Automatic floor plan creation (tracing) system using artificial intelligence (AI)
The AI-based floor plan creation system addresses the inefficiencies of conventional methods by automating the process, reducing time and errors, and ensuring standardized, high-quality output.
Patent Information
- Application Number
- JP2020165433
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2020-09-30
- Publication Date
- 2024-12-05
- Estimated Expiration
- 2040-09-30
AI Technical Summary
Conventional floor plan creation systems require significant human resources, are time-consuming, and prone to errors, with non-standardized formats leading to inconsistencies in real estate property representations.
An automated floor plan creation system using AI that includes a content management device, information extraction device, style conversion device, verification device, and machine learning device to quickly generate accurate floor plans by extracting and converting raster images into vector images while adhering to predefined specifications.
The system significantly reduces creation time and minimizes human errors, enabling rapid production of numerous high-quality floor plans with standardized specifications.
Smart Images

Figure 0007594727000001 
Figure 0007594727000002 
Figure 0007594727000003
Abstract
Description
[Technical Field]
[0001] This invention relates to a system that uses artificial intelligence (AI) to automatically create (trace) floor plans for real estate. [Background technology]
[0002] A real estate floor plan is a drawing used to show and describe a real estate property such as a building or apartment, containing both geometric (recognizable by its shape, e.g., the structure of the rooms) and semantic (meaningful) information (e.g., door types, building materials, fixtures, etc.).
[0003] Currently, floor plans are made publicly available as raster images (bitmap images).
[0004] Floor plans serve the purpose of visualizing real estate properties for real estate agents, buyers, and renters, and are used for property advertising. Therefore, when advertising properties, many real estate companies create (traces) their own floor plans based on architectural drawings of the properties they handle or floor plans created by other companies that abstract architectural drawings for advertising purposes. The original floor plans are often scanned PDFs of printed plans or JPEG data posted online, and in most cases are raster images. When automatically creating (tracing) floor plans using artificial intelligence (AI), the raster images are converted into vector images (a method of representing images and text on a computer using parameters related to geometric elements such as lines and surfaces), creating a format suitable for automatic floor plan creation (tracing) using AI.
[0005] The conversion of real estate floor plans into vector images is based on a variety of raster images with different representations created independently by various architecture and real estate companies, and involves extracting relevant information from the raster images and reconstructing them according to the desired design and specifications.
[0006] In the real estate industry, when publishing (advertising) information about actual real estate properties, floor plans must be created with a certain level of accuracy. For example, the indications of entrances and bathrooms must be consistent across the entire system. In addition, unnecessary information that is not relevant to understanding the property's floor plan must be removed. Unnecessary information, such as handwritten notes that remain on the original floor plan, must also be removed.
[0007] However, the formats and specifications of floor plans provided by real estate agents vary from company to company and are not standardized. For this reason, when automatically creating (tracing) floor plans, a process is required to accurately read the various types of floor plans and standardize them (unify the display). Examples of systems for creating floor plans for real estate include the following cloud-based and software-based floor plan creation systems, as described in non-patent literature at the URL below. [Prior art documents] [Patent documents]
[0008] [Non-Patent Document 1] https: / / madori.jp / [Non-patent document 2] http: / / www.lilacsys.com / mad5 / Summary of the Invention [Problem to be solved by the invention]
[0009] However, with conventional systems, the act of creating floor plans itself is human-reliant, making it difficult to create large quantities at once in a short amount of time, and it takes a long time to create the floor plans.In addition, there are problems with human error when creating floor plans, such as not following the specifications and rules set by each real estate company in advance.
