WO2021196515A1 - 户型图生成方法、装置、计算机设备和存储介质 - Google Patents

户型图生成方法、装置、计算机设备和存储介质 Download PDF

Info

Publication number
WO2021196515A1
WO2021196515A1 PCT/CN2020/111335 CN2020111335W WO2021196515A1 WO 2021196515 A1 WO2021196515 A1 WO 2021196515A1 CN 2020111335 W CN2020111335 W CN 2020111335W WO 2021196515 A1 WO2021196515 A1 WO 2021196515A1
Authority
WO
WIPO (PCT)
Prior art keywords
house
target building
boundary
plan
room
Prior art date
Application number
PCT/CN2020/111335
Other languages
English (en)
French (fr)
Inventor
胡瑞珍
黄惠
张皓
黄泽宇
汤雨涵
Original Assignee
深圳大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 深圳大学 filed Critical 深圳大学
Publication of WO2021196515A1 publication Critical patent/WO2021196515A1/zh
Priority to US17/885,317 priority Critical patent/US20220383572A1/en

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T11/00—2D [Two Dimensional] image generation
    • G06T11/60—Editing figures and text; Combining figures or text
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/10—Geometric CAD
    • G06F30/12—Geometric CAD characterised by design entry means specially adapted for CAD, e.g. graphical user interfaces [GUI] specially adapted for CAD
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/10—Geometric CAD
    • G06F30/13—Architectural design, e.g. computer-aided architectural design [CAAD] related to design of buildings, bridges, landscapes, production plants or roads
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T11/00—2D [Two Dimensional] image generation
    • G06T11/20—Drawing from basic elements, e.g. lines or circles
    • G06T11/203—Drawing of straight lines or curves
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T11/00—2D [Two Dimensional] image generation
    • G06T11/20—Drawing from basic elements, e.g. lines or circles
    • G06T11/206—Drawing of charts or graphs
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
    • G06V10/449—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters
    • G06V10/451—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters with interaction between the filter responses, e.g. cortical complex cells
    • G06V10/454—Integrating the filters into a hierarchical structure, e.g. convolutional neural networks [CNN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/778—Active pattern-learning, e.g. online learning of image or video features
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00—Details relating to CAD techniques
    • G06F2111/04—Constraint-based CAD
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2210/00—Indexing scheme for image generation or computer graphics
    • G06T2210/04—Architectural design, interior design
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2210/00—Indexing scheme for image generation or computer graphics
    • G06T2210/12—Bounding box

