Furniture identification and matching method and related device

Abstract

The invention discloses a furniture identification and matching method and a related device. Firstly, a house plane graph, furniture demand description and user preferences can be obtained, and all furniture space parameters in the house plane graph are extracted to obtain a furniture space parameter set. And performing furniture retrieval in the multi-modal furniture knowledge base based on the furniture space parameter set and the furniture demand feature description in the furniture demand description to obtain p matched furniture diagrams, and selecting q matched furniture diagrams from the p matched furniture diagrams based on user preferences, and displaying the selected furniture diagrams to the user for selection to obtain a user selection result. And determining a target furniture drawing in q (q is smaller than p) matched furniture drawings based on a user selection result, and performing three-dimensional model conversion on the target furniture drawing to obtain a three-dimensional model of the matched furniture drawings. According to the method, the furniture is accurately matched in the multi-modal furniture knowledge base by combining the house planar graph, the furniture demand description and the user preference, and an interactive selection mechanism is provided, so that the design individuation and the user satisfaction are ensured.

Description

Technical Field

[0001] This application relates to the technical field of computer vision and image processing, and particularly to a furniture recognition and matching method and related devices. Background Art

[0002] With the development of information technology, computer-aided home design tools have gradually entered people's lives. The earliest design software such as AutoCAD and 3ds Max can provide accurate floor plans and 3D models for professional designers, but due to complex operations, it is difficult for ordinary users to master. To meet the needs of the public, more user-friendly home design software has been introduced on the market, such as Sweet Home 3D and Planner 5D. These tools use graphical interfaces and provide preset furniture libraries, allowing users to complete interior layout design and preview the effects by simply dragging and dropping. However, the current technology still faces significant challenges in accurately matching user needs and is difficult to accurately respond to users' personalized requirements.

[0003] Therefore, how to improve the personalized and accurate matching ability of home design to better meet users' diverse and unique design needs is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0004] Based on the above problems, this application provides a furniture recognition and matching method and related devices, which can improve the personalized and accurate matching ability of home design to better meet users' diverse and unique design needs.

[0005] The embodiments of this application disclose the following technical solutions:

[0006] A furniture recognition and matching method, the method includes:

[0007] Obtain a floor plan of a house, a furniture requirement description, and user preferences; both the floor plan of the house and the furniture requirement description are provided by the user; the furniture requirement description includes a description of furniture requirement features;

[0008] Extract all furniture space parameters in the floor plan of the house to obtain a furniture space parameter set;

[0009] Based on the furniture space parameter set and the description of furniture requirement features, perform furniture retrieval in a multimodal furniture knowledge base to obtain p matching furniture diagrams, and select q from the p matching furniture diagrams based on the user preferences to display to the user for selection to obtain a user selection result; the multimodal furniture knowledge base includes multiple furniture diagrams, and each furniture diagram is marked with a corresponding furniture introduction and furniture size parameters; both p and q are positive integers, and q is less than p;

[0010] Determine a target furniture image among the q matching furniture images based on the user selection result;

[0011] The target furniture image is converted into a three-dimensional model to obtain a three-dimensional model of the matching furniture image.

[0012] In a possible implementation, performing furniture retrieval in a multimodal furniture knowledge base based on the furniture space parameter set and the furniture demand feature description to obtain p matching furniture graphs includes:

[0013] Vectorizing the furniture demand feature description to obtain a demand feature vector, and vectorizing each furniture image and furniture description of the furniture image in the multimodal furniture knowledge base to obtain multiple furniture image feature vectors and furniture description feature vectors;

[0014] Calculating the similarity between the plurality of furniture profile feature vectors and the requirement feature vector, and selecting the furniture images corresponding to the x furniture profile feature vectors with the top x similarities as the first candidate furniture images; x is a positive integer;

[0015] Calculate the similarity between the furniture image feature vectors of the x first candidate furniture images and the requirement feature vector, and select the furniture corresponding to the y furniture image feature vectors with the top y similarities as the second candidate furniture image; y is a positive integer, and y is less than x;

[0016] The y second candidate furniture images are screened based on the furniture space parameters in the furniture space parameter set to obtain z third candidate furniture images; the furniture size parameters of the z third candidate furniture images are all less than or equal to the furniture space parameters in the furniture space parameter set; z is a positive integer, and z is less than y;

[0017] p fourth candidate furniture images are selected from the z third candidate furniture images as the matching furniture images; the p fourth candidate furniture images are the p third candidate furniture images whose differences between furniture size parameters and the furniture space parameters are ranked last; and p is less than z.

[0018] In a possible implementation, extracting all furniture space parameters from the house plan to obtain a furniture space parameter set includes:

[0019] Converting the image format of the house plan into a target format to obtain a target plan;

[0020] Preprocessing the target plane map to obtain a preprocessed plane map;

[0021] Recognizing architectural elements and furniture symbols in the pre-processed floor plan using a deep neural network based on the YOLO architecture;

[0022] Convert the building elements and the furniture symbols into corresponding furniture space parameters one by one according to the principles of computational geometry;

[0023] Integrate all the furniture space parameters to obtain the set of the furniture space parameters.

[0024] In a possible implementation, when the target floor plan is a raster image, the preprocessing of the target floor plan to obtain a preprocessed floor plan includes:

[0025] Use the Gaussian filtering algorithm to perform noise reduction processing on the target floor plan to obtain a noise-reduced floor plan;

[0026] Apply the histogram equalization algorithm to enhance the contrast of the noise-reduced floor plan to obtain an enhanced floor plan;

[0027] Use the edge detection algorithm to extract the key edge features of the enhanced floor plan to obtain an edge feature map;

[0028] Perform resolution normalization on the edge feature map to obtain a first normalized image;

[0029] Wherein, the first normalized image is the preprocessed floor plan.

[0030] In a possible implementation, when the target floor plan is a vector image, the preprocessing of the target floor plan to obtain a preprocessed floor plan includes:

[0031] Perform vector parsing on the target floor plan to obtain vector data;

[0032] Perform raster image conversion on the vector data to obtain a rasterized image;

[0033] Perform resolution normalization on the rasterized image to obtain a second normalized image;

[0034] Wherein, the second normalized image is the preprocessed floor plan.

[0035] In a possible implementation, the conversion of the target furniture image into a three-dimensional model of the matching furniture image includes:

[0036] Use the Remove Background Model (RMBG model) to perform foreground and background separation processing on the matching furniture image to obtain a foreground furniture image;

[0037] Use the Trellis three-dimensional reconstruction model to perform three-dimensional reconstruction on the foreground furniture image to obtain a three-dimensional furniture model;

[0038] Based on the YOLO12 object detection algorithm, perform size ratio calibration on the three-dimensional furniture model to obtain the three-dimensional model of the matching furniture image.

[0039] In a possible implementation, the method further includes:

[0040] Performing three-dimensional scene rendering on the three-dimensional model of the matched furniture diagram, so that the user can rotate or scale the perspective of the three-dimensional model in the three-dimensional scene through mouse or touch operations.

[0041] In a possible implementation, the furniture demand description further includes a furniture demand size description;

[0042] The step of selecting q from the p matched furniture diagrams based on the user preference and presenting them to the user for selection to obtain a user selection result includes:

[0043] Screening the p matched furniture diagrams based on the furniture demand size description to obtain n final matched furniture diagrams; the furniture size parameters of the final matched furniture diagrams are all less than or equal to the furniture demand size description;

[0044] Selecting q from the n final matched furniture diagrams based on the user preference and presenting them to the user for selection to obtain the user selection result; n is a positive integer, and n is less than p and greater than q.

