segmentation algorithm so as to learn the position and boundary of the object and detect the object; recognizing and extracting measurement information including information outside the drawing and text from the preprocessed image of the 2D drawing; converting the measurement information through optical character recognition (OCR) and digitizing same; integrating and mapping learned data to convert components of the object into 3D; and generating BIM by using the measurement information and the recognized components of the object.

Description

Method for creating architectural information modeling using 2D drawings

[0001] The present invention relates to a method for generating building information modeling (BIM) using 2D drawings by learning 2D drawings of a building and generating BIM.

[0002] Building Information Modeling (BIM) technology is a technology that creates and utilizes various building information throughout the entire life cycle of a building, thereby improving efficiency across the entire construction industry, including design, engineering, construction, and maintenance.

[0003] These BIMs can contain not only 3D visual information about buildings and their components, but also various attribute information. This building information can be interpreted in a computer environment, enabling a variety of applications. Representative applications include automated review of building design quality, quantity estimation, constructability review, and building energy simulation.

[0004] Among the aforementioned BIM applications, BIM-based automated building design quality review technology quantitatively and automatically reviews design requirements, such as permitting regulations and design guides, for architectural designs. This complements the time-consuming and labor-intensive design quality review process previously performed manually by architects and building code experts, enabling more efficient, rapid, and accurate performance.

[0005] However, in the practical process of implementing actual BIM design plans, problems often arise where accurate review is not conducted. One cause of these problems is the quality of BIM data, where the BIM model does not contain the data required for review or is poorly defined.

[0006] In addition, there is a problem that since aging houses exist in the form of drawings in the form of images, the drawings must be manually re-created and BIM created using these, which causes a burden of time and money.

[0007] The present invention provides a method for creating building information modeling (BIM) using 2D drawings, which classifies and recognizes objects of a drawing from a 2D drawing image, recognizes information outside of the drawing, and utilizes it for smart maintenance through digital architecture, analysis for constructability, various simulations, and visualization.

[0008] The technical problems to be solved by the present invention are not limited to the technical problems mentioned above, and other technical problems not mentioned can be clearly understood by a person having ordinary skill in the technical field to which the present invention belongs from the description below.

[0009] The method for generating building information modeling (BIM) using a 2D drawing of the present invention for achieving the above object comprises the steps of: inputting a 2D drawing image and drawing information of a building into a main server through a user terminal; preprocessing the 2D drawing image of the building; recognizing an object in the drawing using a deep learning model that has already been trained from the preprocessed 2D drawing image and learning metadata that maps object information and attribute information; applying a semantic segmentation model that segments components of each object at the pixel level using a semantic segmentation algorithm to learn the location and boundary of the object and detect the object; recognizing and extracting measurement information including non-drawing information and text from the preprocessed 2D drawing image; converting the measurement information into digital information using OCR (Optical Character Recognition) and digitizing it; integrating and mapping learning data to 3Dize components of the object; and generating a BIM (Building Information Modeling) using the measurement information and the recognized components of the object.

[0010] In addition, the method further includes a step of comparing the BIM with a 2D drawing image of the building to detect an error, wherein the type of the error is characterized by at least one of an error in position mismatch of the object, an error in size and shape, and an error in missing or erroneous information.

[0011] In addition, when an error in the BIM is detected, the method further comprises a step of automatically adjusting the BIM based on a 2D drawing image of the building according to the type of the error.

[0012] In addition, the drawing information is characterized in that it includes the type of drawing and the type of building of the 2D drawing image.

[0013] In addition, the step of preprocessing the floor plan image of the building is characterized by filtering the drawing image, adjusting a threshold value, and removing noise.

[0014] Additionally, the object is characterized by being divided into space, object, and structure.

[0015] In addition, the measurement information is characterized by being at least one of dimension information, material property information, distance information, scale information, and name information related to the object.

[0016] In addition, the semantic segmentation model step is characterized by including a step of designating a segmentation mask according to pre-specified information; a step of identifying an object by combining a bounding box, which is attribute information corresponding to the pixel; and a step of learning by combining the segmentation mask and the bounding box.

