A method for self-generating views for interior decoration design

Abstract

The present invention discloses a method for self-generating views for interior decoration design, which relates to the field of image generation, including: three-dimensional grid modeling of the user's indoor space; scanning the original structure of the user's indoor space; analyzing the user's preference text through a semantic recognition model; obtaining a decoration design model through a three-dimensional self-generated neural network; and performing two-dimensional projection on the decoration design model to obtain a decoration design view. The present invention has high automation, reduces manual operations, improves design efficiency, can generate customized design solutions according to the user's personalized needs, and provides intuitive view display, which is convenient for users to understand and evaluate the design effect.

Description

Technical Field

[0001] The invention relates to the field of image generation, and in particular to a view self-generation method for interior decoration design. Background Art

[0002] With the advancement of technology, Computer Aided Design (CAD) has been widely used in interior decoration design. However, most of the existing design tools require designers to manually draw or adjust the design plan, which is time-consuming and inefficient.

[0003] At present, interior decoration has also seen a complex and changeable development with the improvement of people's living standards, such as the emergence of interior landscape design. Interior landscape design is a new type of design method that introduces the landscape in nature into the interior and realizes the interior landscape effect through reasonable design and artistic layout. It simulates the scenery of nature in a limited space, borrows the mountains, rocks, flowers and trees in nature, and integrates them with water bodies and buildings, so that the buildings and mountains are combined and reflected in the water, which not only has the shape and spirit of natural mountains, rocks, flowers and trees, but also has the function of conveying emotions.

[0004] For interior decoration design, especially interior landscape design, customers lack complex professional knowledge and experience, but have extremely high design requirements. Therefore, traditional interior decoration CAD tools are difficult to quickly display a preview of the decoration effect that satisfies them. Therefore, it is particularly urgent to develop a method that can automatically generate interior decoration, especially interior landscape design views. Summary of the invention

[0005] In view of the above-mentioned deficiencies in the prior art, the present invention provides a method for self-generating views for interior decoration design, which solves the problem that existing interior decoration CAD tools require designers to manually draw and adjust design plans, and cannot meet customers' complex and changeable decoration needs, especially interior landscape design needs.

[0006] In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is:

[0007] A method for self-generating views for interior decoration design comprises the following steps:

[0008] S1, perform three-dimensional mesh modeling on the user's indoor space to obtain an indoor three-dimensional model;

[0009] S2, scanning the original structure of the user's room, pre-processing the scanned data, and filling it into the indoor three-dimensional model;

[0010] S3, analyzing user preference texts through semantic recognition models to obtain user style features;

[0011] S4. According to the user's style characteristics, the three-dimensional self-generated neural network is used to select various decoration materials in the three-dimensional decoration material library, and the decoration materials are superimposed and combined in the indoor three-dimensional model to obtain a decoration design model;

[0012] S5. Perform two-dimensional projection on the decoration design model to obtain a decoration design view.

[0013] The beneficial effects of the present invention are as follows: the present invention has a high degree of automation, reduces manual operations, improves design efficiency, can generate customized design solutions according to the personalized needs of users, and provides intuitive view displays to facilitate users to understand and evaluate design effects.

[0014] Furthermore, the S1 performs three-dimensional grid modeling on the user's indoor space, and a method for obtaining the indoor three-dimensional model is: dividing the user's indoor space into N×M×L cube grids; N, M and L are all positive integers.

[0015] The beneficial effects of the above further scheme are: the interior space is divided into gridded finite elements, and the decoration design is converted into a translational superposition of layers and elements to facilitate subsequent further processing.

[0016] Furthermore, the S2 comprises the following sub-steps:

[0017] S21, performing a three-dimensional scan of the original structure of the user's room through a 3D scanner to obtain point cloud data of the original structure of the room;

[0018] S22. Perform noise reduction filtering on the indoor original structure point cloud data through the following formula:

[0019] ,

[0020] Among them, f(x+i,y+j,z+k) is the point cloud value of the coordinates of the original indoor structure (x+i,y+j,z+k), f fil (x, y, z) is the point cloud value of the coordinates of the original indoor structure (x, y, z) after denoising and filtering, and x, y, z, i, j and k are the integer numbers of the coordinates;

[0021] S23, performing maximum pooling down-sampling on the indoor original structure point cloud data after denoising filtering, and filling the down-sampled data into the indoor three-dimensional model.