[0045] A furniture recognition and matching device, the device includes:

[0046] An acquisition unit, configured to acquire a floor plan of a house, a furniture demand description, and a user preference; both the floor plan of the house and the furniture demand description are provided by the user; the furniture demand description includes a furniture demand feature description;

[0047] A parameter extraction unit, configured to extract all furniture space parameters in the floor plan of the house to obtain a furniture space parameter set;

[0048] A furniture retrieval unit, configured to perform furniture retrieval in a multi-modal furniture knowledge base based on the furniture space parameter set and the furniture demand feature description to obtain p matched furniture diagrams;

[0049] A furniture diagram display unit, configured to select q from the p matched furniture diagrams based on the user preference and present them to the user for selection to obtain a user selection result; the multi-modal furniture knowledge base includes multiple furniture diagrams, and each furniture diagram is marked with a corresponding furniture introduction and furniture size parameters; both p and q are positive integers, and q is less than p;

[0050] A determination unit, configured to determine a target furniture diagram among the q matched furniture diagrams based on the user selection result;

[0051] A three-dimensional model conversion unit for converting the target furniture drawing into a three-dimensional model of the matching furniture drawing.

[0052] A furniture recognition and matching device includes: a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the furniture recognition and matching method as described above is implemented.

[0053] Compared with the prior art, the present application has the following beneficial effects:

[0054] The present application provides a furniture recognition and matching method and related device. Specifically, when implementing the furniture recognition and matching method provided in the embodiments of the present application, first, obtain the house floor plan, furniture demand description, and user preferences from the user. These input information provide a basis for subsequent precise matching. Then, extract all furniture space parameters from the house floor plan to form a furniture space parameter set. This step helps to determine the available space range and size limitations. Next, based on the extracted furniture space parameter set and the furniture demand feature description in the user's furniture demand description, perform furniture retrieval in the multi-modal furniture knowledge base to obtain p matching furniture drawings. Each furniture drawing is marked with the corresponding furniture introduction and size parameters. Subsequently, select q from these p matching furniture drawings according to the user preferences and display them to the user for selection, and finally obtain the user's selection result (where p and q are both positive integers, and q is less than p). Then, determine the target furniture drawing among the q matching furniture drawings according to the user's final selection result. Finally, perform three-dimensional model conversion on the selected target furniture drawing to generate its three-dimensional model. The present application can more precisely match the personalized needs of users by performing furniture retrieval in the multi-modal furniture knowledge base through the house floor plan and furniture demand description provided by the user. At the same time, the present application provides an interactive selection mechanism based on user preferences, selects q from p matching furniture drawings and displays them to the user, allowing the user to participate in the final selection, further ensuring user satisfaction and design personalization. Description of the Drawings

[0055] To more clearly illustrate the technical solutions in the embodiments or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0056] Figure 1 It is a flowchart of a furniture recognition and matching method provided by an embodiment of the present application;

[0057] Figure 2The flowchart of a method for retrieving a matching furniture diagram provided by an embodiment of the present application;

[0058] Figure 3 The flowchart of a method for constructing a furniture space parameter set provided by an embodiment of the present application;

[0059] Figure 4 The flowchart of a method for preprocessing a floor plan provided by an embodiment of the present application;

[0060] Figure 5 The flowchart of another method for preprocessing a floor plan provided by an embodiment of the present application;

[0061] Figure 6 The flowchart of a method for converting a three-dimensional model of a furniture diagram provided by an embodiment of the present application;

[0062] Figure 7 The structural schematic diagram of a furniture recognition and matching device provided by an embodiment of the present application. Detailed implementation manners

[0063] To facilitate the understanding of the technical solutions provided by the embodiments of the present application, the background technologies related to the embodiments of the present application will be described first below.

[0064] Home decoration is a life event that most people have experienced or will experience. Traditional processes include steps such as measuring the space size, selecting furniture, and planning the layout. In this process, people need to rely on imagination to predict the final effect, which is a challenge for ordinary consumers. With the development of information technology, computer-aided design tools such as AutoCAD and 3ds Max have been introduced. Although these tools can create accurate floor plans and three-dimensional models, the learning curve is steep and it is difficult for ordinary users to master.

[0065] To meet the needs of ordinary users, more user-friendly home design software has emerged on the market, such as SweetHome 3D and Planner 5D. These software adopt graphical interfaces and provide a preset furniture model library. Users can arrange furniture through simple drag-and-drop operations and preview the effects through 3D views. However, the current technology has not effectively combined the floor plan information and the user's furniture description requirements, resulting in the inability to accurately match the user's needs and thus unable to accurately respond to the user's personalized needs.

[0066] To solve this problem, an embodiment of the present application provides a furniture recognition and matching method and related device. First, relevant data is obtained through the floor plan of the house, furniture demand description, and user preferences provided by the user. Then, all furniture space parameters are extracted from the floor plan of the house to form a furniture space parameter set. Next, using this parameter set and the furniture demand feature description in the furniture demand description, furniture retrieval is performed in the multimodal furniture knowledge base to obtain multiple matching furniture diagrams (p pieces). According to the user's preferences, q pieces are selected from these p matching furniture diagrams and presented to the user for selection. Finally, according to the user's selection result, one target furniture diagram is determined, and this furniture diagram is converted into a 3D model. The present application uses the floor plan of the house and furniture demands provided by the user to perform furniture search through the multimodal furniture knowledge base to more accurately meet the personalized needs of the user. At the same time, this application also introduces an interactive selection system based on user preferences, screening out the selected furniture presented to the user from multiple matching furniture diagrams, allowing the user to participate in the selection process, which helps to improve user satisfaction and achieve personalized design.

[0067] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of the present application.

[0068] See Figure 1 , which is a flowchart of a furniture recognition and matching method provided by an embodiment of the present application. As Figure 1 shown, the furniture recognition and matching method may include steps S101 - S105:

[0069] S101: Obtain the floor plan of the house, furniture demand description, and user preferences.

[0070] At the beginning of this process, the user needs to provide some key information to help the system understand their needs. Specifically, the user can upload the floor plan file of their own house or use the tools provided by the system to draw the floor plan of the house. The floor plan includes the dimensions and layouts of each room in the house so that the system can understand the available space. In addition, the user also needs to provide a furniture demand description, explaining what kind of furniture they need (such as style, function, etc.). At the same time, the user's personal preferences also need to be considered, such as preferences for color, material, brand, etc. This information will help the system recommend more suitable furniture for the user and ensure that the recommended furniture can adapt to the actual space layout of the house.

[0071] Among them, the furniture demand description must include the description of furniture demand characteristics, but may also include the description of furniture demand characteristic dimensions. Specifically, the description of furniture demand characteristics involves the user's requirements for the appearance, style, material, color, etc. of the furniture, reflecting the user's overall preference and aesthetics for the furniture. Through these characteristics, the general image and style of the furniture can be clearly outlined. The description of furniture demand characteristic dimensions focuses more on the specific dimension requirements of the furniture to help ensure that the furniture can adapt to the actual use space. The dimension description not only includes the overall size of the furniture, but may also involve the specific dimensions of each part, such as the height and width of the tabletop and the seat, which are crucial for the user's actual use experience. Through these detailed demand descriptions, the specific requirements of the user can be accurately conveyed, providing a clear basis for furniture matching.

[0072] For example, the user may describe "I need a modern-style wooden dining table with a 30*30 tabletop and a dark brown color". Among them, "modern style" is the demand characteristic of the style, "wooden" is the demand characteristic of the material, "dark brown" is the demand characteristic of the color, and "dining table" is the demand characteristic of the furniture type, while "30*30 tabletop" is the demand characteristic of the dimension, specifically indicating the size of the tabletop.

[0073] It should be noted that the floor plan of the house is a two-dimensional drawing of the house layout, showing the actual dimensions of the building elements such as the walls, doors, windows, and columns of the room, as well as the actual dimensions of the furniture such as sofas, beds, tables, and chairs. It helps the system understand the actual space structure of the house so as to ensure that the furniture can adapt to the actual space layout when selecting furniture. The floor plan of the house provides the system with the limiting conditions for furniture placement. Through this information, the system can avoid recommending furniture that does not match the house size.

[0074] User preference refers to the specific preferences and requirements of the user when choosing furniture. These preferences may involve aspects such as color, material, brand, and price range. User preferences help the system further refine furniture recommendations. For example, if the user prefers solid wood furniture rather than metal furniture, the system will give priority to recommending solid wood furniture and provide choices that meet the user's budget and style within the optional range.

[0075] S102: Extract all furniture space parameters in the floor plan of the house to obtain a set of furniture space parameters.