[0017] As explained above,

[0018] The present invention has the effect of being able to utilize smart maintenance through digital architecture, analysis for constructability, various simulations and visualizations, etc. by classifying and recognizing objects in drawings from a 2D drawing image of the building for which no drawing in image file format exists, and recognizing information outside of the drawing.

[0019] In addition, the present invention has the effect of saving time and cost by creating BIM by digitizing 2D drawing images.

[0020] FIG. 1 is a flowchart illustrating a method for creating building information modeling (BIM) using 2D drawings according to an embodiment of the present invention.

[0021] FIG. 2 is a flowchart illustrating a deep learning method for each data set in a method for creating building information modeling (BIM) using 2D drawings according to an embodiment of the present invention.

[0022] FIG. 3 is a drawing showing a structural element extracted in a method for creating building information modeling (BIM) using a 2D drawing according to an embodiment of the present invention.

[0023] FIG. 4 is a drawing showing a spatial element extracted in a method for creating building information modeling (BIM) using a 2D drawing according to an embodiment of the present invention.

[0024] FIG. 5 is a drawing showing information outside of a drawing in a method for creating building information modeling (BIM) using a 2D drawing according to an embodiment of the present invention.

[0025] FIG. 6 is a drawing showing a BIM layer generated by a method for generating building information modeling (BIM) using a 2D drawing according to an embodiment of the present invention.

[0026] FIG. 7 is a drawing showing a BIM generated by a method for generating building information modeling (BIM) using a 2D drawing according to an embodiment of the present invention.

[0027] Hereinafter, embodiments of the present invention will be described in detail with reference to the attached drawings.

[0028] In addition, the size and shape of components illustrated in the drawings may be exaggerated for clarity and convenience of explanation, and terms specifically defined in consideration of the configuration and operation of the present invention may vary depending on the intention or custom of the user or operator, and the definitions of these terms should be made based on the contents throughout this specification.

[0029] FIG. 1 is a flowchart illustrating a method for creating building information modeling (BIM) using 2D drawings according to an embodiment of the present invention.

[0030] As illustrated in Fig. 1, a method for creating building information modeling using a 2D drawing according to an embodiment of the present invention first inputs a 2D drawing image and drawing information of a building into a main server through a user terminal (S100).

[0031] The user terminal may be a terminal provided with a program or application that can input 2D drawing images and drawing information into the main server, and the method of connection between the user terminal and the main server may be an indirect connection or a direct connection through a network, and is not limited thereto.

[0032] In addition, the present invention shows a method for creating building information modeling (BIM) using 2D drawings, and a systematic description thereof is omitted.

[0033] The drawing information entered into the main server includes at least one of the drawing type of the building drawing image, the building type, and the floor information.

[0034] For example, the type of drawing refers to what the 2D drawing image represents, such as a floor plan, a cross-section, an elevation, and a structural drawing, and the type of building can be categorized by the building's purpose, such as an apartment, a single-family home, a townhouse, and a commercial building. Furthermore, the floor information can further include information about the floor, such as which floor the 2D drawing image represents.

[0035] Next, the 2D drawing image of the building is preprocessed (S200). At this time, the preprocessing of the 2D drawing image includes processes such as filtering, threshold adjustment, and noise removal to clearly identify the image.

[0036] Next, deep learning is performed for each data set (S300). This recognizes objects within the drawing using a preprocessed 2D drawing image using a pre-trained deep learning model, and learns metadata that maps object information and attribute information.

[0037] At this time, as illustrated in Figure 2, the deep learning learning method for each data set is as follows. First, objects are classified from a 2D drawing image (S310). Objects are classified into structure, space, object, and non-drawing information.

[0038] Here, structure, space, and objects are depicted in the image of the drawing and correspond to information within the drawing, and information outside the drawing refers to additional information of the drawing, such as information in the form of letters and dimension lines included in the 2D drawing image, and all information other than structure, space, and objects.