[0022] The beneficial effects of the above further scheme are: denoising filtering is performed through three-dimensional neighborhood mean operation to remove the salt and pepper noise of the digital point cloud three-dimensional image; and through maximum pooling sampling, the texture information of the three-dimensional point cloud data is effectively retained, and the data is compressed, effectively saving the amount of calculation.

[0023] Furthermore, the user preference text in S3 includes: a plurality of decoration keywords and emotion words corresponding to each decoration keyword;

[0024] The decoration keywords include the following categories: space function, wall color, furniture style, lighting style, kitchen and bathroom style and landscape style;

[0025] The emotion words include the following words: "like very much", "like moderately", "no feeling", "dislike" and "dislike very much".

[0026] Furthermore, the semantic recognition model in S3 includes:

[0027] The decoration keyword search module is used to search for various decoration keywords in the user's favorite text;

[0028] The decoration keyword scoring module is used to identify the emotional words corresponding to each decoration keyword, and calculate the preference value of each decoration keyword based on the emotional words to form a vector sequence as the user style feature.

[0029] Furthermore, each of the decoration keywords is provided with a one-to-one corresponding decoration keyword J-bit code, where J is a positive integer;

[0030] Each of the sentiment words has a corresponding 3-bit score value:

[0031] The binary codeword for the “very like” rating value is “100”;

[0032] The binary code for the “generally like” rating value is “011”;

[0033] The binary code for the “no feeling” rating value is “010”;

[0034] The binary codeword for the “disgust” score value is “001”;

[0035] The binary codeword for the “very disgusting” rating value is “000”;

[0036] The decoration keyword preference value is a J+3-bit encoding value synthesized by concatenating the decoration keyword J-bit encoding and the emotional word 3-bit rating value.

[0037] The beneficial effect of the above further solution is that the above process establishes a convenient, efficient and accurate semantic coding method for interior decoration, and effectively converts semantic and digital information. It can identify user decoration preferences and perform digital calculations without complex and tedious deep machine learning and time recursive calculations.

[0038] Furthermore, the types of materials in the three-dimensional decoration material library in S4 include: inventory furniture, kitchen and bathroom appliances, lighting, latex wall, wallpaper and indoor landscape data obtained by three-dimensional scanning with a 3D scanner.

[0039] Furthermore, the three-dimensional self-generated neural network in S4 includes:

[0040] The decoration material screening sub-network is used to select materials from the 3D decoration material library based on the user's style characteristics;

[0041] The graphic overlay module is used to spatially translate the selected decoration materials in the indoor three-dimensional model according to the decoration specification constraints, so as to overlay and combine the decoration materials to obtain the decoration design model.

[0042] Furthermore, the decoration material screening sub-network includes N1 input neurons and N2 output neurons; N1 is the total number of vocabulary of decoration keywords, and N2 is the total number of materials in the three-dimensional decoration material library;

[0043] The function expression of the decoration material screening sub-network is:

[0044] ,

[0045] Among them, c n is the output value of the nth output neuron of the decoration material screening subnetwork. If its value is 0, it means that the nth material of the three-dimensional decoration material library is not selected. If its value is not 0, it means that the nth material of the three-dimensional decoration material library is selected. σ(∙) is the sigmoid function, w n,m is the weight coefficient connecting the mth input neuron and the nth output neuron, d m [J+2:3] is the value of the 3rd to J+2th bits of the mth decoration keyword preference value of the user style feature, with the high bit first. m [2:0] is the value of the 0th to 2nd bits of the mth decoration keyword preference value of the user style feature, with the high bit first. n is the bias coefficient of the nth output neuron, b ias is the global bias coefficient.