[0076] After obtaining the floor plan of the house, all spatial information related to furniture placement is retrieved from the floor plan of the house and organized into a set. These spatial parameters may include the area, shape, door and window positions, wall lengths, etc. of each room. Through these parameters, the system can identify the areas in the house suitable for placing furniture and determine which areas have the dimensions and layouts most suitable for specific furniture placement requirements. Finally, all these spatial parameters are collected and organized into a set. This process ensures that the system can accurately understand the actual dimensions and layout of the room, providing an accurate data basis for subsequent furniture matching.

[0077] S103: Based on the set of furniture spatial parameters and the furniture requirement description, perform furniture retrieval in the multimodal furniture knowledge base to obtain p matching furniture diagrams, and select q from the p matching furniture diagrams based on the user preferences to display to the user for selection to obtain the user selection result.

[0078] By comprehensively analyzing the relevant parameters of the house space (i.e., the set of furniture spatial parameters) and the user's description of furniture requirements, the system will retrieve in a knowledge base containing various furniture information to find several furniture images and related information that meet the requirements. Specifically, the system searches for p furniture images that meet the conditions according to these spatial parameters and requirement descriptions. These images come from the same "multimodal furniture knowledge base", which contains multiple furniture diagrams, and each image is accompanied by a detailed furniture introduction and dimension parameters to help the system understand and screen suitable furniture.

[0079] Next, based on the user's personal preferences, the system selects q of the most matching furniture from these p eligible furniture for display. The q pieces of furniture displayed to the user will be available for the user to choose, and q being less than p means that not all eligible furniture will be displayed, but rather through further screening, only a few pieces of furniture that best meet the user's needs will be shown. Finally, the user makes their selection result by choosing one of these displayed furniture. Both p and q are positive integers.

[0080] Exemplarily, assume that the user uploads a floor plan of his bedroom, and the system analyzes the spatial parameters of the bedroom (e.g., the bedroom area is 15 square meters, the length of the room is 4 meters, and the width is 3.5 meters). The user describes the type of bed they wish to purchase: "I need a modern-style bed for two people, with a padded headboard, and the length of the bed is about 2 meters." (i.e., the furniture requirement description). The system retrieves in the multimodal furniture knowledge base based on the spatial parameters (bedroom size and shape) and furniture requirements (modern style, suitable for two people, padded headboard, etc.) provided by the user. The multimodal furniture knowledge base contains a large number of furniture images, and each image has a corresponding furniture profile and size parameters. The system retrieves 8 matching bed images according to these conditions (i.e., p = 8). The system further filters these 8 beds according to the user's preferences (such as padded headboard, modern style, etc.) and selects 4 beds that best meet the requirements (i.e., q = 4). These beds are filtered from the 8 matching bed images and meet all the user's requirements. The system shows these 4 beds to the user, and the user can choose the bed they like according to the images and descriptions. Finally, the user selects a modern-style bed with a padded headboard, suitable for two people, and the size is 2 meters.

[0081] S104: Determine the target furniture image among the q matching furniture images based on the user selection result.

[0082] According to the user's selection result among the q displayed matching furniture images, the system will determine the final target furniture image, that is, the piece of furniture selected by the user. Assume that the system displays 4 furniture images that meet the user's requirements (i.e., q = 4), and each furniture image represents the design and specifications of a piece of furniture. When the user makes a choice from these 4 options (e.g., the user selects one of the bed images), the system will record this selection result and determine the selected furniture image as the final target furniture image, that is, the furniture that the user ultimately wants to purchase or view.

[0083] "Determine the target furniture image among the q matching furniture images based on the user selection result" means determining, through the user's selection, the piece of furniture image that the user is interested in and finally selects.

[0084] S105: Perform a 3D model conversion on the target furniture image to obtain a 3D model of the matching furniture image.

[0085] After the user selects the target furniture image, the system will convert this furniture image into a 3D model, thereby obtaining a furniture model that can be displayed and operated in 3D space.

[0086] Specifically, the target furniture diagram is a two-dimensional image that shows the appearance and design of the furniture. To enable users to more intuitively perceive the three-dimensional effect of the furniture, the system will perform three-dimensional modeling based on the information in this two-dimensional image to create a three-dimensional model. This three-dimensional model can not only display the shape of the furniture but also show various details of the furniture, such as dimensions, materials, proportions, etc. Through the three-dimensional model, users can view the furniture from different angles and perspectives, enhancing the accuracy of the purchase decision.

[0087] Based on the content of S101 - S105, first, obtain the floor plan of the house, furniture demand description, and user preference information provided by the user. The furniture demand feature descriptions in the floor plan of the house and the furniture demand description are used as key input data directly provided by the user, ensuring the personalization and pertinence of the recommendation results. Next, by analyzing the floor plan of the house, extract all relevant furniture space parameters to form a set of furniture space parameters. These space parameters include information such as the layout of the rooms and the dimensions of the furniture placement spaces, providing accurate spatial basic data for subsequent furniture matching. Based on these extracted space parameters and the furniture demand feature descriptions provided by the user, the system will retrieve in a multi-modal furniture knowledge base containing rich and diverse furniture information. This knowledge base contains a large number of furniture images, and each image not only shows the appearance of the furniture but also comes with a corresponding furniture introduction and detailed dimension parameters. This multi-modal knowledge base supports multi-dimensional furniture matching and can screen out multiple eligible furniture images according to different space parameters and demand descriptions, returning p matching results. To further improve the accuracy of the recommendation and meet the personalized needs of users, the p matching furniture diagrams can be screened according to the user's preferences, and q furniture diagrams that best meet the user's needs are displayed. At this time, q is a positive integer less than p. After the user makes a choice from the q displayed furniture diagrams, according to the user's choice result, determine the target furniture diagram, that is, the finally selected furniture diagram. Next, convert the target furniture diagram into a three-dimensional model, and this conversion process enables users to view the selected furniture in a virtual environment and understand its effect and dimensions in the actual space. This application accurately retrieves furniture in a multi-modal furniture knowledge base by using the floor plan of the house and the furniture demand description provided by the user, thus better meeting the personalized needs of users. In addition, this application also provides an interactive selection mechanism based on user preferences, selecting q from the p matching furniture diagrams to display to the user and allowing the user to participate in the final selection, further ensuring user satisfaction and design personalization.

[0088] See Figure 2 , Figure 2This is a flowchart of a method for retrieving matching furniture diagrams provided by an embodiment of the present application. Correspondingly, in step S103, p matching furniture diagrams are retrieved from the multimodal furniture knowledge base based on the furniture space parameter set and the furniture demand feature description, which can be specifically implemented through steps S201 - S205:

[0089] S201: Vectorize the furniture demand feature description to obtain a demand feature vector, and vectorize each furniture diagram and its furniture introduction in the multimodal furniture knowledge base to obtain a plurality of furniture picture feature vectors and furniture introduction feature vectors.

[0090] To determine the matching furniture diagrams, first, the furniture demand feature description provided by the user (such as required functions, styles, materials, etc.) is transformed into a mathematical vector form to obtain a demand feature vector (for example, [0.8, 0.3, 1.0], where each value represents different feature weights or attributes). This vector can help the computer understand the specific content of the demand and can be compared with other data. Then, each furniture diagram in the multimodal furniture knowledge base and its associated furniture introduction (such as dimensions, materials, styles, etc.) are respectively transformed into a furniture picture feature vector (for example, [0.6, 0.5, 0.9], representing the visual features of the picture) and a furniture introduction feature vector (for example, [0.7, 0.4, 0.8], representing the text features of the introduction). These feature vectors represent the main information of the furniture diagrams and introductions, enabling the system to perform effective matching and screening through the similarity between the vectors. In other words, this process transforms complex text and picture information into a digital form that can be understood and processed by the machine, thus providing a basis for subsequent screening and recommendation.

[0091] S202: Calculate the similarity between the plurality of furniture introduction feature vectors and the demand feature vector, and screen the furniture diagrams corresponding to the top x furniture introduction feature vectors with the highest similarity as the first candidate furniture diagrams.