[0010] The present invention has been made in consideration of the above circumstances, and aims to provide a floor plan creation (tracing) system that can quickly create a large number of floor plans while minimizing human error. [Means for solving the problem]
[0011] In order to solve the above problem, the automatic floor plan creation (tracing) system of claim 1 comprises: a content management device for inputting floor plans; an information extraction device that extracts information necessary for creating an output floor plan from the input floor plan input to the content management device and groups the information to form an intermediate product; a style conversion device that converts the intermediate product created by the information extraction device into a vector image to obtain the output floor plan; the input floor plan, the output floor plan, Information required to create an output floor plan a record storage device for storing past verification results and corrections to the output floor plan; a verification device for verifying whether the output floor plan has been converted in accordance with a predetermined floor plan creation specification, determining the quality of the output floor plan, and correcting the output floor plan if the quality is not sufficient; a machine learning device that learns from the verification and correction by the verification device and improves the accuracy of the output floor plan; the information extraction device includes three deep learning devices, namely, a floor plan skeleton recognition device, a symbol recognition device, and an optical character recognition device, and the three deep learning devices extract skeleton information, symbol information representing facility information, and text information as information necessary for creating the output floor plan, and group them to form the intermediate product, which is then transferred to the style conversion device; the style conversion device comprises: a logical inference device that extracts basic elements of a structured vector image, which is a collection of points and lines connecting points, from the intermediate product created by the information extraction device and facility information and character information linked to the intermediate product; and a decorator device that re-decorates the floor plan by applying specifications specific to specific symbols and text to complete the conversion of the input floor plan; an output of the machine learning device is input to the floor plan skeleton recognizer, the symbol recognizer, and the optical character recognizer of the information extraction device; The machine learning device trains the three deep learning devices to extract information with higher accuracy, thereby improving the accuracy of the three deep learning devices.
[0012] An automated floor plan tracing system using artificial intelligence (AI) has been implemented to assist the manual creation process using traditional floor plan creation software, thereby changing the role of the creator. The system includes steps to receive a floor plan, automatically convert it into the desired style, and receive feedback on the resulting floor plan (output floor plan). Rather than drawing everything from scratch, floor plan creators simply run quality assurance checks on the output of the AI system and make changes as needed, significantly reducing the workload and time required for each floor plan creation. Includes a memory mechanic to save feedback / changes, further improving the AI system to address past auto-creation mistakes.
[0013] The floor plan creation method according to claim 2 comprises: inputting a floor plan via a content management means; - extracting information from the input floor plan; - a step of performing style conversion; - a record-keeping process; - a step of performing verification; -Machine learning process It is characterized by having: [Effects of the Invention]
[0014] This allows you to drastically reduce the creation time and quickly create a large number of floor plans at once without the information loss that is required when automatically creating floor plans. [Brief explanation of the drawings]
[0015] [Figure 1] This is a block diagram showing the configuration and overall flow of a system for automatically creating (tracing) floor plans using artificial intelligence (AI). [Figure 2] This is a block diagram showing the interface that connects a client with an automatic floor plan creation (tracing) system that uses artificial intelligence (AI). [Figure 3] This is a block diagram showing three information recognition models that read input floor plan information. [Figure 4] This figure shows (1) the input floor plan (F), (2) the skeleton floor plan (S) output from the skeleton recognition device 310, (3) the symbol-recognized floor plan, and (4) the optical character-recognized floor plan. [Figure 5] This is a block diagram explaining the style conversion device that integrates the read floor plan information and converts it into a floor plan with a new design. [Figure 6] This is a block diagram showing a floor plan data storage device including two different purpose databases, a content management database and a learning database, and their relationships. [Figure 7] This is a block diagram showing the verification interface, correction interface, verifier, and verification device, showing their relationships with each other. [Figure 8] This is a block diagram showing the relationship between the machine learning device, the training database, and the three information extraction models.
[0016] Below, we will explain the entire invention and the process of automatically creating (tracing) floor plans using artificial intelligence (AI), with reference to the attached drawings.
[0017] Figure 1 shows an overview of an automatic floor plan creation (tracing) system 100 using artificial intelligence (AI).
[0018] The device 100 includes three asynchronous tasks surrounding the database: autonomous (automatic) style conversion using an AI system, manual verification by humans, and an AI learning process. Note that the verification task is essential to improve the accuracy of the automated generation of real estate floor plans.
[0019] The asynchronous nature of the three tasks is due to differences in processing time and characteristics, which enables the simultaneous mass processing (conversion) of many floor plans.
[0020] First, the style conversion task for a floor plan begins with a client, such as a real estate company, inputting the floor plan via the content management device 200. Then, the floor plan information is transmitted in order to the information extraction device 300 and the style conversion device 400.