Definitions

  • This application relates to the field of computer technology, and in particular to a method, device, computer equipment and storage medium for generating a house plan.
  • the related method of automatically generating house plans can quickly generate house plans, but because it only generates house plans through simple search and matching in the data set, the quality of the generated house plans is not high.
  • a method for generating a house plan includes:
  • the layout constraints include the room type, the number of rooms, the location of the room, and the adjacency relationship between the rooms;
  • each second house plan For each second house plan, apply the layout constraints of the second house plan to the boundary of the target building to obtain a layout plan of the target building corresponding to each second house plan;
  • the predicted house map includes the room border of each room in the target building and the positional relationship of each room border in the boundary of the target building.
  • a plurality of first house plans are given according to the layout constraints of the target building, including:
  • a plurality of first house plans that meet the layout constraints of the target building are retrieved in the pre-stored house plan data set.
  • a plurality of second house plans are filtered from the plurality of first house plans, and the degree of matching between the boundaries of the plurality of second house plans and the boundary of the target building satisfies a preset condition ,include:
  • the plurality of first house plans corresponding to the cumulative difference being less than the preset difference threshold is used as the plurality of second house plans.
  • the layout constraints of the multiple second house plans are applied to the boundary of the target building to obtain the layout of the target building corresponding to each second house plan Figures, including:
  • the room type, the number of rooms, the location of the room, and the adjacency relationship between the rooms of the adjusted second house plan are arranged in the boundary of the target building to obtain each second house plan.
  • the layout plan of the target building corresponding to the second-family plan.
  • the layout drawing of the target building and the boundary of the target building are input into the house plan generation network to obtain the target building output by the house plan generation network
  • the forecasted house plan including:
  • the bounding box optimization network includes a third convolutional neural network and a region of interest Pooling layer and second multilayer perceptron.
  • the method further includes:
  • the training data set including a plurality of house plan data
  • the parameters of the trained house map generation network are adjusted to obtain the house map generation network.
  • the method further includes:
  • a device for generating a house plan including a processor and a memory, and a computer program is stored in the memory.
  • the processor executes the computer program:
  • the layout constraints include the room type, the number of rooms, the location of the room, and the adjacency relationship between the rooms;
  • each second house plan For each second house plan, apply the layout constraints of the second house plan to the boundary of the target building to obtain a layout plan of the target building corresponding to each second house plan;
  • the predicted house map includes the room border of each room in the target building and the positional relationship of each room border in the boundary of the target building.
  • a computer device including a memory and a processor, the memory stores a computer program, and the processor implements the following steps when the computer program is executed:
  • the layout constraints include the room type, the number of rooms, the location of the room, and the adjacency relationship between the rooms;
  • each second house plan For each second house plan, apply the layout constraints of the second house plan to the boundary of the target building to obtain a layout plan of the target building corresponding to each second house plan;
  • the predicted house map includes the room border of each room in the target building and the positional relationship of each room border in the boundary of the target building.
  • a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:
  • the layout constraints include the room type, the number of rooms, the location of the room, and the adjacency relationship between the rooms;
  • each second house plan For each second house plan, apply the layout constraints of the second house plan to the boundary of the target building to obtain a layout plan of the target building corresponding to each second house plan;
  • the predicted house map includes the room border of each room in the target building and the positional relationship of each room border in the boundary of the target building.
  • the above-mentioned method, device, computer equipment and storage medium of the house plan obtain the boundary of the target building and the layout constraints of the target building.
  • the layout constraints include the room type, the number of rooms, the location of the room, and the adjacency relationship between the rooms;
  • a plurality of first house plans are given; then a plurality of second house plans are filtered from the plurality of first house plans, and the boundary of the plurality of second house plans is related to the target building
  • the matching degree between the boundaries meets the preset condition; for each second house plan, the layout constraints of the second house plan are applied to the boundary of the target building to obtain the target corresponding to each second house plan
  • the layout of the building finally for the layout of each target building, input the layout of the target building and the boundary of the target building into the house plan generation network to obtain the target building output by the house plan generation network
  • a predicted house map of the object, the predicted house map includes the room frame of each room in the target building and the positional relationship of each room frame in the boundary of the
  • the house plan generation method not only considers the boundaries of the building when generating the house plan, but also fully considers the layout constraints of the building, so that the final predicted house plan is more in line with actual needs, and the generated house plan is improved.
  • the quality of the graph is improved.
  • FIG. 1 is a schematic flowchart of a method for generating a house plan in an embodiment
  • FIG. 2 is a schematic flowchart of a method of filtering multiple second house plans from a plurality of first house plans in a method for generating house plans in an embodiment
  • Fig. 3 is an embodiment of the method for generating a house plan for each second house plan, applying the layout constraints of the second house plan to the boundary of the target building to obtain the target building corresponding to each second house plan Schematic diagram of the flow chart of the layout method;
  • Figure 4 shows the layout of each target building in the method for generating a house plan in another embodiment.
  • the layout of the target building and the boundary of the target building are input into the house plan generation network to obtain the output of the house plan generation network.
  • Fig. 5 is a schematic diagram of a specific structure of a house plan generation network in an embodiment
  • FIG. 6 is a schematic diagram of a specific structure of a bounding box optimization network in an embodiment
  • FIG. 7 is a schematic flowchart of a method of training a house diagram generation network in a house diagram generation method in an embodiment
  • FIG. 8 is a schematic diagram of the overall flow of a method for generating a house plan in an embodiment
  • Figure 9 is a schematic diagram of the boundary and steering function of the target building in an embodiment
  • FIG. 10 is a schematic diagram of the alignment adjustment of the predicted house plan in the method for generating the house plan in an embodiment
  • FIG. 11 is a schematic diagram of a set of example results generated by different boundary inputs and different user constraint conditions in the method for generating a house plan in an embodiment
  • FIG. 12 is a structural block diagram of an apparatus for generating a house plan in an embodiment
  • FIG. 13 is a structural block diagram of an apparatus for generating a house plan in another embodiment
  • Fig. 14 is a diagram of the internal structure of a computer device in an embodiment.
  • a method for generating a house plan is provided.
  • the method is applied to a terminal as an example for description.
  • the terminal may be, but not limited to, various personal computers, notebook computers, and smart phones. , Tablet computers and portable wearable devices, the method includes the following steps:
  • Step 101 The terminal obtains the boundary of the target building and the layout constraints of the target building.
  • the layout constraints include the room type, the number of rooms, the location of the room, and the adjacency relationship between the rooms.
  • the boundary of the target building mentioned in this step refers to the outline shape of the target building.
  • the user can manually input the boundary of the target building into the terminal.
  • the measured house profile data is sent to the terminal.
  • the terminal has a measurement function, and can directly measure the target building to obtain the boundary of the target building.
  • the layout constraints of the target building are input to the terminal by the user according to actual needs.
  • a dialog box can be displayed on the terminal so that the user can enter the required room type, the number of rooms, the location of the room, and the room in the dialog box of the terminal.
  • the terminal can display the boundary of the target building to the user.
  • the user can use an external device such as a mouse to roughly draw the desired room type, the number of rooms, the location of the room, and the adjacency relationship between the rooms on the boundary of the target building. Wait.
  • Step 102 The terminal outputs a plurality of first house plans according to the layout constraints of the target building.
  • the terminal will give a first house plan that meets the layout constraints required by the user.
  • the number of rooms, room types, room locations, and the adjacency relationship between the rooms of these first floor plans can fully meet user requirements.
  • the layout constraints of the output first house plans may also partially meet user requirements. For example, if the number of rooms required by the user is 4, the number of rooms in the output of the first floor plan can be 3 or 5, and the degree of matching between the layout constraints of the first floor plan and the layout constraints entered by the user It can be preset.
  • Step 103 The terminal screens out a plurality of second house plans from the plurality of first house plans, and the degree of matching between the boundaries of the plurality of second house plans and the boundary of the target building meets a preset condition.
  • the layout constraints of the first house plan meet the needs of users, but the boundary of the first house plan does not necessarily conform to the boundary of the target building. Therefore, it is necessary to filter the first house plan again in order to obtain a house plan with layout constraints and boundaries that meet the requirements.
  • the method for judging the degree of matching between the boundary of the first house plan and the boundary of the target building may be to enclose the area enclosed by the boundary of the first house plan and the boundary of the target building.
  • the area is overlapped and compared, the ratio of the area of the overlapping part to the area enclosed by the boundary of the target building is calculated, and the first house plan whose ratio exceeds the preset area threshold is selected as the second house plan.
  • the terminal displays each first floor plan within the boundary of the target building input by the user, so that the user can intuitively analyze the matching degree between the first floor plan and the boundary, and thereby select the second floor plan.
  • the method of judging the degree of matching between the boundary of the first house plan and the boundary of the target building may also be to calculate the boundary perimeter and the number of corners of the first house plan, calculate the boundary perimeter and the number of corners of the target building, and then The difference between the boundary perimeter of the first house plan and the boundary perimeter of the target building is smaller than the preset perimeter difference threshold, and the difference between the number of corners of the first house plan and the number of corners of the target building is smaller than the preset
  • the first house plan of the quantity difference threshold is filtered out and used as the second house plan.
  • the steering function of the boundary of the first house plan and the steering function of the boundary of the target building can be calculated, and then the two steering functions are compared, and the second house plan can be filtered from the first house plan through the comparison result.
  • Step 104 For each second house plan, the terminal applies the layout constraints of the second house plan to the boundary of the target building to obtain a layout plan of the target building corresponding to each second house plan.
  • the layout of the second house plan can be directly applied to the boundary of the target building.
  • the filtered second house plans still have multiple layout forms. Because of the same or similar boundary between house plans, the applicability of layout constraints will be stronger after they are directly used. Therefore, the layout constraints in the second house plan can be directly applied to the boundary of the target building to obtain multiple layout plans of the target building.
  • Step 105 For the layout plan of each target building, the terminal inputs the layout plan of the target building and the boundary of the target building into the house plan generation network to obtain the prediction of the target building output by the house plan generation network A house map, where the predicted house map includes the room frame of each room in the target building and the positional relationship of each room frame in the boundary of the target building.
  • the layout drawings of multiple target buildings obtained in the above steps are only multiple layouts applicable to the boundaries of the target buildings, and do not form a specific floor plan of the target buildings. Therefore, it is necessary to input the layout plan of each target building into the house plan generation network to obtain a more accurate house plan.
  • the multiple predicted house plans obtained in this step are available for users to choose and guide users to plan the target buildings.
  • the form of the predicted house plans can be displayed by the terminal in a visual form.
  • the content includes displaying the boundary shape of the target building and displaying the room frame of each room at the corresponding position in the boundary shape.
  • the boundary of the target building and the layout constraints of the target building are obtained.
  • the layout constraints include the room type, the number of rooms, the location of the room, and the adjacency relationship between the rooms; then according to the layout of the target building Constrain to give a plurality of first house plans; then filter out a plurality of second house plans from the plurality of first house plans, and the degree of matching between the boundaries of the plurality of second house plans and the boundary of the target building Satisfy the preset conditions; for each second house plan, apply the layout constraints of the second house plan to the boundary of the target building to obtain the layout plan of the target building corresponding to each second house plan; and finally For the layout plan of each target building, input the layout plan of the target building and the boundary of the target building into the house plan generation network to obtain the predicted house plan of the target building output by the house plan generation network.
  • the predicted house map includes the room border of each room in the target building and the positional relationship of each room border in the boundary of the target building. Because the house plan generation method provided by this application not only considers the boundaries of the building when generating the house plan, but also fully considers the layout constraints of the building, so that the final predicted house plan is more in line with actual needs, and the generated house plan is improved. The quality of the graph.
  • a plurality of first house plans are given according to the layout constraints of the target building, including:
  • a plurality of first house plans that meet the layout constraints of the target building are retrieved in the pre-stored house plan data set.
  • a large amount of house plan data can be stored in the terminal in advance, and then these large amounts of house plan data can be encoded and then stored.
  • the advantage of encoding is that in the subsequent steps, the layout constraints of the target building input by the user can be quickly Retrieve the corresponding first floor plan.
  • the way of encoding the house map may be to encode the room type, the number of rooms, the location of the room, and the adjacency relationship between the rooms in the house map.
  • each room in the house map can be represented as a node. If two rooms have an adjacent relationship in the house map, an edge is added between the nodes corresponding to the room.
  • the room types are divided into 13 types, including living room, master bedroom, second bedroom, guest room, children's room, study room, dining room, bathroom, kitchen, balcony, storage room, wardrobe and front hall. You can use numbers to code each room type as 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, and 13 in the order listed above.
  • the boundary of the target building can be supplemented as a rectangular box to obtain the bounding box of the target building, and then the rectangular box is divided into a K ⁇ K grid, where K is greater than or equal to 1.
  • the grid coordinates of the center of each room in the floor plan are used as the location code of the room.
  • For the adjacency relationship between the rooms first find all the indoor doors in the floor plan, and then use the pair of rooms on both sides of the door as the adjacent pair of rooms. Next, you can check whether the distance between any two rooms is less than a given threshold relative to the room's bounding box, which is equivalent to judging whether the two rooms are adjacent to each other sharing a wall.
  • the edges connecting the two rooms are coded.
  • the coding content can be left, right, top, bottom, top left, top right, bottom left, and bottom right. Indicates the relative positional relationship between two rooms.
  • the edge connecting A and B can be coded as up; but if Using A as the benchmark, B is below A, and the edge connecting A and B can be coded as the bottom.
  • the two coding forms are different, they are essentially the same. Therefore, in actual coding, two relative codes can be stored at the same time for an edge, or a relative code can be randomly selected for storage.
  • the size of the room can also be coded, and the ratio of the area of the room to the area of the entire floor plan is used as the code of the room size.
  • the above coding method can be used on the layout constraints entered by the user. After using the same coding method for the layout constraints entered by the user, the first house plan that meets the user's expectations can be filtered out through relatively simple coding, which greatly improves the terminal Filter the efficiency of the first layout.
  • FIG. 2 provides a method for filtering multiple second house plans from multiple first house plans in a method for generating house plans.
  • the method includes:
  • Step 201 The terminal obtains the steering function of the boundary of the plurality of first house plans and the steering function of the boundary of the target building.
  • the turning function can convert the two-dimensional polygon shape of the boundary into a one-dimensional contour line segment to achieve the purpose of dimensionality reduction, so as to facilitate the comparison of the degree of boundary matching.
  • the one-dimensional contour line segment is located in the coordinate system formed by the normalized boundary circumference and the accumulated angle.
  • the steering function can record the cumulative value of the angle of each turning point (turning angle) in the two-dimensional polygon of the boundary.
  • the cumulative value of the angle of each turning point (turning angle) in the two-dimensional polygon of the boundary may be recorded in a clockwise direction.
  • the location of the front door needs to be considered when comparing the two boundaries, so choose from The position of the front door starts to record the turning function of the target building and the turning function of the first floor plan.
  • Step 202 The terminal calculates the cumulative difference between the steering function of the boundary of the plurality of first house plans and the steering function of the boundary of the target building.
  • the cumulative difference between the two steering functions After obtaining the steering function of the boundary of the target building and the steering function of the first house plan, it is necessary to calculate the cumulative difference between the two steering functions.
  • the difference between the two steering functions and the area enclosed by the coordinate axis may be used as the cumulative difference.
  • Step 203 The terminal uses the plurality of first house plans corresponding to the accumulated difference value less than the preset difference threshold as the plurality of second house plans.
  • a preset difference threshold is set to filter out the second house plan. At this time, there may still be some errors between the boundary of the second house plan selected and the boundary of the target building, but it is within an acceptable range.
  • the first house plan with a cumulative difference of zero can be selected.
