Method for converting two-dimensional image of building component into three-dimensional model

Abstract

A method for converting a two-dimensional image of a building component into a three-dimensional model, relating to the technical field of image processing. The method comprises: comparing the size of a three-dimensional model of a component with the size of an actual building component, so as to obtain error data between the two; generating a corresponding accuracy from a model error data set; if the accuracy of the three-dimensional model of the component does not exceed an expectation, performing overlap analysis on the three-dimensional model of the component and the building component, so as to obtain a degree of overlap, and generating a complete coefficient from the degree of overlap; according to the complete coefficient, screening out defect regions in the three-dimensional model of the component, and associating and obtaining optimization priorities of the defect regions; and after a building component model optimization knowledge graph provides a model optimization solution for the defect regions, sequentially optimizing the defect regions according to the optimization priorities. By using the building component model optimization knowledge graph to provide the corresponding model optimization solution for the defect regions, optimization for each defect region is more targeted.

Description

A method for converting two-dimensional images of building components into three-dimensional models TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, in particular to a method for converting two-dimensional images of building components into three-dimensional models. BACKGROUND

[0002] In the construction industry, building information modeling and management (BIM) is widely recognized as a new tool that can improve the efficiency of building and large infrastructure (such as airports, subway stations, and train stations) operation and asset management, but many buildings and infrastructures built before the widespread use of BIM technology still use traditional construction drawings as the design and construction delivery; it often takes a lot of time and manpower to rebuild BIM models from old CAD engineering drawings. Therefore, developing a technology that can automatically convert CAD building drawings into BIM models can save the financial and material resources consumed by manual remodeling, and immediately realize the value of BIM models in the operation stage of buildings and infrastructures.

[0003] At present, there are some researches and patent inventions on two-dimensional drawings for recognition and three-dimensional reconstruction. The existing advanced methods basically use computer vision technology based on deep learning to recognize and semantically segment drawings converted into pixel pictures. For example, in the Chinese invention patent with the authorization announcement number CN110909650B, a CAD drawing recognition method and device based on domain knowledge and target detection are disclosed, which includes: a CAD drawing analysis module for obtaining drawing information in a to-be-recognized CAD drawing, and converting the to-be-recognized CAD drawing into multiple to-be-recognized pixel drawings; a target detection module for performing aggregated component recognition on the multiple to-be-recognized pixel drawings to determine a component information set of components in the to-be-recognized CAD drawing; a building space file analysis module for obtaining spatial information in the to-be-recognized CAD drawing; a system information fusion module for obtaining coordinate position information of the components in the building space according to the spatial information and the component information set, and combining the drawing information to determine component system information by searching the topology of the components, and fusing the coordinate position information and the component system information to generate component recognition information. The method improves the accuracy of component recognition and information extraction, and enhances the expandability of CAD drawing recognition.

[0004] Although these existing methods achieve high recognition accuracy, the drawing rules of equipment graphics are different in different projects, and it is difficult to use deep learning models to recognize drawings with untrained drawing methods. Therefore, many researchers also reconstruct three-dimensional models of components by recognizing images and point clouds collected on the construction site; however, although these methods can better capture the real position and geometric shape of two-dimensional components,

[0005] But after the building component design is completed and the corresponding two-dimensional design drawing is obtained, and the component three-dimensional model is generated from the two-dimensional design drawing, the generated component three-dimensional model may still have some defect areas, for example, poor precision, lower smoothness of the model surface, etc., so targeted maintenance is still needed for these defect areas, but in the existing maintenance method, the overall optimization is usually performed after the generation of the component three-dimensional model, but the targeting of this optimization is poor, which is easy to cause the generation efficiency of the component three-dimensional model to be low.

[0006] Therefore, the present application provides a method for converting a two-dimensional image of a building component into a three-dimensional model. SUMMARY

[0007] (I) Technical problems to be solved

[0008] In view of the deficiencies in the prior art, the present application provides a method for converting a two-dimensional image of a building component into a three-dimensional model, by generating a corresponding accuracy from a model error data set, if the accuracy of the component three-dimensional model does not exceed the expectation, performing coincidence analysis on the component three-dimensional model and the building component and obtaining a coincidence degree, generating a completeness coefficient from the coincidence degree, screening out defect areas in the component three-dimensional model according to the completeness coefficient, correlating to obtain the optimization priority of each defect area, and then giving a model optimization scheme for the defect areas from the building component model optimization knowledge graph, and optimizing each defect area in turn according to the optimization priority, the targeting of the optimization is higher when optimizing each defect area; thereby solving the technical problems raised in the background art.