[0021] The verification device 600 and the machine learning device 700 are components that handle verification and machine learning training, and the record storage device 500 handles record storage.
[0022] The content management device 200 is an interface through which users (clients such as real estate agents) communicate (interact) with the system.
[0023] The information extraction device 300 processes the floor plan input to the system as follows, and extracts the relevant information necessary for automatic floor plan creation (tracing).
[0024] The extraction of relevant information is carried out in the following three steps:
[0025] The first step (function) is to recognize the skeleton (i.e., sketch / property structure).
[0026] The skeleton recognition function extracts structural information from the floor plan on a pixel-by-pixel basis. The floor plan input from the content management device 200 is drawn as a raster image made up of a collection of square pixels, but the skeleton recognition function reads only structural information such as walls, indoors, and outdoors (where is a wall, where is indoors, and where is outdoors) as pixel information and performs pixel-to-pixel mapping.
[0027] Note that floor plan information other than structural information is ignored in this step.
[0028] Next is the symbol recognition function, which identifies technical symbols representing facility information (doors, windows, toilets, sinks, kitchens, etc.) on the floor plan that were ignored by the first skeleton recognition function. When the symbol recognition function recognizes facility information, it identifies the type and size of the related fixture information, as well as its coordinate position on the floor plan, and extracts the information.
[0029] The last step is optical character recognition (OCR), which converts text into a string of characters.
[0030] This function is responsible for recognizing and guessing letters, words, and lines.
[0032] On the one hand, this style transfer device 400 performs inference on the intermediate artifacts and extracts the information needed to construct vector images (points, paths, etc.).
[0033] Additionally, the extracted information is correlated with each other as described above to restore the information hierarchy (e.g., matching walls to rooms, doors to walls, etc.).
[0034] The record storage device 500 stores the input and output floor plans and related information, as well as past verification results and modifications to the output floor plans.
[0035] The recorded and saved information not only serves as learning data to train the AI-based automatic floor plan creation (tracing) system, but can also be used to track the progress of floor plan conversion work and estimate work times.
[0036] The verification device 600 verifies whether the automated floor plan creation (tracing) system using artificial intelligence (AI) has converted the input floor plan in accordance with the pre-specified floor plan creation specifications. This verification also includes manual verification.
[0037] Once the validation results are available, the floor plan verifier will make any necessary corrections to the output floor plan.
[0038] The verification results and corrections to the output floor plan are recorded to help improve the information extraction accuracy in the three steps of the information extraction device 300 described above.
[0039] The machine learning device 700 performs training so that the three functions mentioned above, namely, the skeleton recognition function, the symbol recognition function, and the optical character recognition function, can extract information with higher accuracy.
[0040] Although the content learned by each function differs depending on its role, they extract and review information using the same input floor plan as a reference source. This machine learning device 700 improves the accuracy of the entire system for automatically creating (tracing) floor plans using artificial intelligence (AI), and it continues to learn over time so that it can handle a variety of input floor plans.
[0041] 2 is a block diagram showing a client, such as a real estate agent, communicating with an artificial intelligence (AI)-based floor plan automatic creation (tracing) system 100. The content management device 200 includes, for example, functions for inputting (uploading) floor plans, displaying verification results, and other management-related functions (e.g., order history, output floor plan download history).
[0042] The content management device 200 can be connected to a client 201 such as a real estate agent, and is connected to the information extraction device 300 and the record storage device 600 via a user interface 210 .
[0043] When a client sends an input floor plan (F) to the information extracting device 300 via the user interface 210, the information extracting device 300 forwards a floor plan record (R) to the record storage device 500, while notifying the client 201 that the processing of the input floor plan (F) is for that particular client.
[0044] The progress of the processing can be made public to the client 201 by searching the storage history (R) in the record storage device 500.
[0045] Once the processing is complete, the results are updated in the record (R) in the record storage device 500 and communicated to the client 201 via the user interface 210.
[0046] The same procedure is followed for creating the input floor plan for another client 201.
[0047] Interaction records relating to the user (client 201), including both input and output floor plans, can be searched and retrieved from record storage device 500 as needed.