[0092] By calculating the similarity between the feature vector of each furniture introduction in the multimodal furniture knowledge base and the feature vector of the user's demand description, it is determined which furniture introductions best meet the user's needs. Specifically, the system calculates the similarity between each furniture introduction feature vector and the demand feature vector (for example, using methods such as cosine similarity), and then selects the top x furniture introductions that best match the user's needs according to the similarity. Finally, the furniture diagrams corresponding to these selected furniture introductions will be used as the first candidate furniture diagrams for further screening and processing. Here, x is a positive integer, indicating how many of the most matching candidate furniture diagrams are selected.

[0093] In a possible implementation manner, the similarity calculation formula can be but is not limited to:

[0094] Among them, q d is the demand feature vector, and t d is the furniture profile feature vector.

[0095] Exemplarily, assume that the demand feature vector of the user is q d = [0.8, 0.3, 1.0], and there are 5 furniture pictures and their furniture profile feature vectors in the furniture knowledge base as follows: Furniture Figure 1 : t d 1 = [0.7, 0.4, 0.9]; Furniture Figure 2 : t d 2 = [0.6, 0.5, 0.8]; Furniture Figure 3 : t d 3 = [0.9, 0.2, 1.0]; Furniture Figure 4 : t d 4 = [0.5, 0.6, 0.7]; Furniture Figure 5 : t d 5 = [0.8, 0.3, 0.9].

[0096] Calculate the similarity: sim(q d [[ID=*36]], t d 1) = 0.95; sim(q d , t d 2) = 0.90; sim(q d , t d 3) = 0.98; sim(q d , t d 4) = 0.88; sim(q d , t d 5) = 0.97.

[0097] Assume x = 3, then select the top 3 furniture pictures with the highest similarity: Furniture Figure 3 (similarity 0.98), Furniture Figure 5 (similarity 0.97), Furniture Figure 1 (similarity 0.95).

[0098] S203: Calculate the similarity between the furniture picture feature vectors of the x first candidate furniture pictures and the demand feature vector, and screen the furniture corresponding to the y furniture picture feature vectors with the top y similarities as the second candidate furniture pictures.

[0099] After screening out the first x candidate furniture pictures that best meet the user's needs, the system further screens these candidate furniture pictures. Specifically, it calculates the similarity between the furniture picture feature vectors of these x furniture pictures and the user's need feature vector to evaluate the matching degree of these furniture pictures with the user's needs. Then, according to the similarity level, the top y furniture pictures with the highest similarity are selected as the second candidate furniture pictures. Here, y is a positive integer and y is less than x, which means the system further refines from the initial x candidates and retains the y furniture pictures that best match the needs for further processing or recommendation.

[0100] In a possible implementation, the similarity calculation formula can be but is not limited to:

[0101] where q d is the need feature vector, and i d is the furniture picture feature vector.

[0102] Exemplarily, assume that there are 3 first candidate furniture pictures (x = 3) initially screened out, which are furniture Figure 1 、 3 and 5, and their furniture picture feature vectors are as follows: Furniture Figure 1 : i d 1 = [0.8, 0.3, 1.0]; Furniture Figure 3 : i d 3 = [0.6, 0.5, 0.8]; Furniture Figure 5 : i d 5 = [0.7, 0.4, 0.9]. The user's need feature vector is q d = [0.8, 0.3, 1.0].

[0103] Calculate the similarity: sim(q d , i d 1) = 0.99; sim(q d , i d 3) = 0.90; sim(q d , i d 5) = 0.95.

[0104] Assume y = 2, then select the top 2 furniture pictures with the highest similarity: Furniture Figure 1 (similarity 0.99) and Furniture Figure 5 (similarity 0.95) as the second candidate furniture pictures.

[0105] S204: Screen the y second candidate furniture pictures based on the furniture space parameters in the furniture space parameter set to obtain z third candidate furniture pictures.

[0106] After the system has selected the top y second-candidate furniture diagrams that best meet the user's requirements, it further filters them based on the spatial parameters of each piece of furniture. Specifically, the spatial parameters of furniture include information such as its dimensions, floor area, etc. By comparing the dimension parameters of each piece of furniture in the second-candidate furniture diagrams with a preset set of furniture spatial parameters (which could be the maximum dimension requirements of the user's space, etc.), the system filters out those furniture diagrams where the furniture dimension parameters are less than or equal to the requirements specified in the spatial parameter set. Finally, z furniture diagrams that meet the conditions are selected as the third-candidate furniture diagrams, and the dimension parameters of these furniture diagrams all meet the spatial restrictions of the user's requirements. Here, z is a positive integer, and z is less than y, meaning that the system further finely filters from the second candidates and retains those furniture diagrams with suitable spaces.

[0107] It should be noted that the furniture dimension parameters and furniture spatial parameters generally both include the following parameters:

[0108] Area: The planar dimension of the furniture, usually length × width, used to calculate the space occupied by the furniture;

[0109] Height: The vertical dimension of the furniture, usually referring to the height or depth of the furniture;

[0110] Depth: For some furniture (such as sofas, desks, wardrobes, etc.), the depth refers to the dimension of the furniture from front to back;

[0111] Length: The horizontal dimension of the furniture, usually referring to the length of the furniture from one end to the other;

[0112] Width: The width of the furniture, usually referring to the dimension of the furniture from one side to the other.

[0113] Exemplarily, in this example, only length, width, and height are taken as examples for illustration. Suppose the furniture spatial parameters provided by the user are: The actual length of the bed: 300 cm (centimeters); The actual width of the bed: 200 cm; The actual height of the bed: 250 cm. And there are 2 (y = 2) second-candidate furniture diagrams (i.e., candidate bed diagrams) initially selected, and their furniture dimension parameters are as follows: Selected bed type Figure 1 : Length 200 cm, width 150 cm, height 80 cm; Selected bed type Figure 5 : Length 350 cm, width 200 cm, height 100 cm. Filtering according to the furniture spatial parameters provided by the user: The dimensions of the candidate bed type Figure 1 (200 cm × 150 cm × 80 cm) are less than the actual dimensions of the bed (300 cm × 200 cm × 250 cm), meeting the conditions. The candidate bed type Figure 2The size (350 cm × 200 cm × 100 cm) exceeds the actual size of the bed (especially the length exceeds 300 cm) and does not meet the requirements. Therefore, there is only 1 third candidate furniture drawing finally determined (at this time z = 1): the candidate bed type Figure 1 (i.e., the furniture Figure 1 ).

[0114] S205: Screen p fourth candidate furniture drawings from the z third candidate furniture drawings as the matching furniture drawings.

[0115] Finally, from the z third candidate furniture drawings that have been screened out, the system further screens according to the difference between the size parameters of each furniture drawing and the preset furniture space parameters. Specifically, calculate the difference between the size of each furniture drawing and the space parameters, and then select the p furniture drawings with the smallest differences as the fourth candidate furniture drawings, that is, the size parameters of these furniture drawings are closest to the space requirements of the user. Here, p is a positive integer and p is less than z, which means that the system further finely screens from the third candidates and retains those furniture drawings with smaller size differences, and finally uses them as the recommended matching furniture drawings.

[0116] Exemplarily, in this example, only the length, width and height are taken as examples for illustration. Assume that after the previous screening, there are 3 third candidate furniture drawings (z = 3), and their furniture size parameters are as follows: the candidate bed type Figure 1 (i.e., the furniture Figure 1 ): length 20 cm, width 150 cm, height 80 cm; the candidate bed type Figure 2 (i.e., the furniture Figure 2 ): length 250 cm, width 180 cm, height 100 cm; the candidate bed type Figure 3 (i.e., the furniture Figure 3 ): length 220 cm, width 160 cm, height 90 cm. The furniture space parameters provided by the user are: the actual length of the bed: 300 cm, the actual width of the bed: 200 cm, the actual height of the bed: 250 cm.