  • the two-dimensional boundary comparison problem is converted into the one-dimensional contour line segment mathematical calculation problem, which greatly improves the efficiency and accuracy of screening the second floor plan.
  • Fig. 3 provides a method for generating a house plan for each second house plan.
  • the layout constraints of the second house plan are applied to the boundary of the target building to obtain each second house plan.
  • the method for the layout drawing of the target building corresponding to the two-family plan includes:
  • Step 301 The terminal adjusts each second house plan until the angle between the front door direction of the second house plan and the front door direction of the target building is less than a preset angle threshold, and the adjusted second house plan is obtained. picture.
  • the second house plan needs to be adjusted so that the adjusted second house plan's boundary coincides with the boundary of the target building
  • the maximum degree Since the calculation of the degree of boundary overlap is based on the steering function starting from the front door, the front door is used as the reference point for alignment. First align the front doors of the two borders, which also prevents the front door from being blocked by any room.
  • the criterion for judging whether the front door is aligned is to determine whether the angle between the front door direction of the rotated second floor plan and the front door direction of the target building is less than A predetermined angle, such as 45 degrees, etc.
  • the direction of the front door mentioned in this application is the vector connecting the center of the building's bounding box to the center of the door and the vector connecting the center of the bounding box of the rotated second floor plan to the front door .
  • Step 302 For each adjusted second house plan, the terminal arranges the room type, the number of rooms, the location of the room, and the adjacency relationship between the rooms of the adjusted second house plan to the boundary of the target building. Obtain the layout plan of the target building corresponding to each second house plan.
  • the layout constraints of the adjusted second floor plan can be transferred to the boundary of the target building.
  • the transfer method can be a complete copy, that is, the rooms of the target building are laid out completely according to the room type, the number of rooms, the location of the room, and the adjacency relationship between the rooms in the adjusted second floor plan.
  • the adjustment method for the nodes outside the boundary may be to first establish the bounding box and the grid in the bounding box of the target building, then query the grid where the nodes outside the boundary are located, and then set the nodes outside the boundary Move to the grid within the nearest boundary.
  • the above-mentioned adjustment method for nodes outside the boundary may be automatically completed by the terminal, or the terminal may display the layout diagrams of multiple target buildings to the user, and the user may edit it by himself.
  • the layout constraints of the adjusted second house plan are applied within the boundary of the target building, and the nodes outside the boundary are further adjusted, so that the final The obtained layout drawing of the target building is more accurate and more in line with actual needs.
  • Figure 4 provides a layout diagram for each target building in a method for generating a house plan.
  • the layout diagram of the target building and the boundary of the target building are input into the house plan generation network.
  • the method for obtaining the predicted house-type map of the target building output by the house-type map generation network includes:
  • Step 401 For the layout diagram of each target building, the terminal inputs the layout diagram of the target building into the graph neural network to obtain a room feature vector corresponding to each room in the layout diagram of the target building.
  • Step 402 The terminal inputs the boundary of the target building into the first convolutional neural network to obtain the boundary feature vector of the target building.
  • Step 403 The terminal associates each room feature vector with the boundary feature vector to obtain multiple associated feature vectors.
  • Step 404 The terminal inputs each associated feature vector into a first Multilayer Perceptron (MLP) to obtain an initial bounding box corresponding to each room in the layout diagram of the target building.
  • MLP Multilayer Perceptron
  • Step 405 The terminal uses the initial bounding box corresponding to each room to map the associated feature vector corresponding to each room to obtain multiple first feature maps.
  • step 406 the terminal combines the plurality of first feature maps into a second feature map, and then the terminal inputs the second feature map into a second convolutional neural network to obtain a grid floor plan of the target building.
  • Step 407 The terminal inputs the multiple associated feature vectors, the initial bounding box and the raster house plan into a bounding box optimization network to obtain a predicted house plan of the target building.
  • the bounding box optimization network includes a third convolutional neural network. Network, interest area pooling layer and second multilayer perceptron.
  • the above steps 401 to 407 are now described in conjunction with the specific structure of the house map generation network in Fig. 5.
  • the boundary and layout constraints of the target building in Fig. 5 are input into different neural networks to obtain the correspondence of each room.
  • the associated feature vector; the multi-layer perceptron can be used to predict the initial bounding box of the room in the boundary for each room ( Figure 5 uses Means, where i is any positive integer between 0 and n); use the initial bounding box corresponding to each room to map the associated feature vector corresponding to each room to obtain multiple first feature maps;
  • the first feature maps are combined to obtain a second feature map; then the terminal inputs the second feature map to the second convolutional neural network to obtain a raster house map.
  • the output of the house map generation network can be obtained, including the initial bounding box corresponding to each room, the image of the raster house plan, and the optimized bounding box corresponding to each room.
  • the optimized bounding box can be combined as the prediction of the target building Floor plan.
  • the bounding box optimization network includes a third convolutional neural network, a region of interest pooling layer, and a second multilayer perceptron.
  • the raster house map passes through the third convolutional neural network to obtain a third feature map.
  • the initial bounding box corresponding to each room and the third feature map are input into the region of interest pooling layer to obtain a feature vector of a specific length.
  • the feature vector of the specific length and the associated feature vector corresponding to each room are spliced, and input into the second multilayer perceptron to obtain the optimized bounding box corresponding to each room.
  • the visualized position relationship of the room frame of each room in the boundary of the target building can be inferred according to the optimized bounding box of each room.
  • a house map generation network is formed by using a combination of multiple networks to generate a predicted house map, which can make the finally obtained predicted house map more accurate and higher in quality.
  • FIG. 7 provides a method for training a house diagram generation network in a house diagram generation method, and the method includes:
  • Step 701 The terminal obtains a training data set, where the training data set includes a plurality of house pattern data.
  • a large number of house plans can be obtained from the massive resources on the Internet.
  • the obtained large number of house plans can be preprocessed.
  • the preprocessing can include unified conversion of the color house plans. It is a grayscale image, adjusting the resolution of all the house plans to the same value, adjusting all the house plans to the same size, or encoding all the house plans using the coding method described in the above content.
  • the processed house plans can be stored. In the process of training the house map generation network, in fact, all the networks included in Figure 5 are trained.
  • Step 702 The terminal uses the training data set to train the initial house plan generation network to obtain the trained house plan generation network.
  • Step 703 The terminal uses the cross-entropy loss function, the regression loss function, and the geometric loss function to calculate the loss value of the trained house map generation network.
  • Step 704 The terminal adjusts the parameters of the trained house map generation network according to the loss value to obtain the house map generation network.
  • L pix (I) is the image loss of the cross entropy of the predicted house plan and the real house plan, Is the return loss, It is a geometric loss.
  • the geometric loss only applies to the initial bounding box, the initial bounding box The geometric loss is defined as:
  • the ⁇ in (B) set and the ⁇ bd (B) set represent the inner pixels and outer boundary pixels of the bounding box B, respectively.
  • Coverage loss The boundary of the target building should be completely covered by the combination of all room frames. In one embodiment, at any point p ⁇ in (B) should be at least one room border coverage. Therefore, the coverage loss is defined as follows:
  • mutex loss The overlap between room borders should be as small as possible, so that the rooms can be tightly distributed inside the target building. Therefore, the mutex loss can be defined as:
  • Matching loss the border B i of each room should cover the corresponding true bounding box Cover the same area, that is, B i should be displayed in internal. Therefore, the matching loss can be defined as:
  • the method further includes:
  • the final output of the floor plan generation network is an image of the predicted floor plan and the room border of each room.
  • One problem that may occur with room borders is that they may not be well aligned, and some room borders may overlap in certain areas. Therefore, in the final alignment adjustment step, the image of the grid floor plan is used to determine the room type allocation of the overlapping area.
  • adjacent pairs of rooms are aligned according to the coded spatial relationship in the layout diagram. For example, if room frame A is on the left side of room frame B, align the right edge of room frame A with the left edge of room frame B.
  • the adjacent sides of the two room borders are aligned, because the room borders placed side by side in the floor plan should have aligned walls to maximize Reduce the number of corners to a great extent.
  • One room border may be adjacent to a different room border. Therefore, the edges must be updated multiple times.
  • a flag can be set to indicate whether the edges of the box have been updated. If any edge has been updated, it is fixed, and the other edge is aligned with the fixed edge. If the two edges are not fixed, update them to their average position.
  • the graph is constructed by adding a node representing each room and a directed edge from room 1 to room 2. Then, the goal is to find the order in which all graph nodes satisfy the ordering constraints imposed by the directed edges. In order to find such an order, first randomly select any node with an in-degree of 0, and then delete this node and all the edges that point to the rest of the graph.
  • the in-degree is the number of directed edges that point to the node. Continue to delete nodes with an in-degree of 0 until the graph becomes empty. Note that if there is a cycle in the graph, the linear order of the nodes in the cycle cannot be found. Therefore, randomly select the node with the smallest in-degree to delete and break the loop.
  • the final floor plan can show the specific layout of the target building more accurately.
  • a steering function is used for the screening process of the matching degree of the above boundary.
  • a steering function For the specific form of the steering function, please refer to FIG. 9.
  • the left side of Fig. 9 is the boundary shape of the target building, and the right side of Fig. 9 is the steering function of the boundary of the target building.
  • the process of obtaining the adjusted house plan can refer to Fig. 10.
  • the predicted house plan (refer to a in Fig. 10) is obtained, and then the room border and the target Align the boundaries of the building (refer to b in Figure 10), and finally align the borders of adjacent rooms (refer to c in Figure 10).
  • the method for generating a house plan can accept different types and numbers of user constraints, and generate a corresponding predicted house plan.
  • Figure 11 shows a set of example results generated by different boundary inputs and different user constraints. Each row shows the results of different layout constraints applied to the same boundary, and each column shows the results of the same layout constraints applied to different boundaries.
  • the selected constraint is the number required for the three room types: bedrooms, bathrooms, and balconies. The corresponding constraints on the number of rooms are displayed at the bottom of each column.
  • the method for generating house plans provided in this application can generate diversified house plans according to different boundaries and different layout constraints input by users. Compared with the traditional simple search method of generating house plans, the generated house plans The picture quality is higher.
  • a device 1200 for generating a house plan including: an obtaining module 1201, an output module 1202, a screening module 1203, a first obtaining module 1204, and a second obtaining module 1205, wherein :
  • the obtaining module 1201 is used to obtain the boundary of the target building and the layout constraints of the target building, the layout constraints including the room type, the number of rooms, the location of the room, and the adjacency relationship between the rooms;
  • the output module 1202 is configured to provide a plurality of first house plans according to the layout constraints of the target building;
  • the screening module 1203 is configured to filter out a plurality of second house plans from the plurality of first house plans, and the degree of matching between the boundaries of the plurality of second house plans and the boundary of the target building meets a preset condition;
  • the first obtaining module 1204 is used to apply the layout constraints of the second house plan to the boundary of the target building for each second house plan to obtain the layout plan of the target building corresponding to each second house plan ;as well as
  • the second obtaining module 1205 is used to input the layout diagram of the target building and the boundary of the target building into the house plan generation network for the layout plan of each target building to obtain the target output by the house plan generation network A predicted house plan of the building, the predicted house plan including the room frame of each room in the target building and the positional relationship of each room frame in the boundary of the target building.
  • the output module 1202 is specifically configured to retrieve a plurality of first house plans that satisfy the layout constraints of the target building in a pre-stored house plan data set.
  • the screening module 1203 is specifically configured to obtain the steering function of the boundary of the plurality of first house plans and the steering function of the boundary of the target building; The cumulative difference between the steering function and the steering function of the boundary of the target building; the plurality of first house plans corresponding to the cumulative difference less than the preset difference threshold is used as the plurality of second house plans.
  • the first obtaining module 1204 is specifically used to adjust each second house plan until the angle between the front door direction of the second house plan and the front door direction of the target building Less than the preset angle threshold, the adjusted second house plan is obtained; for each adjusted second house plan, the room type, the number of rooms, the room location and the adjacency between the rooms of the adjusted second house plan The relationship is arranged in the boundary of the target building, and the layout drawing of the target building corresponding to each second house plan is obtained.
  • the second obtaining module 1205 is specifically used to input the layout diagram of the target building into the graph neural network for the layout diagram of each target building to obtain the layout diagram of the target building.
  • Room feature vector corresponding to each room input the boundary of the target building into the first convolutional neural network to obtain the boundary feature vector of the target building; associate each room feature vector with the boundary feature vector to obtain multiple Associated feature vector; input each associated feature vector into the first multi-layer perceptron to obtain the initial bounding box corresponding to each room in the layout of the target building; use the initial bounding box corresponding to each room to Map the associated feature vectors corresponding to the room to obtain multiple first feature maps; combine the multiple first feature maps into a second feature map, and input the second feature map into the second convolutional neural network to obtain the target building Multiple raster house plans of the object; input the multiple associated feature vectors, the initial bounding box and the raster house plans into a bounding box optimization network to obtain a predicted house plan of the target building, and the bounding box
  • FIG. 13 there is provided another house plan generating device 1300.
  • the house plan generating device 1300 includes the modules included in the house plan generating device 1200, and optionally, the house plan
  • the generating device 1300 further includes a training module 1206 and an alignment module 1207.
  • the training module 1206 is used to obtain a training data set, the training data set includes multiple house map data; use the training data set to train the initial house map generation network to obtain the trained house map generation network; use the cross-entropy loss function, The regression loss function and the geometric loss function calculate the loss value of the house map generation network after training; adjust the parameters of the house map generation network after training according to the loss value to obtain the house map generation network.
  • the alignment module 1207 is used to align the room border of the target building with the border of the target building; and align the room borders of the target building.
  • each module in the above-mentioned unit diagram generating device can be implemented in whole or in part by software, hardware, and a combination thereof.
  • the above-mentioned modules may be embedded in the form of hardware or independent of the processor in the computer equipment, or may be stored in the memory of the computer equipment in the form of software, so that the processor can call and execute the operations corresponding to the above-mentioned modules.
  • a computer device is provided.
  • the computer device may be a terminal, and its internal structure diagram may be as shown in FIG. 14.
  • the computer equipment includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus.
  • the processor of the computer device is used to provide calculation and control capabilities.
  • the memory of the computer device includes a non-volatile storage medium and an internal memory.
  • the non-volatile storage medium stores an operating system and a computer program.
  • the internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium.
  • the communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, an operator's network, NFC (near field communication) or other technologies.
  • WIFI wireless fidelity
  • NFC near field communication
  • the computer program is executed by the processor, a method for generating a house plan is realized.
  • the display screen of the computer equipment can be a liquid crystal display screen or an electronic ink display screen
  • the input device of the computer equipment can be a touch layer covered on the display screen, or it can be a button, trackball or touch pad set on the housing of the computer equipment , It can also be an external keyboard, touchpad, or mouse.
  • FIG. 14 is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied.
  • the specific computer device may Including more or fewer parts than shown in the figure, or combining some parts, or having a different arrangement of parts.
  • a computer device including a memory and a processor, and a computer program is stored in the memory, and the processor implements the steps in the foregoing method embodiments when the computer program is executed.
  • a computer-readable storage medium on which a computer program is stored, and the computer program is executed by a processor to implement the steps in the foregoing method embodiments.
  • Non-volatile memory may include read-only memory (Read-Only Memory, ROM), magnetic tape, floppy disk, flash memory, or optical storage.
  • Volatile memory may include random access memory (RAM) or external cache memory.
  • RAM may be in various forms, such as static random access memory (Static Random Access Memory, SRAM) or dynamic random access memory (Dynamic Random Access Memory, DRAM), etc.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Geometry (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Software Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Databases & Information Systems (AREA)
  • Medical Informatics (AREA)
  • Computing Systems (AREA)
  • Multimedia (AREA)
  • General Engineering & Computer Science (AREA)
  • Computer Hardware Design (AREA)
  • Biomedical Technology (AREA)
  • Architecture (AREA)
  • Pure & Applied Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Mathematical Optimization (AREA)
  • Molecular Biology (AREA)
  • Mathematical Analysis (AREA)
  • Computational Mathematics (AREA)
  • Computational Linguistics (AREA)
  • Human Computer Interaction (AREA)
  • Mathematical Physics (AREA)
  • Data Mining & Analysis (AREA)
  • Biophysics (AREA)
  • Biodiversity & Conservation Biology (AREA)
  • Civil Engineering (AREA)
  • Structural Engineering (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Processing Or Creating Images (AREA)