[0048] As shown in FIG. 3 , the information extraction device 300 includes three deep learning devices: a floor plan skeleton recognition device 310, a symbol recognition device 320, and an optical character recognition (OCR) device 330. These three devices are connected to the content management device 200 via a user interface 210, and are also connected in parallel to a style conversion device 400.
[0049] The information extracting device 300 is provided with the floor plan (F) input by the client from the content management device 200.
[0050] The above three devices extract and group the information necessary to create the output floor plan, form an intermediate product (I), and transfer it to the style conversion device 400.
[0051] When the floor plan (F) input by the client 201 is sent from the content management device 200, the skeleton recognition device 310 extracts and retains only the structural information of the building, while discarding most of the detailed information such as facilities.
[0052] Specifically, this process is a pixel mapping operation that replaces an input floor plan raster image (F) with another raster image (S).
[0053] If you think of the input floor plan raster image (F) as a grid (draft), the mapping draws the floor plan image (S) by overwriting it with pixels while referring to the grid.
[0054] This process creates a floor plan (S) that looks like it is tracing the input floor plan (F), and as a result, walls are drawn, making it possible to distinguish between indoor and outdoor areas (see Figure 4).
[0055] By looking at this output skeleton floor plan (S), you can understand the relative structural information of the floor plan, such as the building structure, number of rooms, and their locations.
[0056] The symbol recognizer 320 processes the input floor plan (F) in a different way than the skeleton recognizer, focusing on symbols representing fixtures (e.g., sinks, bathrooms, kitchens) or openings (e.g., doors, windows) (see Figure 3).
[0057] The information extracted by the symbol recognizer 320 is in the form of a set of coordinates and an object type (symbol of equipment, opening, etc.). Specifically, the set of coordinates indicates the boundary of a rectangular box containing the object, from which both the position and size of the object itself can be inferred (see Figure 3).
[0058] The optical character recognition (OCR) device 330 mainly extracts text information from the input floor plan (F) (see Figure 4).
[0059] Each character is recognized along with its coordinates in a similar manner to the way information is extracted by the symbol recognizer 320 (see Figure 4).
[0060] The optical character recognition (OCR) device 330 recognizes a single character that it has read as a word or a sentence by grouping it with surrounding single characters. Multiple characters form a word, multiple words form a line, and multiple lines form a block (which indicates the type of room, such as a Western-style room, Japanese-style room, or living room, or the number of tatami mats that indicate the size) (see Figure 4).
[0061] Because the sole purpose of the information extraction device 300 is to extract the information necessary to perform the conversion from the raster image of the input floor plan (F) to the output floor plan (I), the output floor plan (I) (which is the combined output from the three devices 310, 320, and 330) cannot be directly used to generate a vector image floor plan.
[0062] Therefore, the style conversion device 400 is responsible for converting the intermediate product (I), which combines the floor plan information extracted from the three devices 310, 320, and 330, into a vector image.
[0063] The style conversion device 400 consists of two main steps: a logic reasoner 410 and a decorator device 420.
[0064] As can be seen, the logical reasoner 410 and the decorator 420 are continuously connected, while also being connected to the archival device 500 .
[0065] The main purpose of the logical reasoner 410 is to extract the basic elements of a structured vector image, which is a collection of points and lines connecting the points, from the intermediate product (I) created by the information extraction device 300 and the equipment information and text information associated therewith.
[0066] Incidentally, this conversion from a raster image to a vector image is performed intuitively so as to draw the same structure as the skeleton structure of the floor plan, and the resulting image is similar to the skeleton (structure) (S) of the floor plan of the raster image extracted by the skeleton extraction device 310. However, the image that is actually drawn has been converted into a vector image.
[0067] One option for AI-based automatic creation (tracing) of floor plans is to define and create floor plan information in a bottom-up manner through trial and error, based on knowledge gained through the floor plan information extraction process.
[0068] For example, the endpoints of a wall are considered to be at the bottom of the vector image.
[0069] And the wall is represented as a line, a vector designed to connect the points on either end of the wall, above the endpoints of the wall.
[0070] When countless lines are drawn on a floor plan, the points on either side of the walls inevitably overlap. The areas enclosed by the lines represent rooms, or interior spaces.
[0071] Walls and interior spaces are simply marked on the intermediate artifact skeleton (S) for subsequent inference in floor plan creation (see Figure 4-2).