[0009] (II) Technical solutions

[0010] To achieve the above purpose, the present application is implemented by the following technical solutions: a method for converting a two-dimensional image of a building component into a three-dimensional model, comprising: performing quality analysis on a source image and obtaining quality data, generating an image quality coefficient from the image quality data , screening out low-quality images from a plurality of source images according to the image quality coefficient , matching an optimization scheme for the low-quality images, and performing optimization on the low-quality images according to the optimization scheme;

[0011] Among them, the contrast and the noise density are linearly normalized, and the corresponding data values are mapped to the interval , and the image quality coefficient is generated according to the following formula :

[0012] ;

[0013] weight coefficient, , ;

[0014] The optimized source image is taken as a target image, and after recognizing the edges in the target image, the key points in the target image are extracted, and a semi-global matching algorithm is used to generate a depth map. After converting the depth map into point cloud data in a three-dimensional space, three-dimensional coordinates are obtained from pixel coordinates and corresponding depth values, and a point cloud is constructed;

[0015] A model surface is generated from the point cloud data, the obtained model surface is converted into a polygon mesh model composed of triangular meshes, holes in the polygon mesh are repaired, noise in the polygon mesh is smoothed, and the initial model of the component is adjusted to obtain a three-dimensional model of the building component;

[0016] The size of the component three-dimensional model is compared with the size of the actual building component to obtain error data between the two, and a corresponding accuracy is generated from the model error data set If the accuracy of the component three-dimensional model does not exceed the expectation, a model optimization instruction is sent to the outside;

[0017] The component three-dimensional model and the building component are analyzed for coincidence and a coincidence degree is obtained, and a completeness coefficient is generated from the coincidence degree According to the completeness coefficient , the defect areas are filtered out in the component three-dimensional model, and the optimization priorities of the defect areas are obtained , and then the model optimization scheme is given to the defect areas by the building component model optimization knowledge graph, and the defect areas are optimized in turn according to the optimization priorities.

[0018] Further, source images related to the two-dimensional drawings of the building are collected, the quality of the source images is analyzed and quality data is obtained, and the quality data is summarized to construct a quality data set; an image quality coefficient is constructed from the image quality data set If the obtained image quality coefficient is lower than the preset quality threshold, the corresponding image quality is taken as a low-quality image;

[0019] After filtering out the low-quality image, the quality data of the low-quality image is obtained, the features of the quality data are extracted and the corresponding image features are obtained; a plurality of image optimization schemes are obtained to construct an image optimization scheme library; the optimized source image is obtained, and a three-dimensional reconstruction instruction is sent to the outside.

[0020] Further, after receiving the three-dimensional reconstruction instruction, the optimized source image is obtained and used as the target image, the strong edges in the target image are detected using the Canny edge detection algorithm, and the Sobel operator is applied to detect the horizontal and vertical edges in the target image. After obtaining the edges of the target image, the continuity of the edges is optimized, and a key point detection instruction is sent to the outside.

[0021] Further, after receiving the key point detection instruction, the Harris corner point detection algorithm is used to detect the corner points in the target image, the corner points are used as the key points, and the SURF algorithm is used to calculate the descriptors of the key points. The BFMatcher algorithm is used to identify and match the feature points in the left and right images, and the RANSAC algorithm is used to estimate the model parameters by iteratively selecting a random subset to identify and exclude outliers.

[0022] The disparity map is calculated by finding the best matching pixel blocks in the left and right images, and corresponding processing measures are taken according to the disparity map. After generating the depth map using the semi-global matching algorithm, the SGM algorithm is applied to aggregate the cost in multiple directions.