[0046] The beneficial effects of the above further scheme are: the three-dimensional self-generated neural network converts the selection of decoration materials into a logistic regression operation; the decoration design is converted into the translation and stacking of material graphics in the finite element discrete space under the specification constraints, which effectively realizes the automated design of decoration. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 A flowchart of a method for self-generating views for interior decoration design provided by an embodiment of the present invention;

[0048] Figure 2 This is a structural diagram of a semantic recognition model according to an embodiment of the present invention;

[0049] Figure 3 This is a three-dimensional self-generated neural network structure diagram of an embodiment of the present invention. DETAILED DESCRIPTION

[0050] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0051] like Figure 1 As shown, in one embodiment of the present invention, a method for self-generating views for interior decoration design includes the following steps:

[0052] S1. Perform three-dimensional grid modeling on the user's indoor space, divide the user's indoor space into N×M×L cube grids, and obtain an indoor three-dimensional model. N, M, and L are all positive integers.

[0053] The interior space is divided into gridded finite elements, and the decoration design is converted into layers and elements with translational overlay to facilitate further processing.

[0054] S2. Scan the original structure of the user's room, pre-process the scanned data, and fill it into the indoor three-dimensional model.

[0055] S2 includes the following sub-steps:

[0056] S21, performing a three-dimensional scan of the original structure of the user's room through a 3D scanner to obtain point cloud data of the original structure of the room;

[0057] S22. Perform noise reduction filtering on the indoor original structure point cloud data through the following formula:

[0058] ,

[0059] Among them, f(x+i,y+j,z+k) is the point cloud value of the coordinates of the original indoor structure (x+i,y+j,z+k), f fil (x, y, z) is the point cloud value of the coordinates of the original indoor structure (x, y, z) after denoising and filtering, and x, y, z, i, j and k are the integer numbers of the coordinates;

[0060] S23, performing maximum pooling down-sampling on the indoor original structure point cloud data after denoising filtering, and filling the down-sampled data into the indoor three-dimensional model.

[0061] The 3D neighborhood mean operation is used to perform noise reduction filtering to remove the salt and pepper noise of the digital point cloud 3D image. The maximum pooling sampling is used to effectively retain the texture information of the 3D point cloud data and compress the data, effectively saving the amount of calculation.

[0062] S3. Analyze user preference texts through semantic recognition models to obtain user style features.

[0063] The user preference text includes: multiple decoration keywords and emotional words corresponding to each decoration keyword. The decoration keywords include the following categories: space function, wall color, furniture style, lighting style, kitchen and bathroom style, and landscape style. The emotional words include the following words: "very like", "generally like", "no feeling", "dislike", and "very dislike".

[0064] like Figure 2 As shown, the semantic recognition model includes:

[0065] The decoration keyword search module is used to search for various decoration keywords in the user's favorite text;

[0066] The decoration keyword scoring module is used to identify the emotional words corresponding to each decoration keyword, and calculate the preference value of each decoration keyword based on the emotional words to form a vector sequence as the user style feature.

[0067] In this embodiment, each decoration keyword is provided with a one-to-one corresponding decoration keyword J-bit code, where J is a positive integer.

[0068] Each of the sentiment words has a corresponding 3-bit score value:

[0069] The binary codeword for the “very like” rating value is “100”;

[0070] The binary code for the “generally like” rating value is “011”;

[0071] The binary code for the “no feeling” rating value is “010”;

[0072] The binary codeword for the “disgust” score value is “001”;

[0073] The binary codeword for the “very disgusting” rating value is “000”;

[0074] The decoration keyword preference value is a J+3-bit encoding value synthesized by concatenating the decoration keyword J-bit encoding and the emotional word 3-bit rating value.

[0075] The above process establishes a convenient, efficient and accurate semantic coding method for interior decoration, which effectively converts semantic and digital information. It can identify user decoration preferences and perform digital calculations without complex and tedious deep machine learning and time recursive calculations.

[0076] To facilitate technical personnel to understand, taking this embodiment as an example, among the 6 types of decoration keywords mentioned above, there are 53 specific words, that is, the total number of decoration keyword words N1 is 53. In order to encode each decoration keyword, the J value is set to 6, which can represent a total of 64 decimal values ​​from "0" to "63".

[0077] According to the coding rules of the present invention, if the 6-bit code of the decoration keyword "white light LED chandelier" is set to the binary "000101", and the user feedback is "generally like", then the binary representation of the decoration keyword preference value is "000101011".