[0072] However, this type of bottom-up automated floor plan creation requires extremely high extraction accuracy, as the accuracy of points and walls is the foundation (foundation) of floor plan creation.
[0073] We believe that it would be difficult to generalize and accurately convert all input floor plans because there are countless different designs and specifications depending on the artist and the possibility of mistakes by the creator cannot be ruled out. In other words, we recognize that incorrect inference (conversion) may occur.
[0074] As a result, small, individual incorrect inferences are the primary cause of errors in automated floor plan generation (e.g., misreading a point can result in a room being created without walls).
[0075] Therefore, the inventor and applicant decided to take a different approach to recognizing the skeleton structure (Figure 4-2): a top-down approach. (To make the top-down approach easier to understand, we deliberately explained the bottom-up approach in the previous section.) The top-down approach is an approach that recognizes the space first and then recognizes specific walls and pillars, so as to prevent the superstructure from collapsing due to a series of incorrect inferences. Specifically, we start by understanding the structure of the room.
[0076] When you draw a room, the lines that surround it are recognized as a polygon. By breaking down that polygon, you can imagine the walls, which can be thought of as boundary lines. Then, by understanding the ends of the boundary lines, you will be able to recognize both ends of the walls.
[0077] Specifically, several solutions that are likely to be correct (drawing reading predictions, skeletons) are scored using a preset formula, and the solution with the highest score is selected. The formula uses a cost function to heuristically find the optimal solution.
[0078] For example, when referring to the output floor plan (S) extracted by the skeleton extraction device 310, if the surface of the output floor plan (S) largely matches the indoor area predicted by the logical reasoning device 410, it satisfies the condition of being an indoor area, so a reward is obtained and the score is increased.
[0079] Conversely, if the walled space cannot enclose a room (i.e., the output floor plan (S) does not largely match the predicted indoor area), it is reasonably determined to be surrounded by a wall (the boundary line of the dwelling unit) and is not considered an indoor area, and a penalty will be incurred.
[0080] Such formulas are heuristically designed to approximate the output floor plan (S) as closely as possible.
[0081] If the prediction for the approximate drawing does not have a high enough score (likelihood), then alternatives that contradict the prediction will be considered.
[0082] For example, some indoor areas that were originally predicted to be indoors will now be considered outdoors, and a collection of walls will surround the room.
[0083] In this example, the prediction score for the approximate drawing is lower (it contradicts the prediction, so no reward is given, but a penalty is incurred), but the result is more logically complete (the room structure can be grasped).
[0084] The above flow only infers the relationships between rooms (interior spaces), walls, and wall edges from the floor plan. Detailed information is then added to complete the inference from the floor plan symbol recognizer 320 and optical character recognition (OCR) device 330.
[0085] Each object is placed in its proper location on the skeleton floor plan via coordinates. (For example, where on which wall are doors and windows located?)
[0086] The inherent information of these objects (coordinates, boundaries, size) is maintained while being incorporated into the vector image hierarchy.
[0087] Text information is arranged in a similar way, helping to classify different areas of the room (living room, bathroom, toilet, etc.).
[0088] In particular, while some specific symbols are specific to certain types of rooms (e.g., gas stoves are usually found in kitchens rather than bathrooms), the room type is often apparent from the text information.
[0089] Decorator 420 will autonomously complete the floor plan style conversion process.
[0090] The logical reasoner 410 provides structured details of the input floor plan that are particularly amenable to being converted into a vector image.
[0091] Decorator 420 redecorates floor plans by applying specific rules (specifications) to certain symbols and text to complete the floor plan transformation.
[0092] The specific rules (specifications) for designing and creating floor plans are predefined by us or our clients, such as real estate agencies.
[0093] Such decorative work can range from painting walls with a specific thickness and color to painting fixtures with a new design.
[0094] The result of this step is the final product of an automated floor plan creation (tracing) system using artificial intelligence (AI). The output floor plan retains the original information of the input floor plan while being modified (decorated) to the desired design.
[0095] The output floor plan (O) is transferred to the record storage device 500, which then updates the record (R) and notifies the system that the conversion process has been completed.