[0023] Further, a smooth and continuous surface is generated from the point cloud data using the Poisson surface reconstruction algorithm, which is used as the model surface. The Marching Cubes algorithm is used to extract triangular meshes from the voxel data of the model surface, generating a mesh model composed of triangular meshes, and converting the implicit surface to an explicit polygon mesh to obtain the initial model of the component.

[0024] Further, the texture coordinates of the two-dimensional image are assigned to each vertex of the initial model of the component through UV mapping, ensuring accurate alignment and minimal deformation of the texture map on the model surface. The holes in the polygon mesh are repaired using the mesh repair tool, and the noise in the polygon mesh is smoothed by applying the Taubin smoothing algorithm. The initial model of the component is adjusted according to the actual building size to obtain the three-dimensional model of the building component.

[0025] Further, after obtaining the reconstructed three-dimensional model of the component, it is used as the target model. After dimensioning the target model, the dimensioning data of the target model and the size data of the actual component are obtained respectively. The dimensioning is compared with the size of the actual building component, and the error ratio data between the two is obtained to build the model error data set.

[0026] Further, the corresponding accuracy is generated from the model error data set, wherein the error ratio is linearly normalized, and the corresponding data value is mapped to the interval , according to the following formula:

[0027] ;

[0028] in, Let be the error percentage at the i-th dimension point. This represents the acceptable target value for the error ratio. The standard deviations of several error proportions; weighting coefficients: , ,and .

[0029] Furthermore, upon receiving the model optimization instruction, the target model is divided into several sub-regions, several anchor points are selected within each sub-region, and a face is randomly selected from the regions adjacent to the anchor points as the model face; corresponding faces are selected at the corresponding positions of the actual building components as standard faces; and the overlap between the standard face and the model face is obtained based on the dimensional data between the standard face and the model face.

[0030] Furthermore, the integrity coefficient is generated from the overlap degree. Among them, in terms of overlap When under dimensionless conditions, proceed as follows:

[0031] ;

[0032] Weighting coefficients: , ; Let be the degree of overlap between the i-th standard surface and the model surface within the sub-region. This is the acceptable standard value for overlap. The number of overlaps;

[0033] If the completeness coefficient If the threshold is not exceeded, the corresponding sub-region is designated as a defect region, and each defect region is marked on the 3D model of the component.

[0034] Furthermore, the location and integrity coefficient of each defect area are obtained. Post-association optimization priority acquisition Based on optimization priority Each defective area is marked as follows:

[0035] ;

[0036] in, It is the first The defect area to the first The shortest distance to each defect area Average distance; weighting coefficient: ; a complete coefficient for the i-th defect region, a corresponding qualified standard value, a variance of the complete coefficient.

[0037] Further, model state data in each defect region is acquired, feature extraction is performed on the position of the model state data in the defect region, and corresponding model optimization features are acquired; building component model optimization is taken as a target word, and a building component model optimization knowledge graph is constructed in advance; and the model optimization features and the model optimization scheme in the defect region are used.

[0038] (Three) beneficial effects

[0039] The application provides a method for converting a two-dimensional image of a building component into a three-dimensional model, which has the following beneficial effects:

[0040] 1. Image quality data is used to construct an image quality coefficient According to the acquired image quality coefficient , the image quality of the source image can be judged, low-quality images can be screened out, and the low-quality images can be optimized, thereby improving the model construction quality when constructing a three-dimensional model.

[0041] 2. After feature extraction is performed on the image quality data and corresponding image optimization features are acquired, the low-quality images are optimized according to the image optimization features, thereby improving the efficiency and effect of optimization when the images need to be optimized.

[0042] 3. A semi-global matching algorithm is used to generate a smoother and more accurate depth map, and the SGM algorithm is applied to aggregate costs in multiple directions, thereby effectively reducing noise and discontinuity in the disparity map and generating a more consistent depth map.

[0043] 4. Through accurate surface reconstruction, accurate extraction of mesh models, accurate alignment of texture mapping, integrity of mesh repair, optimization of mesh smoothing, and accurate adjustment of model size, high-quality three-dimensional model construction is realized.