Abstract

本申请涉及一种户型图生成方法、装置、计算机设备和存储介质。所述方法包括：获取目标建筑物的边界和所述目标建筑物的布局约束；根据所述目标建筑物的布局约束输出多个第一户型图；从所述多个第一户型图中筛选出多个第二户型图；对于每个所述第二户型图，将所述第二户型图的布局约束应用在所述目标建筑物的边界中，得到每个所述第二户型图对应的所述目标建筑物的布局图；对于每个所述目标建筑物的布局图，将所述目标建筑物的布局图和所述目标建筑物的边界输入户型图生成网络中，得到所述户型图生成网络输出的所述目标建筑物的预测户型图。

Description

户型图生成方法、装置、计算机设备和存储介质
本申请要求于2020年4月3日提交中国专利局，申请号为2020102571436，申请名称为“户型图生成方法、装置、计算机设备和存储介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及计算机技术领域，特别是涉及一种户型图生成方法、装置、计算机设备和存储介质。
背景技术
在日常生活中，经常需要对建筑物进行设计，例如，对建筑物进行户型设计或对建筑物进行布局设计。由于人为设计往往耗时较长，因此出现了自动生成户型图的方法。
相关的自动生成户型图的方法中，需要首先获取建筑物的边界形状，然后在预先存储的大量户型图数据中搜索与建筑物边界形状相符合的户型图，然后输出搜索到的户型图。
然而，相关的自动生成户型图的方法，虽然可以快速生成户型图，但是因为其只是通过在数据集中进行简单的搜索匹配来生成户型图，因此生成的户型图质量不高。
发明内容
基于此，有必要针对上述技术问题，提供一种能够提高质量的户型图生成方法、装置、计算机设备和存储介质。
第一方面，提供一种户型图生成方法，该方法包括：
获取目标建筑物的边界和目标建筑物的布局约束，该布局约束包括房间类型、房间数量、房间位置和房间之间的邻接关系；
根据该目标建筑物的布局约束给出多个第一户型图；
从该多个第一户型图中筛选出多个第二户型图，该多个第二户型图的边界与该目标建筑物的边界之间的匹配度满足预设条件；
对于每个第二户型图，将该第二户型图的布局约束应用在该目标建筑物的边界中，得到每个第二户型图对应的目标建筑物的布局图；以及
对于每个目标建筑物的布局图，将该目标建筑物的布局图和该目标建筑物的边界输入户型图生成网络中，得到该户型图生成网络输出的该目标建筑物的预测户型图，该预测户型图包括该目标建筑物中每个房间的房间边框和每个房间边框在目标建筑物的边界中的位置关系。
在其中一个实施例中，根据该目标建筑物的布局约束给出多个第一户型图，包括：
在预先存储的户型图数据集中检索满足该目标建筑物的布局约束的多个第一户型图。
在其中一个实施例中，从该多个第一户型图中筛选出多个第二户型图，该多个第二户型图的边界与该目标建筑物的边界之间的匹配度满足预设条件，包括：
获取该多个第一户型图的边界的转向函数和该目标建筑物的边界的转向函数；
计算该多个第一户型图的边界的转向函数和该目标建筑物的边界的转向函数之间的累计差值；以及
将该累计差值小于预设差值阈值所对应的该多个第一户型图作为该多个第二户型图。
在其中一个实施例中，对于每个第二户型图，将该多个第二户型图的布局约束应用在该目标建筑物的边界中，得到每个第二户型图对应的目标建筑物的布局图，包括：
对于每个该第二户型图进行调整，直至该第二户型图的前门方向与该目标建筑物的前门方向之间的夹角小于预设角度阈值，得到调整后的第二户型图；以及
对于每个调整后的第二户型图，将该调整后的第二户型图的房间类型、房间数量、房间位置和房间之间的邻接关系布置到该目标建筑物的边界中，得到每个第二户型图对应的目标建筑物的布局图。
在其中一个实施例中，对于每个目标建筑物的布局图，将该目标建筑物的布局图和该目标建筑物的边界输入户型图生成网络中，得到该户型图生成网络输出的目标建筑物的预测户型图，包括：
对于每个目标建筑物的布局图，将该目标建筑物的布局图输入图神经网络，得到该目标建筑物的布局图中每个房间对应的房间特征向量；
将该目标建筑物的边界输入第一卷积神经网络，得到该目标建筑物的边界特征向量；
将每个房间特征向量与该边界特征向量关联起来，得到多个关联特征向量；
将每个该关联特征向量输入第一多层感知器中，得到该目标建筑物的布局图中每个房间对应的初始边界框；
利用每个房间对应的初始边界框对每个房间对应的关联特征向量进行映射，得到多个第一特征图；
将该多个第一特征图组合为第二特征图，将该第二特征图输入第二卷积神经网络，得到该目标建筑物的栅格户型图；以及
将该多个关联特征向量、该初始边界框和该栅格户型图输入边界框优化网络中，得到该目标建筑物的预测户型图，该边界框优化网络包括第三卷积神经网络、兴趣区域池化层和第二多层感知器。
在其中一个实施例中，该方法还包括：
获取训练数据集，该训练数据集包括多个户型图数据；
利用该训练数据集对初始户型图生成网络进行训练，得到训练后的户型图生成网络；
利用交叉熵损失函数、回归损失函数和几何损失函数计算该训练后的户型图生成网络的损失值；以及
根据该损失值调整训练后的户型图生成网络的参数，得到户型图生成网络。
在其中一个实施例中，该方法还包括：
将目标建筑物的房间边框与目标建筑物的边界对齐；以及
将目标建筑物的房间边框之间对齐。
第二方面，提供一种户型图生成装置，包括处理器和存储器，所述存储器中存储有计算机程序，当所述计算机程序被所述处理器执行时，所述处理器：
获取目标建筑物的边界和目标建筑物的布局约束，该布局约束包括房间类型、房间数量、房间位置和房间之间的邻接关系；
根据该目标建筑物的布局约束输出多个第一户型图；
从该多个第一户型图中筛选出多个第二户型图，该多个第二户型图的边界与该目标建筑物的边界之间的匹配度满足预设条件；
对于每个第二户型图，将该第二户型图的布局约束应用在该目标建筑物的边界中，得到每个第二户型图对应的目标建筑物的布局图；以及
对于每个目标建筑物的布局图，将该目标建筑物的布局图和该目标建筑物的边界输入户型图生成 网络中，得到该户型图生成网络输出的该目标建筑物的预测户型图，该预测户型图包括该目标建筑物中每个房间的房间边框和每个房间边框在目标建筑物的边界中的位置关系。
第三方面，提供一种计算机设备，包括存储器和处理器，该存储器存储有计算机程序，该处理器执行该计算机程序时实现以下步骤：
获取目标建筑物的边界和目标建筑物的布局约束，该布局约束包括房间类型、房间数量、房间位置和房间之间的邻接关系；
根据该目标建筑物的布局约束给出多个第一户型图；
从该多个第一户型图中筛选出多个第二户型图，该多个第二户型图的边界与该目标建筑物的边界之间的匹配度满足预设条件；
对于每个第二户型图，将该第二户型图的布局约束应用在该目标建筑物的边界中，得到每个第二户型图对应的目标建筑物的布局图；以及
对于每个目标建筑物的布局图，将该目标建筑物的布局图和该目标建筑物的边界输入户型图生成网络中，得到该户型图生成网络输出的该目标建筑物的预测户型图，该预测户型图包括该目标建筑物中每个房间的房间边框和每个房间边框在目标建筑物的边界中的位置关系。
第四方面，提供一种计算机可读存储介质，其上存储有计算机程序，该计算机程序被处理器执行时实现以下步骤：
获取目标建筑物的边界和目标建筑物的布局约束，该布局约束包括房间类型、房间数量、房间位置和房间之间的邻接关系；
根据该目标建筑物的布局约束给出多个第一户型图；
从该多个第一户型图中筛选出多个第二户型图，该多个第二户型图的边界与该目标建筑物的边界之间的匹配度满足预设条件；
对于每个第二户型图，将该第二户型图的布局约束应用在该目标建筑物的边界中，得到每个第二户型图对应的目标建筑物的布局图；以及
对于每个目标建筑物的布局图，将该目标建筑物的布局图和该目标建筑物的边界输入户型图生成网络中，得到该户型图生成网络输出的该目标建筑物的预测户型图，该预测户型图包括该目标建筑物中每个房间的房间边框和每个房间边框在目标建筑物的边界中的位置关系。
上述户型图生成方法、装置、计算机设备和存储介质，通过获取目标建筑物的边界和目标建筑物的布局约束，该布局约束包括房间类型、房间数量、房间位置和房间之间的邻接关系；然后根据该目标建筑物的布局约束给出多个第一户型图；接着从该多个第一户型图中筛选出多个第二户型图，该多个第二户型图的边界与该目标建筑物的边界之间的匹配度满足预设条件；对于每个该第二户型图，将该第二户型图的布局约束应用在该目标建筑物的边界中，得到每个第二户型图对应的目标建筑物的布局图；最后对于每个目标建筑物的布局图，将该目标建筑物的布局图和该目标建筑物的边界输入户型图生成网络中，得到该户型图生成网络输出的该目标建筑物的预测户型图，该预测户型图包括目标建筑物中每个房间的房间边框和每个房间边框在目标建筑物的边界中的位置关系。因为本申请提供的户型图生成方法在生成户型图的时候不仅考虑到了建筑物的边界，还充分考虑到了建筑物的布局约束，使得最终生成的预测户型图更加符合实际需求，提高了生成的户型图的质量。
附图说明
图1为一个实施例中户型图生成方法的流程示意图；
图2为一个实施例中户型图生成方法中从多个第一户型图中筛选出多个第二户型图的方法的流程示意图；
图3为一个实施例中户型图生成方法中对于每个第二户型图，将第二户型图的布局约束应用在目标建筑物的边界中，得到每个第二户型图对应的目标建筑物的布局图的方法的流程示意图；
图4为另一个实施例中户型图生成方法中对于每个目标建筑物的布局图，将目标建筑物的布局图和目标建筑物的边界输入户型图生成网络中，得到户型图生成网络输出的目标建筑物的预测户型图的方法的流程示意图；
图5为一个实施例中户型图生成网络的具体结构示意图；
图6为一个实施例中边界框优化网络的具体结构示意图；
图7为一个实施例中户型图生成方法中训练户型图生成网络的方法的流程示意图；
图8为一个实施例中户型图生成方法的整体流程示意图；
图9为一个实施例中目标建筑物的边界和转向函数的示意图；
图10为一个实施例中户型图生成方法中对预测户型图进行对齐调整的示意图；
图11为一个实施例中户型图生成方法中一组由不同的边界输入和不同用户约束条件生成的示例结果的示意图；
图12为一个实施例中户型图生成装置的结构框图；
图13为另一个实施例中户型图生成装置的结构框图；
图14为一个实施例中计算机设备的内部结构图。
具体实施方式
为了使本申请的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本申请进行进一步详细说明。应当理解，此处描述的具体实施例仅仅用以解释本申请，并不用于限定本申请。
在本申请实施例中，如图1所示，提供了一种户型图生成方法，以该方法应用于终端为例进行说明，该终端可以但不限于是各种个人计算机、笔记本电脑、智能手机、平板电脑和便携式可穿戴设备，该方法包括以下步骤：
步骤101，终端获取目标建筑物的边界和该目标建筑物的布局约束，该布局约束包括房间类型、房间数量、房间位置和房间之间的邻接关系。
本步骤中提到的目标建筑物的边界指的是目标建筑物的户型轮廓形状，可选的，用户可以手动将目标建筑物的边界输入终端，测量设备也可以对目标建筑物进行测量之后将测量得到的户型轮廓数据发送给终端。在一实施例中，终端具有测量功能，可以直接对目标建筑物进行测量，得到目标建筑物的边界。
目标建筑物的布局约束是用户按照实际需求输入到终端的，可选的，终端上可以展示一个对话框，使用户可以在终端的对话框中输入需要的房间类型，房间数量、房间位置以及房间之间的邻接关系等信息。此外，终端可以将目标建筑物的边界展示给用户，用户可以通过鼠标等外接设备，在目标建筑物的边界中大略绘制出想要的房间类型、房间数量、房间位置和房间之间的邻接关系等。
步骤102，终端根据该目标建筑物的布局约束输出多个第一户型图。
本步骤中，终端在获取到目标建筑物的布局约束后，会给出满足用户要求的布局约束的第一户型图。在一实施例中，同样的布局约束可以存在多种布局形式。为了达到精确性的目的，这些第一户型图的房间数量、房间类型、房间位置和房间之间的邻接关系可以是完全满足用户要求的。但是，在一 些可能的情况下，为了给出尽可能多样化的户型图，输出的第一户型图的布局约束也可以是部分满足用户要求的。例如，用户需要的房间数量是4个，输出的第一户型图的房间数量可以是3个，也可以是5个，第一户型图的布局约束与用户输入的布局约束之间的的匹配程度可以是预先设定的。
步骤103，终端从该多个第一户型图中筛选出多个第二户型图，该多个第二户型图的边界与该目标建筑物的边界之间的匹配度满足预设条件。
第一户型图的布局约束是满足用户需求的，但是第一户型图的边界不一定是符合目标建筑物的边界的。因此，需要对第一户型图进行再一次的筛选，目的是为了得到布局约束和边界都符合要求的户型图。
在一实施例中，本步骤中判断第一户型图的边界与目标建筑物的边界的匹配度的方法可以是，将第一户型图的边界围合的面积和目标建筑物的边界围合的面积进行重叠对比，计算重叠部分的面积与目标建筑物的边界围合的面积的比值，将比值超过预设面积阈值的第一户型图筛选出来，作为第二户型图。在一实施例中，终端将每个第一户型图显示在用户输入的目标建筑物的边界内，以便用户能够直观地分析第一户型图与边界的匹配度，从而选择出第二户型图。
此外，判断第一户型图的边界与目标建筑物的边界的匹配度的方法还可以是，计算第一户型图的边界周长和拐角数量，计算目标建筑物的边界周长和拐角数量，然后将第一户型图的边界周长与目标建筑物的边界周长之间的差值小于预设周长差值阈值，且第一户型图的拐角数量与目标建筑物的拐角数量的差值小于预设数量差值阈值的第一户型图筛选出来，作为第二户型图。
在一实施例中，可以计算第一户型图的边界的转向函数和目标建筑物的边界的转向函数，然后比较两个转向函数，通过比较结果从第一户型图中筛选出第二户型图。
步骤104，对于每个第二户型图，终端将该第二户型图的布局约束应用在该目标建筑物的边界中，得到每个第二户型图对应的目标建筑物的布局图。
在筛选出布局约束和边界都符合用户要求的第二户型图之后，可以直接将第二户型图中的布局应用在目标建筑物的边界中。在同样的布局约束下，户型图仍然存在多种形式，因此，筛选出的第二户型图仍然存在多种布局形式。因为边界相同或相似的户型图之间，布局约束在被直接利用后的适用性就越强。因此，可以直接将第二户型图中的布局约束完整应用在目标建筑物的边界中，得到目标建筑物的多种布局图。