[0039] Each data set is trained separately according to the classified objects.

[0040] First, in the case of a structure among objects, the structure is recognized based on information already learned from a 2D drawing image (S311), and the class according to the structure is classified (S312).

[0041] The class according to structure recognizes the structure of a wall (310), a door (320), and a window (330), as shown in Fig. 3.

[0042] Here, the door (320) and window (330) can be recognized by subdividing them according to type.

[0043] For example, doors can be subdivided into sliding doors and swing doors, and windows can be subdivided into sliding windows and swing windows.

[0044] Then, a semantic segmentation model that segments each object's components at the pixel level using a semantic segmentation algorithm is applied to learn the location and boundary of the object (S313) and detect the object (S314).

[0045] Semantic segmentation labels every pixel in an image.

[0046] In other words, semantic segmentation is a method of extracting objects of interest on a pixel-by-pixel basis. When you want to know where an object is located in an image, what shape that object is, which pixel belongs to which object, etc., you segment the image and label each pixel in the image.

[0047] In more detail, the semantic segmentation model of the semantic segmentation algorithm specifies a segmentation mask according to pre-specified information, identifies an object by combining a bounding box, which is attribute information corresponding to a pixel, and learns by combining the segmentation mask and the bounding box.

[0048] At this time, learning to recognize the location and boundaries of objects is performed using the Deeplap algorithm.

[0049] Deeplab v3+ itself focuses primarily on semantic segmentation and does not directly provide object detection capabilities. Therefore, it is desirable to design a multi-task learning structure that combines object detection models with deeplab v3+.

[0050] This is a form of combining information for object detection with segmentation location information of the architectural structure.

[0051] After recognizing the location and boundaries of the object, object detection is performed.

[0052] At this time, the Yolov5 (You only look once) algorithm is used to detect the object and guess the type and location of the object.

[0053] In this way, the results of semantic segmentation and object detection are combined (S315).

[0054] More preferably, the results of object detection are filtered using the positions and boundaries of structural elements to extract only objects related to the structural elements, or the results are corrected to obtain accurate position and boundary information of objects.

[0055] Additionally, in the case of space, the space is recognized from a 2D drawing image (S321) and the class according to the space is classified (S322).

[0056] As shown in Figure 4, the classes according to space are recognized by classifying the space into bedroom, bathroom, dressing room, living room, balcony, kitchen, entrance, multipurpose space, outdoor machine room, stairwell, elevator, and elevator hall.

[0057] For semantic segmentation of space, learning to recognize the location and boundaries of spatial objects is performed using the deeplap algorithm (S323), and spatial object detection is performed using Yolov5 (S324).

[0058] The results of space segmentation and object detection are combined (S325).

[0059] Additionally, for objects, objects are recognized from 2D drawing images (S331) and classes are classified according to objects (332).

[0060] Classes by object classify and recognize objects such as sinks, gas ranges, washbasins, toilets, and bathtubs.

[0061] For semantic segmentation of objects, learning to recognize the location and boundaries of objects is performed using the deeplap algorithm (S333), and spatial object detection is performed using Yolov5 (S334).

[0062] The results of object segmentation and object detection are combined (S335). In the embodiments of the present invention, the data set is described in the order of structure, space, and object for illustrative purposes. However, learning priorities are not a consideration, and the data integration process is performed after learning is completed for each independent node.

[0063] Lastly, among the data sets, measurement information including non-drawing information and text is recognized and extracted from the preprocessed 2D drawing image (S341).

[0064] At this time, the measurement information includes at least one of dimensional information, material property information, distance information, scale information, and name information related to the object.

[0065] To recognize measurement information, text and numbers are extracted (S342), and symbols and patterns indicating dimension lines are extracted (S343).

[0066] That is, the form of measurement information includes all forms used to describe the drawing, such as letters, numbers, pointers, dimension lines, symbols, and patterns.