[0044] 5. The model size is compared with the size of the actual building component, error data between the two is acquired, and an accuracy is generated according to the error data , thereby judging and evaluating the precision of the component three-dimensional model, and if the current accuracy of the component three-dimensional model does not meet the expectation, the component three-dimensional model can be further optimized in time, thereby improving the model construction effect.

[0045] 6. According to the acquired complete coefficient The consistency between each sub-region and the actual component is judged, if the consistency fails to reach the expectation, the defect region is screened out after the judgment is completed, the optimization target can be reduced when the target is optimized, which helps to realize targeted optimization and improve the model optimization efficiency.

[0046] 7. Analyzing to obtain the optimization priority of each defect region After the defect regions are sorted, the efficiency of optimization can be improved by optimizing each defect region in turn; after the defect regions are selected, the corresponding model optimization scheme is given to the defect region by the building component model optimization knowledge graph after data collection and feature extraction in the defect region, the optimization is more targeted when each defect region is optimized, which further improves the optimization effect. BRIEF DESCRIPTION OF DRAWINGS

[0047] Fig. 1 is a flowchart of the method for converting a two-dimensional image into a three-dimensional model. DETAILED DESCRIPTION

[0048] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0049] Please refer to Fig. 1, the present application provides a method for converting a two-dimensional image of a building component into a three-dimensional model, comprising:

[0050] Step one, quality analysis of the source image and obtaining quality data, generating image quality coefficient from image quality data According to the image quality coefficient In several source images, low-quality images are screened out, and optimization schemes are matched for the low-quality images, then the optimization schemes are executed to optimize the low-quality images;

[0051] The step one includes the following contents;

[0052] Step 101, when it is needed to convert a two-dimensional image into a three-dimensional model, source images related to the two-dimensional drawing of building construction are collected, quality analysis of the source images is performed and quality data are obtained, including contrast, brightness and image noise, etc., after summarizing, a quality data set is constructed;

[0053] Image quality coefficient is constructed from image quality data set According to the image quality coefficient, the image quality of the source image is judged, the mode is as follows: the contrast and noise density Linear normalization processing is performed to map the corresponding data value to the interval , according to the following formula:

[0054]

[0055] weight coefficient, , The weight coefficient can be obtained by referring to the analytic hierarchy process;

[0056] If the obtained image quality coefficient is lower than the preset quality threshold, the corresponding image quality is regarded as low-quality image;

[0057] In use, after obtaining a plurality of two-dimensional drawing image data, an image quality coefficient is constructed from the image quality data, and the image quality of the source image can be judged according to the obtained image quality coefficient . After filtering out low-quality images, the low-quality images can be optimized, which can improve the quality of model construction when constructing a three-dimensional model;

[0058] Step 102, after filtering out low-quality images, obtain the quality data of the low-quality images, set the qualified standard by referring to various image quality indicators, and then extract the features of the quality data and obtain the corresponding image features; a plurality of image optimization schemes are obtained by offline collection and online retrieval, and an image optimization scheme library is generated by summarizing the image optimization schemes;

[0059] According to the correspondence between the image features and the optimization schemes, the image optimization scheme library is used to match the specific optimization scheme for the low-quality image, and the low-quality image is optimized by executing the optimization scheme;

[0060] For example, adjust the contrast of the image to make the brightness distribution more uniform. Use the curve adjustment or histogram adjustment tool to enhance the contrast of the image; apply median filtering or Gaussian filtering to remove noise points in the image; use the four-point perspective correction method to select four corner points in the image for correction, and adjust the scale of the image according to the known size of the reference object;

[0061] After executing the optimization scheme, the optimized source image is obtained, and a three-dimensional reconstruction instruction is issued to the outside;

[0062] In use, the contents in steps 101 and 102 are combined:

[0063] After filtering out low-quality images, the image quality data is extracted to obtain the corresponding image optimization features, and the low-quality images are optimized according to the image optimization features, which can improve the efficiency and effect of optimization when the image needs to be optimized.