步骤105，对于每个目标建筑物的布局图，终端将该目标建筑物的布局图和该目标建筑物的边界输入户型图生成网络中，得到该户型图生成网络输出的该目标建筑物的预测户型图，该预测户型图包括该目标建筑物中每个房间的房间边框和每个房间边框在目标建筑物的边界中的位置关系。
上述步骤中得到的多个目标建筑物的布局图仅仅是适用于目标建筑物的边界的多种布局，而没有形成具体的目标建筑物的户型图。因此，需要将每个目标建筑物的布局图输入到户型图生成网络中，得到更加准确的户型图。本步骤中得到的多个预测户型图可供用户选择，并指导用户对目标建筑物进行户型规划，本步骤中对于预测户型图的表现形式可以是终端以可视化的形式展示预测户型图，展示的内容包括展示目标建筑物的边界形状以及在该边界形状中的相应位置上展示各个房间的房间边框。
上述户型图生成方法中，通过获取目标建筑物的边界和目标建筑物的布局约束，该布局约束包括房间类型、房间数量、房间位置和房间之间的邻接关系；然后根据该目标建筑物的布局约束给出多个第一户型图；接着从该多个第一户型图中筛选出多个第二户型图，该多个第二户型图的边界与该目标建筑物的边界之间的匹配度满足预设条件；对于每个该第二户型图，将该第二户型图的布局约束应用在该目标建筑物的边界中，得到每个第二户型图对应的目标建筑物的布局图；最后对于每个目标建筑 物的布局图，将该目标建筑物的布局图和该目标建筑物的边界输入户型图生成网络中，得到该户型图生成网络输出的该目标建筑物的预测户型图，该预测户型图包括目标建筑物中每个房间的房间边框和每个房间边框在目标建筑物的边界中的位置关系。因为本申请提供的户型图生成方法在生成户型图的时候不仅考虑到了建筑物的边界，还充分考虑到了建筑物的布局约束，使得最终生成的预测户型图更加符合实际需求，提高了生成的户型图的质量。
在本申请实施例中，根据该目标建筑物的布局约束给出多个第一户型图，包括：
在预先存储的户型图数据集中检索满足该目标建筑物的布局约束的多个第一户型图。
可选的，终端中可以预先存储大量的户型图数据，然后对这些大量的户型图数据进行编码然后存储，进行编码的优点是，在后续步骤中针对用户输入的目标建筑物的布局约束可以快速地检索到对应的第一户型图。
在一实施例中，对户型图进行编码的方式可以是，对户型图中的房间类型、房间数量、房间位置和房间之间的邻接关系进行编码。
现对上述编码做详细说明，首先可以将户型图中的每个房间表示为一个节点，如果某两个房间在户型图中具有邻接关系，则在该房间对应的节点之间添加一条边。将房间类型划分为13种，包括客厅、主卧、次卧、客房、儿童室、书房、餐厅、浴室、厨房、阳台、储物室、衣橱和前厅。可以按照上述列举顺序利用数字对每种房间类型编码为1、2、3、4、5、6、7、8、9、10、11、12和13。然后为了对房间位置进行编码，可以将目标建筑物的边界补充为一个矩形框，得到目标建筑物的边界框，然后将该矩形框划分为K×K的网格，K为大于或等于1的正整数，将户型图中每个房间的中心所在的网格坐标作为房间的位置编码。对于房间之间的邻接关系，首先在户型图中找到所有的室内门，然后将门两侧的房间对作为相邻的房间对。接下来，可以检查任意两个房间之间的距离是否小于相对于房间边界框的给定阈值，相当于判断两个房间之间是否为共用一堵墙的隔壁邻接关系。对于每对相邻的房间，对连接两个房间的边进行编码，编码内容包括可以是左、右、上、下、左上、右上、左下和右下。表示两个房间之间的相对位置关系，由于房间之间的相对位置关系是相对的，例如，以B为基准，A在B的上方，可以将连接A和B的边编码为上；但如果以A为基准的话，B在A的下方，可以将连接A和B的边编码为下。虽然两种编码形式不同，但实质相同，因此，在实际编码时，对于一条边可以同时存储两种相对关系的编码，或者随机选择一种相对关系的编码进行存储。除此之外，还可以对房间的大小进行编码，将房间的面积与整个户型图的面积的比值作为房间大小的编码。
上述编码方法可以利用在用户输入的布局约束上，在对用户输入的布局约束使用同样的编码方式后，可以通过比较简单的编码就可以筛选出符合用户期望的第一户型图，大大提高了终端筛选第一布局图的效率。
在本申请实施例中，请参考图2，提供了一种户型图生成方法中从多个第一户型图中筛选出多个第二户型图的方法，该方法包括：
步骤201，终端获取该多个第一户型图的边界的转向函数和该目标建筑物的边界的转向函数。
转向函数(Turning Function)可以将边界的二维多边形形状转换成一维轮廓线段，达到降维的目的，从而便于进行边界匹配程度的对比。该一维轮廓线段位于归一化的边界周长和累计角度构成的坐标系中。转向函数可以记录边界的二维多边形中每个转折点(转角)的角度累计值。
在一实施例中，可以从户型图的前门开始，沿顺时针方向记录边界的二维多边形中每个转折点(转角)的角度累计值。在一种可能的情况下，即使两个具有完全相同边界的建筑物，不同的前门位置也会导致户型图的显著变化，这意味着在比较两个边界时需要考虑前门的位置，因此选择从前门的位置开 始记录目标建筑物的转向函数和第一户型图的转向函数。
步骤202，终端计算该多个第一户型图的边界的转向函数和该目标建筑物的边界的转向函数之间的累计差值。
在得到目标建筑物的边界的转向函数和第一户型图的转向函数之后，需要计算两个转向函数之间的累计差值。在一实施例中，可以将两个转向函数的差值与坐标轴围成的面积大小作为累计差值。
步骤203，终端将该累计差值小于预设差值阈值所对应的该多个第一户型图作为该多个第二户型图。
目标建筑物的边界的转向函数与第一户型图的边界的转向函数之间的累计差值越小，则说明目标建筑物的边界与第一户型图的边界之间的相似程度越高。本步骤中设置了预设的差值阈值，来筛选出第二户型图。这时筛选出的第二户型图的边界与目标建筑物的边界可能仍然会存在一些误差，但是在可以接受的范围内。
此外，若期望筛选出与目标建筑物的边界完全匹配的第二户型图，可以选择累计差值为零的第一户型图。
在本申请实施例中，通过利用转向函数，将二维的边界比较问题转换成了一维轮廓线段的数学计算问题，大大提高了筛选第二户型图的效率和准确度。
在本申请实施例中，请参考图3，提供了一种户型图生成方法中对于每个第二户型图，将第二户型图的布局约束应用在目标建筑物的边界中，得到每个第二户型图对应的目标建筑物的布局图的方法，该方法包括：
步骤301，终端对于每个该第二户型图进行调整，直至该第二户型图的前门方向与该目标建筑物的前门方向之间的夹角小于预设角度阈值，得到调整后的第二户型图。
本步骤中，在将第二户型图的布局约束应用在目标建筑物的边界中之前，需要对第二户型图进行调整，使得调整后的第二户型图的边界与目标建筑物的边界的重合程度最大，由于边界重合程度的计算是基于从前门出发的转向函数，因此前门作为对齐的参考点。首先对齐两个边界的前门，这也防止前门被任何房间阻挡，判断前门是否对齐的标准是，判断旋转后的第二户型图的前门方向与目标建筑物的前门方向之间的夹角是否小于预定角度，诸如45度等，本申请中提到的前门方向是将建筑物边界框的中心连接到门的中心的向量和将旋转后的第二户型图的边界框的中心连接到前门的向量。
需要注意的是，在旋转第二户型图时，旋转后的第二户型图的底边必须保持水平，而竖边尽可能大多数保持垂直，这样的户型图才是有效的户型图。
步骤302，对于每个调整后的第二户型图，终端将该调整后的第二户型图的房间类型、房间数量、房间位置和房间之间的邻接关系布置到该目标建筑物的边界中，得到每个第二户型图对应的目标建筑物的布局图。
在将第二户型图调整之后，就可以将调整后的第二布局图的布局约束转移到目标建筑物的边界内。转移的方式可以是完全复制，即，将目标建筑物的房间完全按照调整后的第二户型图中的房间类型、房间数量、房间位置和房间之间的邻接关系进行布局。
但是，由于调整后的第二布局图的边界与目标建筑物的边界仍然存在差异，因此，在将调整后的第二户型图的布局约束直接应用在目标建筑物的边界中之后，可能会有一些房间节点被布局在边界之外，这显然不符合实际需求。因此，可以对得到的每个目标建筑物的布局图再次进行调整，保证所有的房间节点都落在目标建筑物的边界内。在一实施例中，对于在边界之外的节点的调整方式可以是，首先建立目标建筑物的边界框和边界框中的网格，然后查询边界外节点所在的网格，其次将边界外节 点移动至最近的边界内的网格中。如果最近的边界内的网格中已经存在节点，那么将网格中已经存在的节点沿相同的方向移动至新的网格中。如果沿相同方向移动到边界中的最后一个网格，但网格中仍然已经存在节点，则在该网格中保留两个节点。上述沿相同方向指的是，与边界外的节点移动至最近的边界内的网格中所采取的方向相同。
上述对于边界外节点的调整方式可以是终端自动完成的，也可以是终端将多个目标建筑物的布局图展示给用户，用户自行编辑完成的。
本申请实施例中，通过对第二户型图进行调整，将调整后的第二户型图的布局约束应用在目标建筑物的边界内，并且对于处于边界外的节点进行了进一步的调整，使得最终得到的目标建筑物的布局图更加准确，更加符合实际需求。
在本申请实施例中，请参考图4，提供了一种户型图生成方法中对于每个目标建筑物的布局图，将目标建筑物的布局图和目标建筑物的边界输入户型图生成网络中，得到户型图生成网络输出的目标建筑物的预测户型图的方法，该方法包括：
步骤401，对于每个目标建筑物的布局图，终端将该目标建筑物的布局图输入图神经网络，得到该目标建筑物的布局图中每个房间对应的房间特征向量。
步骤402，终端将该目标建筑物的边界输入第一卷积神经网络，得到该目标建筑物的边界特征向量。
步骤403，终端将每个房间特征向量与该边界特征向量关联起来，得到多个关联特征向量。
步骤404，终端将每个该关联特征向量输入第一多层感知器(MLP，Multilayer Perceptron)中，得到该目标建筑物的布局图中每个房间对应的初始边界框。
步骤405，终端利用每个房间对应的初始边界框对每个房间对应的关联特征向量进行映射，得到多个第一特征图。
步骤406，终端将该多个第一特征图组合为第二特征图，然后终端将该第二特征图输入第二卷积神经网络，得到该目标建筑物的栅格户型图。
步骤407，终端将该多个关联特征向量、该初始边界框和该栅格户型图输入边界框优化网络中，得到该目标建筑物的预测户型图，该边界框优化网络包括第三卷积神经网络、兴趣区域池化层和第二多层感知器。
现结合图5的户型图生成网络的具体结构对上述步骤401到步骤407进行说明，图5中的目标建筑物的边界和布局约束分别被输入到不同的神经网络中，得到了每个房间对应的关联特征向量；利用多层感知器可以为每个房间预测出房间在边界中的初始边界框(图5中用
Figure PCTCN2020111335-appb-000001
表示，其中i为0到n之间的任意正整数)；利用每个房间对应的初始边界框对每个房间对应的关联特征向量进行映射，得到多个第一特征图；将得到的多个第一特征图组合起来得到第二特征图；然后终端将该第二特征图输入到第二卷积神经网络，得到栅格户型图。在得到栅格户型图之后，还可以使用一个边界框优化网络对之前得到的初始边界框进行优化，得到更加准确的优化边界框(图5中用
Figure PCTCN2020111335-appb-000002
表示，其中i为0到n之间的任意正整数)。整个过程中，可以得到户型图生成网络的输出包括各个房间对应的初始边界框、栅格户型图的图像和各个房间对应的优化边界框，该优化边界框可以被组合起来作为目标建筑物的预测户型图。
上述边界框优化网络的具体结构可参考图6，如图6所示，边界框优化网络包括第三卷积神经网络、兴趣区域池化层和第二多层感知器。首先栅格户型图经过第三卷积神经网络得到一个第三特征图，其次每个房间对应的初始边界框和该第三特征图一起被输入兴趣区域池化层得到一个特定长度的特征向量，然后将该特定长度的特征向量和每个房间对应的关联特征向量进行拼接，输入到第二多层感知器中得到每个房间对应的优化边界框。进一步的，可以根据各个房间的优化边界框推测出每个房间的房间边框在目标建筑物的边界中的可视化位置关系。
在本申请实施例中，通过利用多种网络组合的方式形成户型图生成网络，从而生成预测户型图，可以使最终得到的预测户型图更加准确，质量更高。
在本申请实施例中，请参考图7，提供了一种户型图生成方法中训练户型图生成网络的方法，该方法包括：
步骤701，终端获取训练数据集，该训练数据集包括多个户型图数据。
本步骤中，可以从网上的海量资源中获取大量的户型图，在获取到大量的户型图之后，可以对获取到的大量的户型图进行预处理，预处理可以包括将彩色的户型图统一转换为灰度图、将所有户型图的分辨率调整为相同的值、将所有户型图调整为相同大小或对所有的户型图采用上述内容中记载的编码方式进行编码等操作。在对采集到的户型图进行预处理之后，可以将与处理后的户型图进行存储。在对户型图生成网络进行训练的过程，实际上是针对图5中包含的所有网络进行了训练。
步骤702，终端利用该训练数据集对初始户型图生成网络进行训练，得到训练后的户型图生成网络。
步骤703，终端利用交叉熵损失函数、回归损失函数和几何损失函数计算该训练后的户型图生成网络的损失值。
步骤704，终端根据该损失值调整训练后的户型图生成网络的参数，得到户型图生成网络。
为了保证训练后的户型图生成网络具有良好的性能，本步骤中设计了多个类型的损失函数来计算户型图生成网络的损失值。
损失值的计算方式如下面的公式所示：
Figure PCTCN2020111335-appb-000003
其中，L pix(I)是预测户型图和真实户型图的交叉熵的图像损失，
Figure PCTCN2020111335-appb-000004
是回归损失，
Figure PCTCN2020111335-appb-000005
是几何损失。几何损失只适用于初始边界框，初始边界框
Figure PCTCN2020111335-appb-000006
的几何损失定义为：
Figure PCTCN2020111335-appb-000007
L coverage和L interior都约束边界和房间边框之间的空间一致性，L mutex约束任意两个房间边框之间的空间一致性，
Figure PCTCN2020111335-appb-000008
确保预测框与真实框匹配。前三个项确保房间边框对建筑物内部适当覆盖。最后一项，该项将预测框与真实框进行比较，从而确保在训练期间房间边框的位置和尺寸的预测也得到改进。
在给出有关几何损失项的更多详细信息之前，首先定义两个距离函数d in(p,B)和d out(p,B)，d in(p,B)来测量点p距离边界框B内部的距离，d out(p,B)来测量点p距离边界框B外部的距离，点p是任意点。d in(p,B)和d out(p,B)如下所示：