[0117] Calculate the difference between the size of each furniture drawing and the furniture space parameters: the candidate bed type Figure 1 : length difference = 300 - 200 = 100, width difference = 200 - 15 = 50, height difference = 250 - 80 = 170, total difference = 100 + 50 + 170 = 320;

[0118] The candidate bed type Figure 2 : length difference = 300 - 250 = 50, width difference = 200 - 180 = 20, height difference = 250 - 100 = 150, total difference = 50 + 20 + 150 = 220;

[0119] The candidate bed type Figure 3: Length difference = 300 - 220 = 80, width difference = 200 - 160 = 40, height difference = 250 - 90 = 160, total difference = 80 + 40 + 160 = 280.

[0120] Assume that 2 fourth candidate furniture diagrams are to be selected (p = 2), then select the first 2 candidate bed types with the smallest difference: candidate bed type Figure 2 (total difference 220) and candidate bed type Figure 3 (total difference 280).

[0121] Therefore, there are 2 fourth candidate furniture diagrams finally determined (p = 2): candidate bed type Figure 2 and candidate bed type Figure 3 .

[0122] In this way, the system ensures that the recommended furniture not only meets the space requirements, but also is the furniture option most suitable for the user's space conditions.

[0123] See Figure 3 , Figure 3 which is the method flowchart of a method for constructing a furniture space parameter set provided by an embodiment of the present application. Correspondingly, step S102 extracts all furniture space parameters in the house floor plan to obtain a furniture space parameter set, which can be specifically implemented through S301 - S305:

[0124] S301: Convert the image format of the house floor plan into a target format to obtain a target floor plan.

[0125] When constructing a furniture space parameter set, it is first necessary to convert the original house floor plan from one image or file format into a standard format suitable for subsequent processing. Since the house floor plan may be a raster image or a vector image, these standard formats include common image formats such as Joint Photographic Experts Group (JPEG), Pixel Graphic (PxG) format, Tagged Image File Format (TIFF), etc., which are usually used to store picture data, or vector formats such as Drawing (DWG) format, Drawing Exchange Format (DXF), etc., which are usually used to store architectural design drawings and engineering data. Through this conversion, the floor plan can be adapted to different software and processing tools to ensure efficient subsequent image processing, feature recognition, and analysis. Finally, the converted "target floor plan" will be used as the basis for processing and enter the subsequent preprocessing and automatic recognition steps.

[0126] S302: Preprocess the target floor plan to obtain a preprocessed floor plan.

[0127] "Preprocess the target floor plan to obtain a preprocessed floor plan" means that after converting the house floor plan into the target format, a series of processes are carried out to make it more suitable for subsequent automated analysis and recognition. These preprocessing steps may include operations such as removing noise in the image, enhancing the contrast of the image, adjusting the sharpness of the image, and performing edge detection. The purpose of doing this is to improve the quality of the image, making features such as building elements and furniture symbols in it clearer and easier to identify, laying a foundation for subsequent object recognition, spatial parameter extraction, etc. Finally, the preprocessed floor plan will be used as the input data for subsequent analysis and is called the "preprocessed floor plan".

[0128] S303: Use a deep neural network based on the YOLO architecture to identify building elements and furniture symbols in the preprocessed floor plan.

[0129] "Use a deep neural network based on the YOLO architecture to identify building elements and furniture symbols in the preprocessed floor plan" means using a deep learning model based on the YOLO (You Only Look Once) architecture to analyze and identify the preprocessed floor plan. YOLO is an efficient real-time object detection algorithm that can quickly identify specific objects in an image. In this context, the YOLO deep neural network is trained to identify building elements (such as walls, doors, windows, stairs, etc.) and furniture symbols (such as sofas, tables, chairs, etc.) in the floor plan. Through this process, the model can locate and mark these important elements and symbols in the floor plan, providing support for subsequent automated analysis, spatial planning, and design optimization.

[0130] In a possible implementation, the deep neural network based on the YOLO architecture will perform transfer learning and fine-tuning on a dedicated floor plan dataset using a neural network pre-trained with a general object detection dataset. The input to the network is the preprocessed floor plan image, and through forward propagation, prediction results are generated, including object bounding boxes, class labels, and confidence scores; subsequently, the non-maximum suppression algorithm (NMS) is applied to eliminate redundant results.

[0131] S304: Convert the building elements and the furniture symbols into corresponding furniture space parameters one by one according to the principles of computational geometry.

[0132] "Converting the building elements and the furniture symbols into corresponding furniture space parameters according to the principles of computational geometry" means that through the application of relevant theories and methods of computational geometry, geometric analysis is carried out on the identified building elements and furniture symbols, so as to extract the specific spatial information of each piece of furniture or building element. The principles of computational geometry mainly involve how to process parameters such as the shape, size, and position of objects in two-dimensional or three-dimensional space. Through this method, each furniture symbol (such as sofas, tables, chairs, etc.) and building element (such as walls, doors, windows, etc.) in the floor plan can be converted into specific spatial parameters, such as the length, width, and height of the furniture.

[0133] The specific steps are as follows: First, identify the set of wall points in the building elements and furniture symbols, and then use the convex hull algorithm for the identified set of wall points to generate a spatial circumscribed polygon; after calculating the pixel area \(A_{pixel}\) of the polygon, the user inputs the actual area \(A_{real}\) to establish the area ratio coefficient \(\alpha\):

[0134] \(\alpha = A_{real} / A_{pixel}\);

[0135] Further calculate the linear conversion factor \(\beta\):

[0136]

[0137] For each furniture target \(i\) in the preprocessed floor plan identified, extract the bounding box coordinates, calculate its pixel size, and then apply the linear conversion factor to obtain the actual size:

[0138] L i = L _pi ×\(\beta\)W i = W pi ×\(\beta\);

[0139] Among them, L i 、W i respectively represent the actual length and width of the preprocessed floor plan \(i\), and L _pi 、W pi represent the pixel length and width of the corresponding furniture target.

[0140] Finally, encapsulate the above calculation results into a standard data structure, including furniture types, dimension parameters, and furniture introductions.

[0141] S305: Integrate all the furniture space parameters to obtain the set of the furniture space parameters.

[0142] Finally, summarize and organize the spatial parameters (such as dimensions, etc.) of each recognized piece of furniture to form a complete data set. These parameters come from the conversion results of different furniture symbols. By integrating this data, a systematic set of furniture spatial parameters can be obtained, which contains detailed information about all furniture in the space. Such a set can provide the necessary data support for further interior design optimization, spatial layout analysis, or automated design generation.

[0143] In a possible implementation, since the floor plan of a house may be a raster image or a vector image, and its essential properties remain unchanged when converting the floor plan of the house to obtain the target floor plan, only the file format is converted. Therefore, the target floor plan may be a raster image or a vector image.

[0144] When the target floor plan is a raster image, the present application provides a preprocessing method, such as Figure 4 , Figure 4 is the flowchart of a floor plan preprocessing method provided by an embodiment of the present application. Correspondingly, step S302 preprocesses the target floor plan to obtain a preprocessed floor plan, which can be specifically implemented through steps S401 - S404:

[0145] S401: Use the Gaussian filtering algorithm to perform noise reduction processing on the target floor plan to obtain a noise-reduced floor plan.

[0146] When the target floor plan is a raster image, the Gaussian filtering algorithm can be applied to perform noise reduction processing on the target floor plan. The purpose of this process is to remove the noise in the image (such as interference generated during scanning or transmission), making the image clearer and providing a cleaner image basis for subsequent processing steps.

[0147] S402: Apply the histogram equalization algorithm to enhance the contrast of the noise-reduced floor plan to obtain an enhanced floor plan.

[0148] Then apply the histogram equalization algorithm to the noise-reduced floor plan to enhance the contrast of the image. This step adjusts the brightness distribution of the image, making the details of the image more obvious, especially in the darker or brighter areas of the image, which helps to highlight the key features in the image.

[0149] S403: Use the edge detection algorithm to extract the key edge features of the enhanced floor plan to obtain an edge feature map.

[0150] Based on the image with enhanced contrast, apply the edge detection algorithm to extract the key edge features in the image. This step helps to identify the important boundary lines in the floor plan (such as walls, doors, and windows), and these edge features are crucial for subsequent image analysis and object recognition.

[0151] S404: Normalize the resolution of the edge feature map to obtain a first normalized image.