[0064] ​Step two, taking the optimized source image as the target image, after recognizing the edges in the target image, extracting the key points in the target image, and using the semi-global matching algorithm to generate a depth map, converting the depth map into point cloud data in three-dimensional space, obtaining three-dimensional coordinates from pixel coordinates and corresponding depth values, and constructing a point cloud;

[0065] The step two includes the following contents:

[0066] Step 201, after receiving the three-dimensional reconstruction instruction, obtaining the optimized source image and taking it as the target image, setting the high and low threshold in advance, using the Canny edge detection algorithm to detect the strong edges in the target image, and applying the Sobel operator to detect the horizontal and vertical edges in the target image, optimizing the continuity of the edges after obtaining the edges of the target image, and issuing a key point detection instruction to the outside;

[0067] When in use, through the Canny edge detection algorithm and the Sobel operator, the strong edges and directional edges in the target image can be accurately identified, providing clear and accurate edge information for subsequent key point detection and feature matching; optimizing the continuity of the edges helps to improve the stability and accuracy of the subsequent processing steps, especially in key point detection and feature matching, continuous edges can provide more reliable feature point positioning.

[0068] Step 202, after receiving the key point detection instruction, setting the sensitivity parameter in advance, using the Harris corner detection algorithm to detect the corners in the target image, taking the corners as the key points, using the SURF algorithm to calculate the descriptors of the key points for subsequent feature matching; using the BFMatcher algorithm to identify and match the feature points in the left and right images, these feature points are regions with significant changes in the image, such as corners, edges or texture-rich regions; and using the RANSAC algorithm to estimate model parameters by iteratively selecting a random subset, and using these parameters to verify other points, identify and exclude outliers, significantly improving the robustness of matching and reducing false matches caused by image noise or occlusion.

[0069] When in use, using the Harris corner detection algorithm combined with the preset sensitivity parameter, the corners in the image can be effectively identified, and these corners serve as key points, providing stable reference points for feature matching. Using the SURF algorithm to calculate the descriptors of the key points, combined with the BFMatcher algorithm for feature matching, can efficiently identify and match the feature points in the left and right images, which are regions with significant changes in the image, such as corners, edges or texture-rich regions.

[0070] Step 203, calculate the disparity map by finding the best matching pixel block in the left and right images, considering that block matching may produce noise and discontinuity, take corresponding processing measures according to the disparity map, such as filling invalid areas, smoothing processing, etc., improve the accuracy and consistency of the disparity map; and use the semi-global matching algorithm to generate a more smooth and accurate depth map, apply the SGM algorithm to effectively reduce the noise and discontinuity in the disparity map by cost aggregation in multiple directions, generate a more consistent depth map;

[0071] Convert the depth map into point cloud data in three-dimensional space, convert the pixel coordinates and corresponding depth values into three-dimensional coordinates, and thus construct a point cloud, and use point cloud processing software to denoise and smooth the point cloud, providing a high-quality data basis for subsequent point cloud analysis, model reconstruction and visualization.

[0072] In use, in combination with the contents in steps 201 to 203:

[0073] Use the semi-global matching algorithm to generate a more smooth and accurate depth map, apply the SGM algorithm to effectively reduce the noise and discontinuity in the disparity map by cost aggregation in multiple directions, generate a more consistent depth map.

[0074] Step three, generate a model surface from the point cloud data, convert the obtained model surface into a polygon mesh model composed of triangular meshes, smooth the noise in the polygon mesh after repairing the holes in the polygon mesh, and adjust the initial model of the component to obtain a three-dimensional model of the building component;

[0075] The step three includes the following contents:

[0076] Step 301, generate a smooth and continuous surface from the point cloud data using the Poisson surface reconstruction algorithm, use the Marching Cubes algorithm to extract triangular meshes from the voxel data of the model surface to generate a mesh model composed of triangular meshes, and convert the implicit surface to an explicit polygon mesh, thus providing accurate model geometry, laying a foundation for subsequent texture mapping and model adjustment; at this time, the preliminary construction of the component three-dimensional model is completed, and the initial model of the component is obtained;

[0077] In use, generate a smooth and continuous surface from the point cloud data using the Poisson surface reconstruction algorithm, which ensures the continuity and smoothness of the model surface, reduces the noise and irregularities that may exist in the point cloud data.