Figure PCTCN2020111335-appb-000009
Figure PCTCN2020111335-appb-000010
其中，Ω in(B)集和Ω bd(B)集分别表示边界框B的内部像素点和外边界像素点。
覆盖损失：目标建筑物的边界应该被所有房间边框的组合完全覆盖。在一实施例中，任何点p∈Ω in(B)应该被至少一个房间边框覆盖。因此，覆盖损失定义如下：
Figure PCTCN2020111335-appb-000011
其中，|Ω in(B)|为在Ω in(B)中像素的数量。
内部损失：每个房间的边框应该坐落在目标建筑物边界框
Figure PCTCN2020111335-appb-000012
内。因此，内部损失可以定义为：
Figure PCTCN2020111335-appb-000013
互斥量损失：房间边框之间的重叠部分应该尽可能小，这样房间才能紧凑地分布在目标建筑物内部。因此，互斥量损失可以定义为：
Figure PCTCN2020111335-appb-000014
匹配损失：每个房间的边框B i应该覆盖与对应的真实边界框
Figure PCTCN2020111335-appb-000015
覆盖的相同的区域，也就是说，B i应该被显示在
Figure PCTCN2020111335-appb-000016
的内部。因此，可以将匹配损失定义为：
Figure PCTCN2020111335-appb-000017
在本申请实施例中，通过设置多样化的损失函数，考虑到多种类型的损失。使得最终训练得到的 户型图生成网络具有更好的预测性能。
在本申请实施例中，在得到户型图生成网络输出的该目标建筑物的预测户型图之后，该方法还包括：
将目标建筑物的房间边框与目标建筑物的边界对齐；以及
将目标建筑物的房间边框之间对齐。
户型图生成网络的最终输出是一个预测户型图的图像和每个房间的房间边框。房间边框可能出现的一个问题是它们可能没有很好地对齐，而且某些房间边框之间可能在某些区域重叠。因此，在最后的对齐调整步骤中，使用栅格户型图的图像来确定重叠区域的房间类型分配。
首先将房间边框与目标建筑物的边界对齐，然后将相邻的房间边框彼此对齐。更具体地说，对于一个房间边框的每条边，找到最近的具有相同方向的边界边，即水平或垂直，如果它们的距离小于给定距离阈值，使房间边框的边与边界边对齐。此外，根据布局图中已编码的空间关系对相邻的房间对进行对齐。例如，如果房间边框A在房间边框B的左侧，将房间边框A的右边缘与房间边框B的左边缘对齐。此外，如果两个房间边框的相邻边的长度小于给定距离阈值，则将两个房间边框的相邻边进行对齐，因为在户型图中并排放置的房间边框最好具有对齐的墙以最大程度地减少转角的数量。一个房间边框可能与不同的房间边框相邻。因此，必须多次更新边。为了避免破坏之前的优化对齐，可以设置一个标志来指示框的边是否已经更新。如果任何一条边已经更新了，则它是固定的，将另一条边与固定边对齐。如果两条边都不固定，将它们更新到它们的平均位置。
此外，需要为房间边框之间的重叠区域确定所属的房间类型。为了实现这一点，对于每一对房间，检查它们是否重叠，并使用生成的栅格户型图的图像来确定它们的相对顺序。更详细地说，对于每个房间对，计算每个房间类型在重叠区域内的像素数量，并决定首先绘制像素数量较少的房间，这样，当两个房间边框之间存在重叠区域的时候，后绘制的房间边框会覆盖在先绘制的房间边框之上，使得重叠区域被归属在后绘制的房间边框内。按照此过程，如果房间1和房间2重叠并且应在房间2之前绘制房间1，则通过添加一个节点表示每个房间以及从房间1到房间2的一个有向边来构建图。然后，目标是找到所有图节点满足有向边施加的排序约束的顺序。为了找到这样的一个排序，首先随机选择任意一个入度为0的节点，然后删除这个节点以及节点指向图其余部分的所有边。这里，入度是指向节点的有向边的数量。继续删除入度为0的节点，直到图变为空。注意，如果图中有一个循环，就无法找到循环中节点的线性顺序。因此，随机选择具有最小入度数的节点来删除和打破循环。
本申请实施例中，通过优化对齐和确定重叠部分的标签，使最终得到的户型图可以更加精确地展示目标建筑物的具体布局。
现针对上述各实施例提供一具体实施例，对本申请的户型图生成方法进行整体流程的说明。
首先，获取用户输入的目标建筑物的边界和目标建筑物的布局约束(可参考图8中a)，接着在预先存储的户型图数据集中检索，满足用户输入的布局约束的第一户型图并展示第一户型图(可参考图8中b)，然后对多个第一户型图进行筛选，包括边界的匹配程度的筛选和旋转调整使前门方向与输入便捷的前门方向一致等，得到第二户型图，接着可以直接将第二户型图的布局约束应用在目标建筑物的边界上并将节点调整到边界内，得到多个目标建筑物的布局图(可参考图8中c)，接着将多个目标建筑物的布局图输入户型图生成网络中，得到户型图生成网络输出的预测户型图和房间边框(可参考图8中d)，最后对预测户型图进行对齐调整，得到调整后的户型图(可参考图8中e)。
对于上述边界的匹配程度的筛选过程利用了转向函数，转向函数的具体形式可参考图9。图9的左边是目标建筑物的边界形状，图9的右边是目标建筑物的边界的转向函数。
此外，对于上述步骤中的对预测户型图进行对齐调整，得到调整后的户型图的过程可参考图10，首先得到了预测户型图(可参考图10中的a)，接着将房间边框与目标建筑物的边界对齐(可参考图10中的b)，最后将相邻的房间边框之间对齐(可参考图10中的c)。
在本申请实施例中，本申请提供的户型图生成方法可以接受不同类型和数量的用户约束，并生成对应的预测户型图。图11显示了一组由不同的边界输入和不同用户约束条件生成的示例结果。每一行显示应用于同一边界的不同布局约束的结果，而每一列显示应用于不同边界的相同布局约束的结果。所选的约束是三种房间类型所需的数量：卧室、浴室和阳台。房间数量的相应约束显示在每一栏的底部。
通过检查图11中的每一行，可以看到生成的预测户型图如何满足给定的房间数量约束并适应边界。不同数量的卧室、浴室和阳台是根据布局约束条件生成的，这些房间的位置会发生变化，以使户型图最符合输入边界。阳台通常在目标建筑物边界上有两个或三个面，反映了公寓中典型的阳台设计。因此，它们的位置取决于输入边界。所有的户型图也都有一个客厅，因为它存在于像这样有复杂边界的建筑物中，并且可以有额外的房间。
从图11中的每一列的结果中，看到相同数量的相同类型房间如何在具有不同边界的建筑物内以不同的方式分布。例如，第三列中的两个浴室有时相邻，有时不相邻，但总是与卧室相邻。在第五列中，阳台从不相邻，通常出现在建筑物的不同位置，显示出不同的户型图。
由上述内容可知，本申请提供的户型图生成方法，可以根据用户输入的不同边界和不同布局约束生成多样化的户型图，相比于传统的简单搜索生成户型图的方法而言，生成的户型图质量更高。
应该理解的是，虽然图1至图11的流程图中的各个步骤按照箭头的指示依次显示，但是这些步骤并不是必然按照箭头指示的顺序依次执行。除非本文中有明确的说明，这些步骤的执行并没有严格的顺序限制，这些步骤可以以其它的顺序执行。而且，图1至图11中的至少一部分步骤可以包括多个步骤或者多个阶段，这些步骤或者阶段并不必然是在同一时刻执行完成，而是可以在不同的时刻执行，这些步骤或者阶段的执行顺序也不必然是依次进行，而是可以与其它步骤或者其它步骤中的步骤或者阶段的至少一部分轮流或者交替地执行。
在本申请实施例中，如图12所示，提供了一种户型图生成装置1200，包括：获取模块1201、输出模块1202、筛选模块1203、第一得到模块1204和第二得到模块1205，其中：
获取模块1201，用于获取目标建筑物的边界和目标建筑物的布局约束，该布局约束包括房间类型、房间数量、房间位置和房间之间的邻接关系；
输出模块1202，用于根据该目标建筑物的布局约束给出多个第一户型图；
筛选模块1203，用于从该多个第一户型图中筛选出多个第二户型图，该多个第二户型图的边界与该目标建筑物的边界之间的匹配度满足预设条件；
第一得到模块1204，用于对于每个第二户型图，将该第二户型图的布局约束应用在该目标建筑物的边界中，得到每个第二户型图对应的目标建筑物的布局图；以及
第二得到模块1205，用于对于每个目标建筑物的布局图，将该目标建筑物的布局图和该目标建筑物的边界输入户型图生成网络中，得到该户型图生成网络输出的该目标建筑物的预测户型图，该预测户型图包括该目标建筑物中每个房间的房间边框和每个房间边框在目标建筑物的边界中的位置关系。
在本申请实施例中，该输出模块1202具体用于，在预先存储的户型图数据集中检索满足该目标建筑物的布局约束的多个第一户型图。
在本申请实施例中，该筛选模块1203具体用于，获取该多个第一户型图的边界的转向函数和该目标建筑物的边界的转向函数；计算该多个第一户型图的边界的转向函数和该目标建筑物的边界的转向函数之间的累计差值；将该累计差值小于预设差值阈值所对应的该多个第一户型图作为该多个第二户型图。
在本申请实施例中，该第一得到模块1204具体用于，对于每个该第二户型图进行调整，直至该第二户型图的前门方向与该目标建筑物的前门方向之间的夹角小于预设角度阈值，得到调整后的第二户型图；对于每个调整后的第二户型图，将该调整后的第二户型图的房间类型、房间数量、房间位置和房间之间的邻接关系布置到该目标建筑物的边界中，得到每个第二户型图对应的目标建筑物的布局图。
在本申请实施例中，该第二得到模块1205具体用于，对于每个目标建筑物的布局图，将该目标建筑物的布局图输入图神经网络，得到该目标建筑物的布局图中每个房间对应的房间特征向量；将该目标建筑物的边界输入第一卷积神经网络，得到该目标建筑物的边界特征向量；将每个房间特征向量与该边界特征向量关联起来，得到多个关联特征向量；将每个该关联特征向量输入第一多层感知器中，得到该目标建筑物的布局图中每个房间对应的初始边界框；利用每个房间对应的初始边界框对每个房间对应的关联特征向量进行映射，得到多个第一特征图；将该多个第一特征图组合为第二特征图，将该第二特征图输入第二卷积神经网络，得到该目标建筑物的多个栅格户型图；将该多个关联特征向量、该初始边界框和该栅格户型图输入边界框优化网络中，得到该目标建筑物的预测户型图，该边界框优化网络包括第三卷积神经网络、兴趣区域池化层和第二多层感知器。
在本申请实施例中，请参考图13，提供了另一种户型图生成装置1300，该户型图生成装置1300除了包含该户型图生成装置1200包含的各模块外，可选的，该户型图生成装置1300还包括训练模块1206和对齐模块1207。
该训练模块1206用于获取训练数据集，该训练数据集包括多个户型图数据；利用该训练数据集对初始户型图生成网络进行训练，得到训练后户型图生成网络；利用交叉熵损失函数、回归损失函数和几何损失函数计算该训练后户型图生成网络的损失值；根据该损失值调整训练后户型图生成网络的参数，得到户型图生成网络。
在本申请实施例中，该对齐模块1207用于将目标建筑物的房间边框与目标建筑物的边界对齐；将目标建筑物的房间边框之间对齐。
关于户型图生成装置的具体限定可以参见上文中对于户型图生成方法的限定，在此不再赘述。上述户型图生成装置中的各个模块可全部或部分通过软件、硬件及其组合来实现。上述各模块可以硬件形式内嵌于或独立于计算机设备中的处理器中，也可以以软件形式存储于计算机设备中的存储器中，以便于处理器调用执行以上各个模块对应的操作。
在本申请实施例中，提供了一种计算机设备，该计算机设备可以是终端，其内部结构图可以如图14所示。该计算机设备包括通过系统总线连接的处理器、存储器、通信接口、显示屏和输入装置。其中，该计算机设备的处理器用于提供计算和控制能力。该计算机设备的存储器包括非易失性存储介质和内存储器。该非易失性存储介质存储有操作系统和计算机程序。该内存储器为非易失性存储介质中的操作系统和计算机程序的运行提供环境。该计算机设备的通信接口用于与外部的终端进行有线或无线方式的通信，无线方式可通过WIFI、运营商网络、NFC(近场通信)或其他技术实现。该计算机程序被处理器执行时以实现一种户型图生成方法。该计算机设备的显示屏可以是液晶显示屏或者电子墨水显示屏，该计算机设备的输入装置可以是显示屏上覆盖的触摸层，也可以是计算机设备外壳上设置的按键、轨迹球或触控板，还可以是外接的键盘、触控板或鼠标等。
本领域技术人员可以理解，图14中示出的结构，仅仅是与本申请方案相关的部分结构的框图，并不构成对本申请方案所应用于其上的计算机设备的限定，具体的计算机设备可以包括比图中所示更多或更少的部件，或者组合某些部件，或者具有不同的部件布置。
在本申请实施例中，提供了一种计算机设备，包括存储器和处理器，存储器中存储有计算机程序，该处理器执行计算机程序时实现上述各方法实施例中的步骤。
在本申请实施例中，提供了一种计算机可读存储介质，其上存储有计算机程序，该计算机程序被处理器执行时实现上述各方法实施例中的步骤。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过计算机程序来指令相关的硬件来完成，所述的计算机程序可存储于一非易失性计算机可读取存储介质中，该计算机程序在执行时，可包括如上述各方法的实施例的流程。其中，本申请所提供的各实施例中所使用的对存储器、存储、数据库或其它介质的任何引用，均可包括非易失性和易失性存储器中的至少一种。非易失性存储器可包括只读存储器(Read-Only Memory，ROM)、磁带、软盘、闪存或光存储器等。易失性存储器可包括随机存取存储器(Random Access Memory，RAM)或外部高速缓冲存储器。作为说明而非局限，RAM可以是多种形式，比如静态随机存取存储器(Static Random Access Memory，SRAM)或动态随机存取存储器(Dynamic Random Access Memory，DRAM)等。
以上实施例的各技术特征可以进行任意的组合，为使描述简洁，未对上述实施例中的各个技术特征所有可能的组合都进行描述，然而，只要这些技术特征的组合不存在矛盾，都应当认为是本说明书记载的范围。
以上所述实施例仅表达了本申请的几种实施方式，其描述较为具体和详细，但并不能因此而理解为对发明专利范围的限制。应当指出的是，对于本领域的普通技术人员来说，在不脱离本申请构思的前提下，还可以做出若干变形和改进，这些都属于本申请的保护范围。因此，本申请专利的保护范围应以所附权利要求为准。