[0096] The record storage device 500 contains two databases with different purposes, as can be seen in Figures 6 and 7. The content management database 510 stores records of floor plans uploaded to the system.
[0097] The Record (R) (see Figure 2) stores transformation-related information (e.g., input floor plan (F), output floor plan (O), intermediate deliverables (I), etc.) and management-related information (e.g., date, time, status, project, client).
[0098] This database allows clients and verifiers to see work in progress and make revisions to previously created floor plans.
[0099] Meanwhile, the training database 520 is a database that provides training for the models 310, 320, and 330, including the input floor plan (F) and the validated intermediate products (I).
[0100] In fact, the training database 520 queries data from the content management database 510, but the use of the training database 520 is asynchronous to 510 since the machine learning device 700 is an independent module of the style transfer process.
[0101] With reference to FIG. 7, the verification device 600 is not synchronized with the style conversion device 400 described above, but is associated with an asynchronous verification task.
[0102] This design allows for high-volume style conversion tasks to be performed without human quality verification being a hindrance to the floor plan style converter 400.
[0103] 6 and 7, the verification device 600 comprises a verification interface 610 and a correction interface 620, both operated by a human verifier 601. The verification interface 610 is connected to the record storage device 500.
[0104] After updating the record (R), the style conversion device 400 notifies the verifier 601 that the output floor plan (0) has been successfully converted and is ready for quality verification.
[0105] Through the verification interface 610, the verifier 601 can determine the quality of the conversion result, whether the output floor plan is correct enough to be delivered to the client.
[0106] If the quality is not sufficient, the verifier 601 must modify the output floor plan (O) using the modification interface 620 and return the modified floor plan (C) to the verification interface 610.
[0107] During the two-way communication between the verification interface 610 and the correction interface 620, the intermediate product (I) is verified and corrected by the machine learning device 700, which automatically learns efficiently to improve the accuracy of floor plan generation in the system. Therefore, when making corrections, it is necessary to make them synchronously on both the output floor plan (O) and the intermediate product (I).
[0108] Once the modifications are complete and the quality is approved, the verified output floor plan (V) (either (O) or (C)) is sent to the record storage device 500 to update the record (R). After the transmission is complete, the client 201 is notified that the record has been stored.
[0109] The machine learning device 700 is responsible for enhancing the accuracy of three devices: the floor plan skeleton recognizer 310, the symbol recognizer 320, and the optical character recognition (OCR) device 330.
[0110] It is well known that deep learning models require a large amount of training data set to achieve sufficient accuracy. Before this system was put into operation, training database 520 was pre-populated with training sample data from floor plan creation work that had been done manually using floor plan creation software.
[0111] Once the system is operational, we will use the verified data that has passed through the system to improve the accuracy of the floor plan creation process.
[0112] As mentioned above, each device (model) focuses on different elements of the floor plan, so the training data required to improve the accuracy of floor plan creation for each device (model) will inevitably be different.
[0113] Each device (model) is trained using components corresponding to intermediate artifacts (I) stored in the database 520. The skeleton recognizer 310 uses a raster image of the floor plan skeleton, the symbol recognizer 320 uses a list of facility information for the floor plan, and the optical character recognition (OCR) device uses a list of characters, among others. [Explanation of symbols]
[0114] 100...Automatic floor plan creation (tracing) system using artificial intelligence (AI) 200...Content management device 201 ...Client 210...User Interface 300...information extraction device 310 ...Skeleton Recognition 320...Symbol Recognition Function 330 ...Optical Character Recognition (OCR) function 400...Style conversion device 410 ...Logical reasoning device 420...Decorator equipment 500...Record-keeping device 510 ... Content Management Database 520 ...Learning database 600 ... Verification device 601 ...Verifier 610 ... Validation Interface 620...Modification Interface 700...Machine learning device F ... Input floor plan (floor plan uploaded to the system) R...Record I...Intermediate deliverables (floor plan) S...Floor plan skeleton (structure) O ...Output floor plan (deliverable) C...Revised floor plan V... Verified floor plans
Claims