[0067] For example, as illustrated in FIG. 5, information outside of the drawing can be extracted from a 2D drawing image, and the dimensions of each part can be identified according to the dimension line (510) illustrated on the left side of the drawing, and bedroom 1 described in a square box (520) described in the drawing corresponds to attribute information representing a space among objects.

[0068] Additionally, the text (530) described according to the pointer in the drawing indicates the property information of the finishing material corresponding to the area indicated by the pointer.

[0069] To digitize measurement information included in non-drawing information, OCR conversion is performed (S344), digitization is performed, and text and numbers corresponding to symbols and patterns are matched (S345).

[0070] To 3Dize the components of an object, learning data learned by the data set is integrated and mapped (S400).

[0071] That is, it maps structural elements to 3D the recognized objects and segmented components, which is intended to be expressed as a 3D scene in the field of computer graphics.

[0072] Consequently, the learned data of the data set according to each independent node is integrated by mapping the relationships between structure, space, objects, and extra-drawing information to combine them into a single model.

[0073] For example, when creating an architectural framework using learning data according to structure, a data structure that can be created by mapping measurement information and attribute information of drawings through OCR is formed.

[0074] This integration allows for a detailed and accurate representation of the building in terms of its physical dimensions and spatial relationships.

[0075] Finally, a Building Information Modeling (BIM) is created using the measurement information and recognized object components (S500). The created BIM can then be output via a 3D engine.

[0076] Additionally, the generated BIM data can be used in accordance with standards such as architectural libraries.

[0077] FIG. 6 is a 3D drawing showing a BIM layer created by a method for creating building information modeling (BIM) using 2D drawings according to an embodiment of the present invention, and FIG. 7 is a drawing showing a BIM created by a method for creating building information modeling (BIM) using 2D drawings according to an embodiment of the present invention.

[0078] Fig. 6 is a BIM fault image generated according to the input of a 2D drawing image, and is generated as a 3D image showing the structure and space of a building. Fig. 7 (a) shows a front view generated as a 3D image in which the BIM faults shown in Fig. 6 are gathered together to form a single building, and Fig. 7 (b) shows a back view generated as a 3D image in which the BIM faults shown in Fig. 7 are gathered together to form a single building.

[0079] The single-layer structure illustrated in Figure 6 is assembled to form multiple layers, the upper portion of which is formed into a different structure and space. In other words, a 2D drawing image can be created into a 3D image, and these can be stacked in multiple layers to form a building shape.

[0080] Additionally, BIM can be compared with 2D drawing images of a building to detect errors, and detected errors can be improved manually or automatically.

[0081] The method of detecting errors in the generated BIM defines in advance the errors that may occur in the generated BIM.

[0082] Here, the type of error refers to one or more of the following: object position mismatch errors, size and shape errors, and information omission or errors. For example, the type of error may be an error depending on the object being extracted, such as structural element position mismatches, size and shape errors of spaces and objects, and information omission or errors in drawings.

[0083] In addition, the BIM error detection method aligns the generated BIM on top of the input original 2D drawing image, overlays it, and compares it to detect errors.

[0084] At this time, the BIM is resized, rotated, and translated to match the scale and orientation of the 2D drawing image.

[0085] Next, key elements such as structural components that form the framework, such as walls, doors, and windows in BIM and 2D drawing images, are identified using pattern recognition or vision algorithms, and the mapped features are extracted.

[0086] Next, spatial analysis is performed using the extracted features to compare the geometric characteristics of the feature, such as location, direction, length, and area, in both BIM and 2D drawing images.

[0087] At this time, for efficient comparison, comparison is made using algorithms such as geometric hashing and spatial indexing that have already been classified based on the BIM standard library.

[0088] Finally, errors are detected by analyzing the differences in attributes to identify inconsistencies.

[0089] For example, if the wall location in BIM deviates from the acceptable threshold in the 2D drawing image, it is detected and flagged as an error.

[0090] Additionally, mismatches detected in BIM are classified by error criteria such as positional errors, dimensional errors, missing elements, or additional elements not present in the original 2D drawing image. In order for a mismatch to be classified as an error, if it exceeds a predetermined threshold, it is desirable to pre-establish a standard for the tolerance level in the architectural design.