Claims (15)

  1. 一种户型图生成方法，包括：
    获取目标建筑物的边界和所述目标建筑物的布局约束，所述布局约束包括房间类型、房间数量、房间位置和房间之间的邻接关系；
    根据所述目标建筑物的布局约束输出多个第一户型图；
    从所述多个第一户型图中筛选出多个第二户型图，所述多个第二户型图的边界与所述目标建筑物的边界之间的匹配度满足预设条件；
    对于每个所述第二户型图，将所述第二户型图的布局约束应用在所述目标建筑物的边界中，得到每个所述第二户型图对应的所述目标建筑物的布局图；以及
    对于每个所述目标建筑物的布局图，将所述目标建筑物的布局图和所述目标建筑物的边界输入户型图生成网络中，得到所述户型图生成网络输出的所述目标建筑物的预测户型图，所述预测户型图包括所述目标建筑物中每个房间的房间边框和每个所述房间边框在所述目标建筑物的边界中的位置关系。
  2. 根据权利要求1所述的方法，其中，所述根据所述目标建筑物的布局约束输出多个第一户型图，包括：
    在预先存储的户型图数据集中检索满足所述目标建筑物的布局约束的所述多个第一户型图。
  3. 根据权利要求1所述的方法，其中，所述从所述多个第一户型图中筛选出多个第二户型图，所述多个第二户型图的边界与所述目标建筑物的边界之间的匹配度满足预设条件，包括：
    获取所述多个第一户型图的边界的转向函数和所述目标建筑物的边界的转向函数；
    计算所述多个第一户型图的边界的转向函数和所述目标建筑物的边界的转向函数之间的累计差值；以及
    将所述累计差值小于预设差值阈值所对应的所述多个第一户型图作为所述多个第二户型图。
  4. 根据权利要求1所述的方法，其中，所述对于每个所述第二户型图，将所述第二户型图的布局约束应用在所述目标建筑物的边界中，得到每个所述第二户型图对应的所述目标建筑物的布局图，包括：
    对于每个所述第二户型图进行调整，直至所述第二户型图的前门方向与所述目标建筑物的前门方向之间的夹角小于预设角度阈值，得到调整后的第二户型图；以及
    对于每个所述调整后的第二户型图，将所述调整后的第二户型图的房间类型、房间数量、房间位置和房间之间的邻接关系布置到所述目标建筑物的边界中，得到每个所述第二户型图对应的目标建筑物的布局图。
  5. 根据权利要求1所述的方法，其中，所述对于每个所述目标建筑物的布局图，将所述目标建筑物的布局图和所述目标建筑物的边界输入户型图生成网络中，得到所述户型图生成网络输出的所述目标建筑物的预测户型图，包括：
    对于每个所述目标建筑物的布局图，将所述目标建筑物的布局图输入图神经网络，得到所述目标建筑物的布局图中每个房间对应的房间特征向量；
    将所述目标建筑物的边界输入第一卷积神经网络，得到所述目标建筑物的边界特征向量；
    将每个所述房间特征向量与所述边界特征向量关联起来，得到多个关联特征向量；
    将每个所述关联特征向量输入第一多层感知器中，得到所述目标建筑物的布局图中每个房间对应 的初始边界框；
    利用所述每个房间对应的初始边界框对所述每个房间对应的关联特征向量进行映射，得到多个第一特征图；
    将所述多个第一特征图组合为第二特征图，将所述第二特征图输入第二卷积神经网络，得到所述目标建筑物的栅格户型图；以及
    将所述多个关联特征向量、所述初始边界框和所述栅格户型图输入边界框优化网络中，得到所述目标建筑物的预测户型图，所述边界框优化网络包括第三卷积神经网络、兴趣区域池化层和第二多层感知器。
  6. 根据权利要求1所述的方法，还包括：
    获取训练数据集，所述训练数据集包括多个户型图数据；
    利用所述训练数据集对初始户型图生成网络进行训练，得到训练后的户型图生成网络；
    利用交叉熵损失函数、回归损失函数和几何损失函数计算所述训练后的户型图生成网络的损失值；以及
    根据所述损失值调整所述训练后的户型图生成网络的参数，得到所述户型图生成网络。
  7. 根据权利要求1所述的方法，还包括：
    将所述目标建筑物的房间边框与所述目标建筑物的边界对齐；以及
    将所述目标建筑物的房间边框之间对齐。
  8. 根据权利要求1所述的方法，其中，所述获取目标建筑物的边界和所述目标建筑物的布局约束，包括：
    对所述目标建筑物进行测量，以得到所述目标建筑物的边界；以及
    接收输入的所述目标建筑物的布局约束。
  9. 根据权利要求1所述的方法，其中，所述从所述多个第一户型图中筛选出多个第二户型图，所述多个第二户型图的边界与所述目标建筑物的边界之间的匹配度满足预设条件，包括：
    将所述第一户型图的边界围合的面积和所述目标建筑物的边界围合的面积进行重叠对比；以及
    计算重叠部分的面积与所述目标建筑物的边界围合的面积的比值；以及
    将比值超过预设面积阈值所对应的第一户型图筛选出来，作为所述第二户型图。
  10. 根据权利要求1所述的方法，其中，所述从所述多个第一户型图中筛选出多个第二户型图，所述多个第二户型图的边界与所述目标建筑物的边界之间的匹配度满足预设条件，包括：
    计算所述多个第一户型图的边界周长和拐角数量，以及所述目标建筑物的边界周长和拐角数量；以及
    将所述多个第一户型图的边界周长与所述目标建筑物的边界周长之间的差值小于预设周长差值阈值，且所述多个第一户型图的拐角数量与所述目标建筑物的拐角数量的差值小于预设数量差值阈值的第一户型图筛选出来，作为第二户型图。
  11. 根据权利要求2所述的方法，其中，在所述在预先存储的户型图数据集中检索满足所述目标建筑物的布局约束的所述多个第一户型图之前，所述根据所述目标建筑物的布局约束输出多个第一户型图，还包括：
    获取户型图数据；以及
    对所述户型图数据进行编码并存储在所述户型图数据集中。
  12. 根据权利要求3所述的方法，其中，所述累计差值为所述多个第一户型图的边界的转向函数 和所述目标建筑物的边界的转向函数之间的差值与坐标轴围成的面积大小。
  13. 一种户型图生成装置，包括处理器和存储器，所述存储器中存储有计算机程序，当所述计算机程序被所述处理器执行时，所述处理器执行：
    获取目标建筑物的边界和所述目标建筑物的布局约束，所述布局约束包括房间类型、房间数量、房间位置和房间之间的邻接关系；
    根据所述目标建筑物的布局约束输出多个第一户型图；
    从所述多个第一户型图中筛选出多个第二户型图，所述多个第二户型图的边界与所述目标建筑物的边界之间的匹配度满足预设条件；
    对于每个所述第二户型图，将所述第二户型图的布局约束应用在所述目标建筑物的边界中，得到每个所述第二户型图对应的所述目标建筑物的布局图；以及
    对于每个所述目标建筑物的布局图，将所述目标建筑物的布局图和所述目标建筑物的边界输入户型图生成网络中，得到所述户型图生成网络输出的所述目标建筑物的预测户型图，所述预测户型图包括所述目标建筑物中每个房间的房间边框和每个所述房间边框在所述目标建筑物的边界中的位置关系。
  14. 一种计算机设备，包括存储器和处理器，所述存储器存储有计算机程序，其特征在于，所述处理器执行所述计算机程序时实现权利要求1所述的方法的步骤。
  15. 一种计算机可读存储介质，其上存储有计算机程序，其特征在于，所述计算机程序被处理器执行时实现权利要求1所述的方法的步骤。
PCT/CN2020/111335 2020-04-03 2020-08-26 户型图生成方法、装置、计算机设备和存储介质 WO2021196515A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US17/885,317 US20220383572A1 (en) 2020-04-03 2022-08-10 Floor plan image generating method and device, computer device and storage medium