[0152] Finally, perform resolution normalization on the extracted edge feature map to ensure the unity of the image resolution, so that there is no distortion or deformation when the image is displayed on different devices and platforms. This step guarantees the accuracy of subsequent image processing or model analysis.

[0153] Ultimately, after these processes, the obtained "first normalized image" is the so-called "preprocessed floor plan". Such an image has removed noise, enhanced contrast, extracted key edge information, and ensured resolution consistency, making it suitable for subsequent analysis tasks such as the recognition of building elements and furniture symbols.

[0154] When the target floor plan is a vector map, the present application provides another preprocessing method, such as Figure 5 , Figure 5 is the flowchart of another floor plan preprocessing method provided by the embodiment of the present application. Correspondingly, step S302 preprocesses the target floor plan to obtain a preprocessed floor plan, which can specifically be implemented through steps S501 - S503:

[0155] S501: Perform vector parsing on the target floor plan to obtain vector data.

[0156] When the target floor plan is a vector map, first perform vector parsing on the target floor plan to obtain vector data, so as to extract the vector information therein and convert it into digital - form data. This process includes decomposing elements such as geometric shapes and lines in the floor plan into a series of mathematical descriptions (such as points, lines, curves, etc.) to form vector data. Vector data can accurately represent the shape and position of the graphic and is independent of resolution, so it will not be distorted when scaled. This process helps to extract the key geometric structure information in the floor plan, facilitating subsequent further processing and analysis.

[0157] S503: Perform raster image conversion on the vector data to obtain a rasterized image.

[0158] Since subsequent processing may require the raster image format, the system can convert the vector image into a high-precision raster image to ensure that details such as lines are retained as much as possible. That is, after obtaining the vector data, the vector data is converted into a pixelated image form. This process includes converting elements such as geometric shapes and lines in the vector data into a raster image composed of pixel points. During the rasterization process, each vector graphic is decomposed into a series of pixel points, which are arranged according to a specific resolution to form the final rasterized image. This process enables the image to be displayed and processed in pixel form on devices such as monitors or printers, facilitating subsequent image processing and analysis tasks.

[0159] S503: Normalize the resolution of the rasterized image to obtain a second normalized image.

[0160] Finally, standardize the resolution of the rasterized image to reach a unified size and clarity standard. This process includes adjusting the pixel density of the rasterized image to ensure that all images have the same resolution. Through this process, images from different sources or with different original resolutions can be uniformly processed, thus maintaining consistency and comparability in subsequent analysis and operations. The finally obtained image is called the second normalized image, which is a rasterized image at a standard resolution.

[0161] Among them, the second normalized image is the preprocessed floor plan.

[0162] The "second normalized image" obtained through these steps of processing is the so-called "preprocessed floor plan". This image has completed a series of processes such as vector parsing, rasterization, and resolution normalization, and can provide a clearer and more unified image basis for subsequent analysis and recognition.

[0163] See Figure 6 , Figure 6 which is the flowchart of a method for converting a 3D model of a furniture drawing provided by an embodiment of this application. Correspondingly, step S105 performs a 3D model conversion on the target furniture drawing to obtain the 3D model of the matching furniture drawing, which can be specifically implemented through steps S601 - S603:

[0164] S601: Use the RMBG model to perform foreground and background separation processing on the matching furniture drawing to obtain a foreground furniture drawing.

[0165] During the process of converting the target furniture diagram into a 3D model, first, the RemoveBackground (RMBG) model is used to separate the foreground (i.e., the image part of the furniture) from the background (such as the background wall, floor, etc.) in the matching furniture diagram. The separated image will contain only the content related to the furniture (foreground furniture diagram), which provides a more focused object for subsequent 3D reconstruction. In this process, the RMBG model identifies and removes the background in the image through an algorithm, retaining only the furniture part, thereby obtaining the foreground furniture diagram. This step helps to remove unnecessary background information, making the subsequent processing more focused on the furniture itself and facilitating further operations such as 3D reconstruction.

[0166] S602: Use the Trellis 3D reconstruction model to perform 3D reconstruction on the foreground furniture diagram to obtain a 3D furniture model.

[0167] Next, use the Trellis 3D Reconstruction Model to perform 3D reconstruction on the foreground furniture diagram obtained in step S601. This step analyzes the structure and features of the furniture diagram through an algorithm, converting it from a 2D image to a 3D model. The 3D reconstruction model generates a realistic 3D model by capturing the depth information and structure in the image, usually including information such as the shape, size, and relative position of the object. In this process, the Trellis 3D reconstruction model analyzes the geometric features and structure in the foreground furniture diagram through an algorithm, reconstructing the three-dimensional shape and spatial position of the furniture, and generating a complete 3D furniture model. This step converts the 2D image into a 3D structure, making the furniture more visually realistic and three-dimensional, facilitating further size calibration and design preview.

[0168] S603: Based on the YOLO12 object detection algorithm, perform size ratio calibration on the 3D furniture model to obtain the 3D model of the matching furniture diagram.

[0169] After the 3D furniture model is generated, it is necessary to perform size ratio calibration on it. This step uses the YOLO12 object detection algorithm. YOLO is a common object detection algorithm that can quickly and accurately identify objects. Through the YOLO12 algorithm, the system can identify the size of the furniture model and make adjustments so that the size of the 3D model matches the size of the actual furniture, thereby obtaining a real and proportional 3D furniture model.

[0170] Specifically, during the process of performing size ratio calibration on the 3D furniture model based on the YOLO12 object detection algorithm, first, obtain the actual area A_real of the furniture's planar projection recognized by the YOLO12 object detection algorithm from the planar diagram.

[0171] Secondly, obtain the actual wall height parameter H_wall input by the user. If the user does not explicitly provide this parameter, the system uses the standard height (e.g., the default value of 2.8 meters).

[0172] Then, extract the dimension parameters from the initial 3D furniture model output by the 3D reconstruction sub-module, including the initial projected area A_model of the model bottom surface and the initial height H_model of the model.

[0173] Next, calculate the area scaling coefficient S_area:

[0174]

[0175] Furthermore, calculate the height scaling coefficient S_height:

[0176]

[0177] In the formula, R_proportion is an empirical proportion coefficient used to control the reasonable proportion range between the furniture height and the room wall height. The value range is set according to the furniture type experience. For example:

[0178] Sofa furniture: R_proportion = 0.3 - 0.4;

[0179] Bookcase furniture: R_proportion = 0.7 - 0.9;

[0180] Table and chair furniture: R_proportion = 0.4 - 0.6.

[0181] Finally, perform a scaling transformation on the 3D model:

[0182] The length and width dimensions of the model are scaled proportionally with S_area as the coefficient;

[0183] The model height is scaled independently with S_height as the coefficient.

[0184] In particular, for specific furniture that needs to fit the wall or floor (such as wardrobes, beds, cabinets), the system can further fine-tune the geometric position parameters of the model to ensure that the placement position accurately fits the corresponding wall or floor position in the actual 3D space.

[0185] The 3D model obtained through steps S601 - S603 provides a real and accurate 3D representation for the subsequent application of the matching furniture drawing, and may be used in multiple fields such as virtual reality (VR), augmented reality (AR), and interior design.

[0186] In a possible implementation manner, the method further includes:

[0187] Perform 3D scene rendering on the 3D model of the matched furniture drawing, so that the user can rotate or zoom the perspective of the 3D model in the 3D scene through mouse or touch operations.

[0188] This step is to place the generated 3D furniture model in a 3D scene for rendering and provide user interaction functions. Specifically, the system can first convert the 3D model in the furniture drawing into a 3D scene that can be displayed in a virtual environment through computer graphics technology. This process is called "3D scene rendering", which will present various details of the model, such as materials, colors, and lighting effects, on the computer screen. Next, the user can interact with this 3D scene through mouse or touch operations. That is to say, the user can rotate the perspective by dragging the mouse or finger to view different angles of the furniture model, or zoom in or out of the model through pinch gestures to help the user better observe and understand the appearance and size of the furniture. This interactive operation method enables the user to freely adjust the perspective in the virtual environment and obtain a more real and intuitive experience, as if personally entering a real space.