1. Floor plan creation system, a content management device for inputting floor plans; an information extraction device for extracting information required for creating an output floor plan from the input floor plan input to the content management device, and grouping the information to form an intermediate product; a style conversion device that converts the intermediate result created by the information extraction device into a vector image to obtain the output floor plan; a record storage device for storing the input floor plan, the output floor plan, information required for creating the output floor plan, past verification results, and corrections to the output floor plan; a verification device for verifying whether the output floor plan has been converted according to a pre-specified floor plan creation specification, determining the quality, and correcting the output floor plan if the quality is not sufficient; a machine learning device that learns from the verification and the correction by the verification device to improve accuracy of the output floor plan; the information extraction device includes three deep learning devices, namely, a floor plan skeleton recognition device, a symbol recognition device, and an optical character recognition device, and the three deep learning devices extract skeleton information, symbol information representing facility information, and text information as information necessary for creating the output floor plan, and group the information to form the intermediate product, which is then transferred to the style conversion device; the style conversion device includes a logical inference device that extracts basic elements of a structured vector image, which is a collection of points and lines connecting points, from the intermediate product created by the information extraction device and facility information and character information linked to the intermediate product, and a decorator device that re-decorates the floor plan by applying specifications unique to specific symbols and text to complete the conversion of the input floor plan; an output of the machine learning device is input to the floor plan skeleton recognizer, the symbol recognizer, and the optical character recognizer of the information extraction device; A floor plan creation system in which the machine learning device trains the three deep learning devices so that they can extract information with greater accuracy, thereby improving the accuracy of the three deep learning devices.
2. The floor plan creation system of claim 1, wherein the extraction operation by the logical inference device is performed with a top-down approach in which the space is recognized first, and then specific walls and pillars are recognized, so as to prevent the superstructure from collapsing due to an accumulation of erroneous inferences.
3. 2. The floor plan creation system according to claim 1, wherein the content management device comprises an interface for inputting floor plans from a user.
4. The floor plan creation system according to claim 2, wherein the logical inference device heuristically finds an optimal solution using a cost function when extracting basic elements of a structured vector image, which is a collection of points and lines connecting points.
5. How to create a floor plan: inputting a floor plan by a content management device; extracting information necessary for creating an output floor plan from the input floor plan input to the content management device by an information extraction device, and grouping the information to form an intermediate product; a step of converting the intermediate product created by the information extraction device into a vector image by a style conversion device to obtain the output floor plan; storing, by a record storage device, the input floor plan, the output floor plan, information required for creating the output floor plan, past verification results, and corrections to the output floor plan; a step of verifying by a verification device whether the output floor plan has been converted according to a pre-specified floor plan creation specification, determining the quality, and correcting the output floor plan if the quality is not sufficient; and a step of learning the verification and the correction by the verification device using a machine learning device to improve accuracy of the output floor plan, the information extraction device includes three deep learning devices, namely, a floor plan skeleton recognition device, a symbol recognition device, and an optical character recognition device, and the step of forming the intermediate product by the information extraction device includes extracting skeleton information, symbol information representing facility information, and text information as information necessary for creating the output floor plan by the three deep learning devices, grouping them to form the intermediate product, and transferring them to the style conversion device; the step of generating the output floor plan by the style conversion device includes a step of extracting basic elements of a structured vector image, which is a collection of points and lines connecting the points, from the intermediate product created by the information extraction device and facility information and character information associated with the intermediate product; and a step of redecorating the floor plan by applying specifications unique to specific symbols and text to complete the conversion of the input floor plan; an output of the machine learning device is input to the floor plan skeleton recognizer, the symbol recognizer, and the optical character recognizer of the information extraction device; A floor plan creation method in which the step of improving the accuracy of the output floor plan by the machine learning device involves training the three deep learning devices so that they can extract information with greater accuracy, thereby enhancing the accuracy of the three deep learning devices.
Citation Information
Patent Citations
Automatic extraction method and system for house type image element based on machine vision
CN108763606A
Semantic segmentation of 2d floor plans with a pixel-wise classifier
EP3506160A1
Building drawing recognizing method
JP1998124547A
Information processing device, information processing method, and program
JP2018081548A
Drawing machine learning support system, drawing machine learning system, drawing structuring system, drawing machine learning support method, and drawing machine learning support program
JP2019101514A