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010257143.6 2020-04-03
CN202010257143.6A CN111598972B (zh) 2020-04-03 2020-04-03 户型图生成方法、装置、计算机设备和存储介质

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US17/885,317 Continuation US20220383572A1 (en) 2020-04-03 2022-08-10 Floor plan image generating method and device, computer device and storage medium

Publications (1)

Publication Number Publication Date
WO2021196515A1 true WO2021196515A1 (zh) 2021-10-07

Family

ID=72187368

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/111335 WO2021196515A1 (zh) 2020-04-03 2020-08-26 户型图生成方法、装置、计算机设备和存储介质

Country Status (3)

Country Link
US (1) US20220383572A1 (zh)
CN (1) CN111598972B (zh)
WO (1) WO2021196515A1 (zh)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114169063A (zh) * 2021-12-27 2022-03-11 北京奥科博思创新科技有限公司 一种建筑平面布局方法、装置及电子设备
CN114187588A (zh) * 2021-12-08 2022-03-15 贝壳技术有限公司 数据处理的方法、装置、存储介质和计算机程序产品
CN114219007A (zh) * 2021-11-23 2022-03-22 慧之安信息技术股份有限公司 基于图像匹配的户型合理度评判方法及装置
CN114972579A (zh) * 2022-06-22 2022-08-30 北京城市网邻信息技术有限公司 户型图构建方法、装置、设备及存储介质
CN115114713A (zh) * 2022-07-26 2022-09-27 江西少科智能建造科技有限公司 一种自动风盘布置方法、系统、存储介质及设备
CN115205418A (zh) * 2022-09-15 2022-10-18 武汉智筑完美家居科技有限公司 户型图重构方法、装置及电子设备、存储介质
EP4293562A1 (en) * 2022-06-16 2023-12-20 MFTB Holdco, Inc. Automated tools for assessing building mapping information generation

Families Citing this family (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20210358202A1 (en) * 2020-05-13 2021-11-18 Electronic Caregiver, Inc. Room Labeling Drawing Interface for Activity Tracking and Detection
CN112464311A (zh) * 2020-11-24 2021-03-09 中天众达智慧城市科技有限公司 一种智慧城市数据中心的设备布置方法
CN112732380A (zh) * 2020-12-31 2021-04-30 珠海格力电器股份有限公司 智能家居的可视化展示方法和装置、存储介质、电子装置
CN114764329A (zh) * 2021-01-14 2022-07-19 华为技术有限公司 一种ui界面自适应约束求解方法及相关装置
US20220327254A1 (en) * 2021-04-09 2022-10-13 Verizon Patent And Licensing Inc. Floor layout generation and unit location determination
CN113345044B (zh) * 2021-04-22 2022-06-10 贝壳找房(北京)科技有限公司 户型图生成方法及装置
CN113434945A (zh) * 2021-06-30 2021-09-24 杭州群核信息技术有限公司 户型方案的生成方法及装置
CN113656877B (zh) * 2021-08-23 2024-04-16 深圳须弥云图空间科技有限公司 多层户型的模型生成方法、装置、介质及电子设备
CN113642090B (zh) * 2021-10-14 2022-04-01 深圳润世华软件和信息技术服务有限公司 建筑布局生成方法、装置、计算机设备和存储介质
CN113920250B (zh) * 2021-10-21 2023-05-23 广东三维家信息科技有限公司 一种户型编码匹配方法及装置
CN114065342A (zh) * 2021-11-08 2022-02-18 深圳须弥云图空间科技有限公司 户型轮廓的生成方法、装置、存储介质及电子设备
CN114241129A (zh) * 2021-12-10 2022-03-25 广东三维家信息科技有限公司 一种室内场景布局方法、装置、电子设备及存储介质
CN114282291A (zh) * 2021-12-13 2022-04-05 郑州智源堂房地产咨询有限公司 一种房屋功能结构布局设计方法
CN113919043B (zh) * 2021-12-14 2022-04-29 深圳小库科技有限公司 根据给定范围自动生成户型布局的方法

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103268621A (zh) * 2013-05-16 2013-08-28 北京链家房地产经纪有限公司 一种房屋实景图生成方法与装置
US20150279000A1 (en) * 2014-03-27 2015-10-01 Panasonic Intellectual Property Corporation Of America Display control method, non-temporary recording medium storing display control program, and information processing terminal
CN110633553A (zh) * 2019-10-09 2019-12-31 郑豪 一种住宅户型平面图自动生成方法和系统

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106295678B (zh) * 2016-07-27 2020-03-06 北京旷视科技有限公司 神经网络训练与构建方法和装置以及目标检测方法和装置
CN106484959B (zh) * 2016-09-19 2019-12-31 上海斐讯数据通信技术有限公司 一种户型图绘制方法及绘制设备
US10460064B1 (en) * 2017-07-13 2019-10-29 Cadence Design Systems, Inc. Partition-aware grid graph based hierarchical global routing
CN107506391B (zh) * 2017-07-28 2021-07-09 上海网罗电子科技有限公司 一种提取cad图纸信息至室内地图的方法
CN109992693A (zh) * 2019-04-17 2019-07-09 广东三维家信息科技有限公司 户型匹配方法及装置

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103268621A (zh) * 2013-05-16 2013-08-28 北京链家房地产经纪有限公司 一种房屋实景图生成方法与装置
US20150279000A1 (en) * 2014-03-27 2015-10-01 Panasonic Intellectual Property Corporation Of America Display control method, non-temporary recording medium storing display control program, and information processing terminal
CN110633553A (zh) * 2019-10-09 2019-12-31 郑豪 一种住宅户型平面图自动生成方法和系统

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
RUIZHEN HU; ZEYU HUANG; YUHAN TANG; OLIVER VAN KAICK; HAO ZHANG; HUI HUANG: "Graph2Plan: Learning Floorplan Generation from Layout Graphs", ARXIV.ORG, CORNELL UNIVERSITY LIBRARY, 201 OLIN LIBRARY CORNELL UNIVERSITY ITHACA, NY 14853, 28 April 2020 (2020-04-28), 201 Olin Library Cornell University Ithaca, NY 14853, XP081653991, DOI: 10.1145/3386569.3392391 *
WU WENMING, FAN LUBIN, LIU LIGANG, WONKA PETER: "MIQP-based Layout Design for Building Interiors", COMPUTER GRAPHICS FORUM : JOURNAL OF THE EUROPEAN ASSOCIATION FOR COMPUTER GRAPHICS, WILEY-BLACKWELL, OXFORD, vol. 37, no. 2, 1 May 2018 (2018-05-01), Oxford, pages 511 - 521, XP055855200, ISSN: 0167-7055, DOI: 10.1111/cgf.13380 *

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114219007A (zh) * 2021-11-23 2022-03-22 慧之安信息技术股份有限公司 基于图像匹配的户型合理度评判方法及装置
CN114187588A (zh) * 2021-12-08 2022-03-15 贝壳技术有限公司 数据处理的方法、装置、存储介质和计算机程序产品
CN114187588B (zh) * 2021-12-08 2023-01-24 贝壳技术有限公司 数据处理的方法、装置和存储介质
CN114169063A (zh) * 2021-12-27 2022-03-11 北京奥科博思创新科技有限公司 一种建筑平面布局方法、装置及电子设备
EP4293562A1 (en) * 2022-06-16 2023-12-20 MFTB Holdco, Inc. Automated tools for assessing building mapping information generation
CN114972579A (zh) * 2022-06-22 2022-08-30 北京城市网邻信息技术有限公司 户型图构建方法、装置、设备及存储介质
CN115114713A (zh) * 2022-07-26 2022-09-27 江西少科智能建造科技有限公司 一种自动风盘布置方法、系统、存储介质及设备
CN115205418A (zh) * 2022-09-15 2022-10-18 武汉智筑完美家居科技有限公司 户型图重构方法、装置及电子设备、存储介质
CN115205418B (zh) * 2022-09-15 2022-12-13 武汉智筑完美家居科技有限公司 户型图重构方法、装置及电子设备、存储介质

Also Published As

Publication number Publication date
CN111598972A (zh) 2020-08-28
US20220383572A1 (en) 2022-12-01
CN111598972B (zh) 2021-06-29

Similar Documents

Publication Publication Date Title
WO2021196515A1 (zh) 户型图生成方法、装置、计算机设备和存储介质
JP6490220B2 (ja) ２次元図面に基づく３次元自動立体モデリング方法及びプログラム
JP6765487B2 (ja) 人工知能を用いるコンピュータ実施方法、ａｉシステム、及びプログラム
WO2020098076A1 (zh) 跟踪目标的定位方法、装置、设备及存储介质
US10930057B2 (en) Generating two-dimensional plan from three-dimensional image data
US11645781B2 (en) Automated determination of acquisition locations of acquired building images based on determined surrounding room data
WO2019058266A1 (en) SYSTEM AND METHOD FOR CONVERTING A 3D SCALE SOIL PLAN FOR THE CREATION AND RENDERING OF ARCHITECTURAL SCENES OF VIRTUAL REALITY, VIDEOS AND VISUAL IMAGES
CN112712584B (zh) 空间建模方法、装置、设备
US11714940B2 (en) Artificial intelligence determination of building metrics for code compliance with user interaction
US20230138762A1 (en) Automated Building Floor Plan Generation Using Visual Data Of Multiple Building Images
US11475174B2 (en) Methods and apparatus for artificial intelligence conversion of a two-dimensional reference into an actionable interface
US20220292421A1 (en) Methods and apparatus for artificial intelligence conversion of change orders into an actionable interface
US20230206393A1 (en) Automated Building Information Determination Using Inter-Image Analysis Of Multiple Building Images
US20230125295A1 (en) Automated Analysis Of Visual Data Of Images To Determine The Images' Acquisition Locations On Building Floor Plans
US20240061969A1 (en) Bim component creating method, digital design resource library application method and device
CN109448018B (zh) 跟踪目标的定位方法、装置、设备及存储介质
CN115330942B (zh) 多层空间三维建模方法、装置和计算机可读存储介质
CN115481458A (zh) 户型布局图的生成方法、装置、设备、存储介质及产品
CN111046467B (zh) 一种户型参数化建模方法及相关设备
US20210357642A1 (en) Image-based property condition determination systems, methods, and devices
US20240161348A1 (en) Automated Inter-Image Analysis Of Multiple Building Images For Building Information Determination
US20240012955A1 (en) Generative network-based floor plan generation
KR20230055329A (ko) 이사비용 산출 방법 및 장치
CN112199748A (zh) 基于人类活动信息的平面图设计方法、装置及终端设备
CN118036115A (zh) 用于建筑物信息确定的多个建筑物图像的自动图像间分析

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20928427

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 24.11.22)

122 Ep: pct application non-entry in european phase

Ref document number: 20928427

Country of ref document: EP

Kind code of ref document: A1