[0189] Among them, 3D scene rendering: Place the 3D furniture model in a 3D scene for high-quality rendering, making the model more visually real and three-dimensional.

[0190] User interaction function: The user can interact with the 3D scene through mouse or touch operations to achieve the following functions:

[0191] (1) Rotate perspective: The user can rotate the 3D model to view the appearance and structure of the furniture from different angles.

[0192] (2) Zoom perspective: The user can zoom in or out of the 3D model to observe the details of the furniture in more detail.

[0193] Through this process, the user can freely view and operate the furniture model in an interactive 3D scene, thereby obtaining a more intuitive and detailed design preview experience.

[0194] In a possible implementation manner, the selecting q from the p matched furniture drawings based on the user preference and presenting them to the user for selection to obtain the user selection result includes:

[0195] Screen the p matched furniture drawings based on the furniture required size description to obtain n final matched furniture drawings, and then select q from the n final matched furniture drawings based on the user preference and present them to the user for selection to obtain the user selection result.

[0196] When selecting furniture from the matching furniture to display to the user for selection, first, based on the furniture demand size description, p matching furniture diagrams can be screened to obtain n final matching furniture diagrams. In this screening process, only those furniture diagrams with furniture size parameters that meet or are smaller than the user's demand size will be selected, so as to ensure that each finally displayed furniture diagram can meet the user's specific requirements for size. Next, based on the user's personal preferences, q are selected from these n final matching furniture diagrams for display, so that the user can select from them and finally obtain the selection result of the user.

[0197] It should be noted that n is a positive integer, and n is less than p and greater than q. This means that the number of screened matching furniture diagrams is less than the total number p of the initial matching furniture diagrams, but still maintains sufficient diversity for the user to choose from, and the q furniture diagrams displayed to the user are the results of further screening based on their preferences.

[0198] See Figure 7 , Figure 7 is a schematic structural diagram of a furniture recognition and matching device provided by an embodiment of the present application. As Figure 7 shown, the furniture recognition and matching device includes:

[0199] An acquisition unit 701, configured to acquire a floor plan of a house, a furniture demand description, and user preferences; both the floor plan of the house and the furniture demand description are provided by the user; the furniture demand description includes a furniture demand feature description;

[0200] A parameter extraction unit 702, configured to extract all furniture space parameters in the floor plan of the house to obtain a furniture space parameter set;

[0201] A furniture retrieval unit 703, configured to perform furniture retrieval in a multimodal furniture knowledge base based on the furniture space parameter set and the furniture demand feature description to obtain p matching furniture diagrams;

[0202] A furniture diagram display unit 704, configured to select q from the p matching furniture diagrams based on the user preferences and display them to the user for selection to obtain a user selection result; the multimodal furniture knowledge base includes multiple furniture diagrams, and each furniture diagram is marked with a corresponding furniture introduction and furniture size parameters; both p and q are positive integers, and q is less than p;

[0203] A determination unit 705, configured to determine a target furniture diagram among the q matching furniture diagrams based on the user selection result;

[0204] A three-dimensional model conversion unit 706, configured to perform three-dimensional model conversion on the target furniture diagram to obtain a three-dimensional model of the matching furniture diagram.

[0205] In a possible implementation, the furniture retrieval unit 703 specifically includes:

[0206] A vectorization unit, configured to vectorize the furniture demand feature description to obtain a demand feature vector, and vectorize each furniture drawing and furniture introduction in the multimodal furniture knowledge base to obtain a plurality of furniture picture feature vectors and furniture introduction feature vectors;

[0207] A first calculation unit, configured to calculate the similarity between the plurality of furniture introduction feature vectors and the demand feature vector;

[0208] A first screening unit, configured to screen the furniture drawings corresponding to the top x furniture introduction feature vectors with the highest similarity as the first candidate furniture drawings; x is a positive integer;

[0209] A second calculation unit, configured to calculate the similarity between the furniture picture feature vectors of the x first candidate furniture drawings and the demand feature vector;

[0210] A second screening unit, configured to screen the furniture corresponding to the top y furniture picture feature vectors with the highest similarity as the second candidate furniture drawings; y is a positive integer, and y is less than x;

[0211] A third screening unit, configured to screen the y second candidate furniture drawings based on the furniture space parameters in the furniture space parameter set to obtain z third candidate furniture drawings; the furniture size parameters of the z third candidate furniture drawings are all less than or equal to the furniture space parameters in the furniture space parameter set; z is a positive integer, and z is less than y;

[0212] A fourth screening unit, configured to screen p fourth candidate furniture drawings from the z third candidate furniture drawings as the matching furniture drawings; the p fourth candidate furniture drawings are the p third candidate furniture drawings with the smallest difference between the furniture size parameters and the furniture space parameters; p is less than z.

[0213] In a possible implementation, the parameter extraction unit 702 specifically includes:

[0214] An image format conversion unit, configured to convert the image format of the house floor plan into a target format to obtain a target floor plan;

[0215] A preprocessing unit, configured to preprocess the target floor plan to obtain a preprocessed floor plan;

[0216] An identification unit, configured to identify building elements and furniture symbols in the preprocessed floor plan based on a deep neural network with the YOLO architecture;

[0217] A furniture space parameter conversion unit, configured to convert the building elements and the furniture symbols into corresponding furniture space parameters one by one according to the principles of computational geometry;

[0218] A data integration unit, configured to integrate all the furniture space parameters to obtain the set of the furniture space parameters.

[0219] In a possible implementation manner, when the target floor plan is a raster image, the preprocessing unit specifically includes:

[0220] A noise reduction unit, configured to perform noise reduction processing on the target floor plan by using a Gaussian filtering algorithm to obtain a noise-reduced floor plan;

[0221] A contrast enhancement unit, configured to enhance the contrast of the noise-reduced floor plan by applying a histogram equalization algorithm to obtain an enhanced floor plan;

[0222] An edge feature extraction unit, configured to extract key edge features of the enhanced floor plan by using an edge detection algorithm to obtain an edge feature map;

[0223] A first normalization unit, configured to perform resolution normalization on the edge feature map to obtain a first normalized image;

[0224] Wherein, the first normalized image is the preprocessed floor plan.

[0225] In a possible implementation manner, when the target floor plan is a vector graph, the preprocessing unit specifically includes:

[0226] A vector analysis unit, configured to perform vector analysis on the target floor plan to obtain vector data;

[0227] A raster image conversion unit, configured to perform raster image conversion on the vector data to obtain a rasterized image;

[0228] A second normalization unit, configured to perform resolution normalization on the rasterized image to obtain a second normalized image;

[0229] Wherein, the second normalized image is the preprocessed floor plan.

[0230] In a possible implementation manner, the 3D model conversion unit 706 specifically includes:

[0231] A foreground-background separation unit, configured to perform foreground and background separation processing on the matched furniture image by using the RMBG model to obtain a foreground furniture image;

[0232] A 3D reconstruction unit, configured to perform 3D reconstruction on the foreground furniture image by using the Trellis 3D reconstruction model to obtain a 3D furniture model;

[0233] A dimension ratio calibration unit for calibrating the dimension ratio of the three-dimensional furniture model based on the YOLO12 object detection algorithm to obtain the three-dimensional model of the matching furniture drawing.

[0234] In a possible implementation, the device further includes:

[0235] A three-dimensional scene rendering unit for performing three-dimensional scene rendering on the three-dimensional model of the matching furniture drawing, so that the user can rotate or zoom the perspective of the three-dimensional model in the three-dimensional scene through mouse or touch operations.

[0236] In a possible implementation, the furniture requirement description further includes a furniture requirement dimension description;

[0237] In a possible implementation, the furniture drawing display unit 704 specifically includes:

[0238] A fifth screening unit for screening the p matching furniture drawings based on the furniture requirement dimension description to obtain n final matching furniture drawings; the furniture size parameters of the final matching furniture drawings are all less than or equal to the furniture requirement dimension description;

[0239] A selection and display unit for selecting q from the n final matching furniture drawings based on the user preference and displaying them to the user for selection to obtain the user selection result; n is a positive integer, n is less than p and greater than q.

[0240] In addition, an embodiment of the present application further provides a furniture recognition and matching device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the furniture recognition and matching method described above is implemented.