[0091] In the above way, when an error in BIM is detected, the BIM can be automatically adjusted based on the 2D drawing image of the building depending on the type of error, and the automatically adjusted detected error can be corrected by adjusting the position, size (dimension), and property of the structure for the detected error, which improves the design to match the BIM with the original drawing to maintain structural integrity and design consistency.

[0092] In an embodiment of the present invention, a machine learning algorithm is used to continuously improve object recognition from a 2D drawing image, BIM generated by recognizing the object, and error detection in BIM, thereby increasing the accuracy of BIM generation and updating related parameters to increase the reliability of BIM.

[0093] Accordingly, the present invention has the effect of being able to utilize smart maintenance through digital architecture, analysis for constructability, various simulations and visualizations, etc. by classifying and recognizing objects in drawings from a 2D drawing image of the building for which no drawing in image file format exists, and recognizing information outside of the drawing.

[0094] In addition, the present invention has the effect of saving time and cost by creating BIM by digitizing 2D drawing images.

[0095] While the embodiments of the present invention have been described above, they are merely exemplary, and those skilled in the art will understand that various modifications and equivalent embodiments are possible. Accordingly, the true technical protection scope of the present invention should be defined by the following claims.

[0096] The present invention relates to a method for generating building information modeling (BIM) using 2D drawings by learning 2D drawings of a building and generating BIM.

[0097] The present invention has the effect of being able to utilize smart maintenance through digital architecture, analysis for constructability, various simulations and visualizations, etc. by classifying and recognizing objects in drawings from a 2D drawing image of the building for which no drawing in image file format exists, and recognizing information outside of the drawing.

Claims

1. A step of inputting a 2D drawing image and drawing information of a building into the main server through a user terminal; A step of preprocessing a 2D drawing image of the above building; A step of learning metadata that maps object information and attribute information by recognizing objects within a drawing using a deep learning model that has already been trained from a preprocessed 2D drawing image; A step of learning the location and boundary of an object and detecting the object by applying a semantic segmentation model that segments the components of each object at the pixel level using a semantic segmentation algorithm; A step of recognizing and extracting measurement information including non-drawing information and text from the above preprocessed 2D drawing image; A step of converting the above measurement information into digital information using OCR (Optical Character Recognition); A step of integrating and mapping learning data to 3Dize components of the above object; and A method for creating a BIM using a 2D drawing, characterized by comprising the steps of: creating a BIM (Building Information Modeling) using the above measurement information and components of the recognized object; 2. In claim 1, Further comprising a step of detecting errors by comparing the BIM with a 2D drawing image of the building, A method for creating BIM using a 2D drawing, characterized in that the above types of errors are at least one of position mismatch errors of the object, size and shape errors, and omission or error errors of information.

3. In claim 2, A method for creating a BIM using a 2D drawing, characterized in that it further includes a step of automatically adjusting the BIM based on a 2D drawing image of the building according to the type of the error when an error in the BIM is detected.

4. In claim 1, A method for creating BIM using a 2D drawing, wherein the above drawing information includes the type of drawing and the type of building of the 2D drawing image.

5. In claim 1, A method for creating a BIM using a 2D drawing, characterized in that the step of preprocessing the plan drawing image of the above building includes filtering the drawing image, adjusting a threshold value, and removing noise.

6. In claim 1, A method for creating BIM using a 2D drawing, characterized in that the above objects are divided into spaces, objects, and structures.

7. In claim 1, A method for creating a BIM using a 2D drawing, wherein the measurement information includes at least one of dimension information, material property information, distance information, scale information, and name information related to the object.

8. In claim 1, The above semantic segmentation model step is a step of specifying a segmentation mask according to pre-specified information; A step of identifying an object by combining a bounding box, which is attribute information corresponding to the above pixel; and A method for creating a BIM using a 2D drawing, characterized in that it comprises a step of learning by combining the segmentation mask and the bounding box.

Images