[0078] S4. According to the user's style characteristics, the three-dimensional self-generated neural network is used to screen various decoration materials in the three-dimensional decoration material library, and they are superimposed and combined in the indoor three-dimensional model to obtain a decoration design model.

[0079] The types of materials in the 3D decoration material library include: inventory furniture, kitchen and bathroom appliances, lighting, latex walls, wallpapers and interior landscape data obtained through 3D scanning with a 3D scanner.

[0080] like Figure 3 As shown, the three-dimensional self-generated neural network includes:

[0081] The decoration material screening sub-network is used to select materials from the 3D decoration material library based on the user's style characteristics;

[0082] The graphic overlay module is used to spatially translate the selected decoration materials in the indoor three-dimensional model according to the decoration specification constraints, so as to overlay and combine the decoration materials to obtain the decoration design model.

[0083] The decoration material screening sub-network includes N1 input neurons and N2 output neurons; N1 is the total number of decoration keyword words, and N2 is the total number of materials in the three-dimensional decoration material library;

[0084] The function expression of the decoration material screening sub-network is:

[0085] ,

[0086] Among them, c nis the output value of the nth output neuron of the decoration material screening subnetwork. If its value is 0, it means that the nth material of the three-dimensional decoration material library is not selected. If its value is not 0, it means that the nth material of the three-dimensional decoration material library is selected. σ(∙) is the sigmoid function, w n,m is the weight coefficient connecting the mth input neuron and the nth output neuron, d m [J+2:3] is the value of the 3rd to J+2th bits of the mth decoration keyword preference value of the user style feature, with the high bit first. m [2:0] is the value of the 0th to 2nd bits of the mth decoration keyword preference value of the user style feature, with the high bit first. n is the bias coefficient of the nth output neuron, b ias is the global bias coefficient.

[0087] Since the concept of "bit" is a professional term in the field of digital logic and computer, in order to facilitate the understanding of technical personnel, we still take the binary code "000101011" of the decoration keyword preference value as an example. In this field, counting starts at 0, and the numerical binary code word composed of the 0th to 2nd bits with the high bit first is "011", and the numerical arithmetic meaning is "3"; the numerical binary code word composed of the 3rd to J+2th bits with the high bit first is "000101", and the numerical arithmetic meaning is "5".

[0088] The three-dimensional self-generated neural network of the present invention converts the selection of decoration materials into a logistic regression operation; and converts the decoration design into the translation and stacking of material graphics in a finite element discrete space under standard constraints, thereby effectively realizing the automated design of decoration.

[0089] In this embodiment, the decoration specification constraint is the spatial stacking rules of each material in the three-dimensional decoration material library set by the designer through iterative training of the graphic overlay module. For example, the material "TV cabinet" cannot be "adhered" to the material "sofa" in three-dimensional space; the spatial distance between the material "TV cabinet" and the material "sofa" cannot exceed 5 meters.

[0090] For the indoor three-dimensional model modeled by three-dimensional grid, each space stacking rule of this embodiment is in the range of 0 to a positive integer value, so it is not difficult for technical personnel to understand.

[0091] The weight coefficient w in this embodiment n,m The designer performs iterative training on the decoration material screening sub-network. The present invention is not limited to a specific iterative training method, so the parameter training process is not described in detail.

[0092] S5. Perform two-dimensional projection on the decoration design model to obtain a decoration design view.

[0093] The two-dimensional projection of three-dimensional data is only used to display the design plan to the user, and simple and feasible projection plans can be found in a large number of existing literature, so this article will not go into details.

[0094] In summary, the present invention has high automation, reduces manual operations, improves design efficiency, can generate customized design solutions according to user's personalized needs, and provides intuitive view display to facilitate users to understand and evaluate design effects.