[0241] The embodiment of the present application provides a furniture recognition and matching device. By using the floor plan of the house and the furniture requirement description provided by the user to search for furniture in the multi-modal furniture knowledge base, it can more accurately meet the personalized needs of the user. At the same time, an interactive selection mechanism based on user preference is adopted to screen out several from multiple matching furniture drawings and display them to the user, allowing the user to participate in the final selection, thereby further improving user satisfaction and the degree of personalization of the design.

[0242] The above has introduced in detail a furniture recognition and matching method and related devices provided by this application. The various embodiments in the specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. For the same or similar parts among the various embodiments, reference can be made to each other. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple. For the relevant parts, reference can be made to the description in the method section. It should be noted that for those of ordinary skill in the art in this technical field, without departing from the principle of this application, several improvements and modifications can be made to this application, and these improvements and modifications also fall within the protection scope of the claims of this application.

[0243] It should be understood that in this application, "at least one (item)" means one or more, and "a plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships can exist. For example, "A and / or B" can mean: only A exists, only B exists, and both A and B exist at the same time. Among them, A and B can be singular or plural. The character " / " generally means that the associated objects before and after are in an "or" relationship. "At least one (piece) of the following" or its similar expression means any combination of these items, including any combination of single item (piece) or plural items (pieces). For example, at least one (piece) of a, b, or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0244] It should also be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of another identical element in the process, method, article or device including the said element.

Claims

1. A furniture recognition and matching method, characterized in that, The method includes: Obtaining a floor plan of a house, a furniture requirement description, and user preferences; both the floor plan of the house and the furniture requirement description are provided by the user; the furniture requirement description includes a description of furniture requirement characteristics; Extracting all furniture space parameters in the floor plan of the house to obtain a furniture space parameter set; Performing furniture retrieval in a multi-modal furniture knowledge base based on the furniture space parameter set and the furniture requirement characteristics to obtain p matching furniture diagrams, and selecting q from the p matching furniture diagrams based on the user preferences to display to the user for selection to obtain a user selection result; the multi-modal furniture knowledge base includes multiple furniture diagrams, and each furniture diagram is labeled with a corresponding furniture introduction and furniture size parameters; both p and q are positive integers, and q is less than p; Determining a target furniture diagram among the q matching furniture diagrams based on the user selection result; Performing three-dimensional model conversion on the target furniture diagram to obtain a three-dimensional model of the matching furniture diagram.

2. The method according to claim 1, wherein The performing furniture retrieval in a multi-modal furniture knowledge base based on the furniture space parameter set and the furniture requirement characteristics to obtain p matching furniture diagrams includes: Vectorizing the furniture requirement characteristics description to obtain a requirement feature vector, and vectorizing each furniture diagram and the furniture introduction of the furniture diagram in the multi-modal furniture knowledge base to obtain multiple furniture picture feature vectors and furniture introduction feature vectors; Calculating the similarity between the multiple furniture introduction feature vectors and the requirement feature vector, and screening the furniture diagrams corresponding to the top x furniture introduction feature vectors with the highest similarity as the first candidate furniture diagrams; x is a positive integer; Calculating the similarity between the furniture picture feature vectors of the x first candidate furniture diagrams and the requirement feature vector, and screening the furniture corresponding to the top y furniture picture feature vectors with the highest similarity as the second candidate furniture diagrams; y is a positive integer, and y is less than x; Screening the y second candidate furniture diagrams based on the furniture space parameters in the furniture space parameter set to obtain z third candidate furniture diagrams; the furniture size parameters of the z third candidate furniture diagrams are all less than or equal to the furniture space parameters in the furniture space parameter set; z is a positive integer, and z is less than y; Screening p fourth candidate furniture diagrams from the z third candidate furniture diagrams as the matching furniture diagrams; the p fourth candidate furniture diagrams are the p third candidate furniture diagrams with the smallest difference between the furniture size parameters and the furniture space parameters; p is less than z.

3. The method according to claim 1, characterized in that The extracting all furniture space parameters in the floor plan of the house to obtain a furniture space parameter set includes: Converting the image format of the floor plan of the house to a target format to obtain a target floor plan; Preprocessing the target floor plan to obtain a preprocessed floor plan; Identifying building elements and furniture symbols in the preprocessed floor plan based on a deep neural network with the YOLO architecture; Converting the building elements and the furniture symbols into corresponding furniture space parameters one by one according to the principles of computational geometry; Integrating all furniture space parameters to obtain the furniture space parameter set.

4. The method according to claim 3, characterized in that, When the target floor plan is a raster image, the preprocessing of the target floor plan to obtain a preprocessed floor plan includes: Performing noise reduction processing on the target floor plan using a Gaussian filtering algorithm to obtain a noise-reduced floor plan; Applying a histogram equalization algorithm to enhance the contrast of the noise-reduced floor plan to obtain an enhanced floor plan; Using an edge detection algorithm to extract the key edge features of the enhanced floor plan to obtain an edge feature map; Performing resolution normalization on the edge feature map to obtain a first normalized image; Wherein, the first normalized image is the preprocessed floor plan.

5. The method according to claim 3, wherein When the target floor plan is a vector image, the preprocessing of the target floor plan to obtain a preprocessed floor plan includes: Performing vector analysis on the target floor plan to obtain vector data; Performing raster image conversion on the vector data to obtain a rasterized image; Performing resolution normalization on the rasterized image to obtain a second normalized image; Wherein, the second normalized image is the preprocessed floor plan.

6. The method according to claim 1, characterized in that, The conversion of the target furniture image into a 3D model to obtain the 3D model of the matching furniture image includes: Using the Remove Background Model (RMBG) to separate the foreground and background of the matching furniture image to obtain a foreground furniture image; Using the Trellis 3D reconstruction model to perform 3D reconstruction on the foreground furniture image to obtain a 3D furniture model; Based on the YOLO12 object detection algorithm, performing size ratio calibration on the 3D furniture model to obtain the 3D model of the matching furniture image.

7. The method according to claim 1, wherein The method further includes: Performing 3D scene rendering on the 3D model of the matching furniture image, so that the user can rotate or zoom the perspective of the 3D model in the 3D scene through mouse or touch operations.

8. The method according to claim 1, characterized in that The furniture requirement description further includes a furniture requirement size description; The selection of q matching furniture images from the p matching furniture images based on the user preference and presenting them to the user for selection to obtain a user selection result includes: Screening the p matching furniture images based on the furniture requirement size description to obtain n final matching furniture images; the furniture size parameters of the final matching furniture images are all less than or equal to the furniture requirement size description; Selecting q from the n final matching furniture images based on the user preference and presenting them to the user for selection to obtain the user selection result; n is a positive integer, n is less than p and greater than q.

9. A furniture recognition and matching device, characterized in that, The device includes: An acquisition unit, configured to acquire a house floor plan, a furniture requirement description, and a user preference; both the house floor plan and the furniture requirement description are provided by the user; the furniture requirement description includes a furniture requirement feature description; A parameter extraction unit, configured to extract all furniture space parameters in the house floor plan to obtain a furniture space parameter set; A furniture retrieval unit, configured to perform furniture retrieval in a multi-modal furniture knowledge base based on the furniture space parameter set and the furniture requirement feature description to obtain p matching furniture images; A furniture diagram display unit for selecting q furniture diagrams from the p matching furniture diagrams based on the user preferences and presenting them to the user for selection to obtain a user selection result; the multimodal furniture knowledge base includes multiple furniture diagrams, and each furniture diagram is labeled with a corresponding furniture introduction and furniture size parameters; both p and q are positive integers, and q is less than p; A determination unit for determining a target furniture diagram among the q matching furniture diagrams based on the user selection result; A three-dimensional model conversion unit for performing three-dimensional model conversion on the target furniture diagram to obtain a three-dimensional model of the matching furniture diagram.

10. A furniture recognition and matching device, characterized in that, Comprising: A memory, a processor, and a computer program stored on the memory and executable on the processor, wherein when the processor executes the computer program, the furniture recognition and matching method according to any one of claims 1-7 is implemented.

Images