[0095] The present invention uses specific embodiments to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

[0096] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. A method for self-generating views for interior decoration design, characterized in that: The following steps are involved: S1, perform three-dimensional mesh modeling on the user's indoor space to obtain an indoor three-dimensional model; S2, scanning the original structure of the user's room, pre-processing the scanned data, and filling it into the indoor three-dimensional model; S3, analyzing user preference texts through semantic recognition models to obtain user style features; The semantic recognition model in S3 includes: The decoration keyword search module is used to search for various decoration keywords in the user's favorite text; The decoration keyword scoring module is used to identify the emotion words corresponding to each decoration keyword, and calculate the preference value of each decoration keyword according to the emotion words to form a vector sequence as the user style feature, where the decoration keyword preference value is a J+3-bit code value spliced ​​by the decoration keyword J-bit code and the emotion word 3-bit score value; S4. According to the user's style characteristics, the three-dimensional self-generated neural network is used to select various decoration materials in the three-dimensional decoration material library, and the decoration materials are superimposed and combined in the indoor three-dimensional model to obtain a decoration design model; S5. Perform two-dimensional projection on the decoration design model to obtain a decoration design view; The three-dimensional self-generated neural network in S4 includes: The decoration material screening sub-network is used to select materials from the 3D decoration material library based on the user's style characteristics; A graphic overlay module is used to spatially translate the selected decoration materials in the indoor three-dimensional model according to the decoration specification constraints, so as to overlay and combine the decoration materials to obtain a decoration design model; The decoration material screening sub-network includes N1 input neurons and N2 output neurons; N1 is the total number of vocabulary of decoration keywords, and N2 is the total number of materials in the three-dimensional decoration material library; The function expression of the decoration material screening sub-network is: ， Among them, c n is the output value of the nth output neuron of the decoration material screening subnetwork. If its value is 0, it means that the nth material of the three-dimensional decoration material library is not selected. If its value is not 0, it means that the nth material of the three-dimensional decoration material library is selected. σ(∙) is the sigmoid function, w n,m is the weight coefficient connecting the mth input neuron and the nth output neuron, d m [J+2:3] is the value of the 3rd to J+2th bits of the mth decoration keyword preference value of the user style feature, with the high bit first. m [2:0] is the value of the 0th to 2nd bits of the mth decoration keyword preference value of the user style feature, with the high bit first. n is the bias coefficient of the nth output neuron, b ias is the global bias coefficient.

2. The method for self-generating views for interior decoration design according to claim 1, characterized in that: The S1 performs three-dimensional grid modeling on the user's indoor space to obtain the indoor three-dimensional model by dividing the user's indoor space into N×M×L cube grids; N, M and L are all positive integers.

3. The method for self-generating views for interior decoration design according to claim 2, characterized in that: The S2 comprises the following sub-steps: S21, performing a three-dimensional scan of the original structure of the user's room through a 3D scanner to obtain point cloud data of the original structure of the room; S22. Perform noise reduction filtering on the indoor original structure point cloud data through the following formula: ， Among them, f(x+i,y+j,z+k) is the point cloud value of the coordinates of the original indoor structure (x+i,y+j,z+k), f fil (x, y, z) is the point cloud value of the coordinates of the original indoor structure (x, y, z) after denoising and filtering, and x, y, z, i, j and k are the integer numbers of the coordinates; S23, performing maximum pooling down-sampling on the indoor original structure point cloud data after denoising filtering, and filling the down-sampled data into the indoor three-dimensional model.

4. The method for self-generating views for interior decoration design according to claim 1, characterized in that: The user preference text in S3 includes: a plurality of decoration keywords and emotion words corresponding to each decoration keyword; The decoration keywords include the following categories: space function, wall color, furniture style, lighting style, kitchen and bathroom style and landscape style; The emotion words include the following words: "very like", "moderately like", "no feeling", "dislike" and "very dislike".

5. The method for self-generating views for interior decoration design according to claim 4, characterized in that: Each of the decoration keywords is provided with a one-to-one corresponding decoration keyword J bit code, where J is a positive integer; Each of the sentiment words has a corresponding 3-bit score value: The binary code for the "very like" rating value is "100"; The binary code for the "generally liked" rating value is "011"; The binary code for the "no feeling" rating value is "010"; The binary code for the "disgust" score value is "001"; The binary code for the "very disgusting" rating value is "000".

6. The method for self-generating views for interior decoration design according to claim 5, characterized in that: The types of materials in the three-dimensional decoration material library in S4 include: inventory furniture, kitchen and bathroom appliances, lighting, latex wall, wallpaper and indoor landscape data obtained by three-dimensional scanning with a 3D scanner.

Images