[0078] Step 302, by UV mapping to assign the texture coordinates of the two-dimensional image to each vertex of the component initial model, ensure the accurate alignment and minimal deformation of the texture map on the model surface; make the three-dimensional model visually more realistic and delicate, enhance the expressiveness and application value of the model;

[0079] Use the mesh repair tool to repair holes in the polygon mesh, ensure the integrity and closure of the model, avoid rendering and analysis problems caused by holes; and apply Taubin smoothing algorithm to smooth the noise of the polygon mesh, further optimize the surface quality of the model, reduce the roughness and irregularity caused by noise. According to the actual building size, adjust the component initial model to obtain the three-dimensional model of the building component, complete the model construction process;

[0080] In use, combined with the contents in steps 301 and 302:

[0081] Through accurate surface reconstruction, accurate extraction of mesh model, accurate alignment of texture mapping, integrity of mesh repair, optimization of mesh smoothing and accurate adjustment of model size, high-quality three-dimensional model construction is realized, which provides reliable three-dimensional model support for architectural design and construction.

[0082] Step four, compare the size of the component three-dimensional model with the size of the actual building component, obtain the error data between the two, generate the corresponding accuracy from the model error data set, if the accuracy of the component three-dimensional model does not exceed the expectation, send the model optimization instruction to the outside;

[0083] The step four includes the following contents:

[0084] Step 401, after obtaining the reconstructed component three-dimensional model, take it as the target model, after size labeling of the target model, obtain the labeled size data of the target model and the size data of the actual component respectively, compare the labeled size with the size of the actual building component, obtain the error proportion data between the two, and then collect and build the model error data set;

[0085] Step 402, generate the corresponding accuracy from the model error data set, verify the accuracy of the target model according to the accuracy, wherein, the error proportion is linearly normalized, and the corresponding data value is mapped to the interval , according to the following formula:

[0086] ;

[0087] Wherein, is the error proportion at the i-th size labeling point, This represents the acceptable target value for the error ratio. The standard deviations of several error proportions; weighting coefficients: , ,and ;

[0088] Based on historical data and management expectations for model accuracy, an accuracy threshold is pre-set; if the obtained accuracy... If the accuracy threshold is exceeded, it indicates that the accuracy of the target model is feasible, and the 3D model of the component is generated. Conversely, if the accuracy of the target model is insufficient, a model optimization command is sent to the outside to optimize the target model.

[0089] When using this method, refer to steps 401 and 402:

[0090] After constructing the 3D model of the component and annotating its dimensions, the model dimensions are compared with the dimensions of the actual building component to obtain the error data and generate an accuracy score. Based on accuracy The accuracy of the 3D model of a component can be judged and evaluated. If the current accuracy of the 3D model of the component is... If the expected results are not achieved, the 3D model of the component can be further optimized in a timely manner to improve the model construction effect.

[0091] Step 5: Perform an overlap analysis between the 3D model of the component and the building component, obtain the overlap degree, and generate an integrity coefficient from the overlap degree. Based on the completeness coefficient Defect regions are identified within the 3D model of the component, and the optimization priority of each defect region is obtained by correlation. Then, the knowledge graph of building component model optimization provides a model optimization scheme for the defect area, and each defect area is optimized in turn according to the optimization priority.

[0092] Step five includes the following:

[0093] Step 501: After receiving the model optimization instruction, divide the target model into several sub-regions, select several anchor points in each sub-region, and randomly select a face in the region adjacent to the anchor point as the model face; select the corresponding face at the corresponding position of the actual building component and use it as the standard face; obtain the overlap between the standard face and the model face based on the dimensional data between the standard face and the model face.

[0094] Step 502: Generate the integrity coefficient from the overlap degree. Based on the completeness coefficient Defective regions are selected from several sub-regions, among which, based on overlap... In the dimensionless condition, the following way is adopted:

[0095] ;

[0096] The weight coefficient: , The weight coefficient can be obtained by the analytic hierarchy process; is the coincidence degree between the i-th standard surface in the sub-region and the model surface, is the qualified standard value of the coincidence degree, is the number of the coincidence degree,

[0097] According to the historical data and the defect tolerance of each sub-region of the target model, the integrity threshold is set in advance;

[0098] If the integrity coefficient does not exceed the integrity threshold, the corresponding sub-region is regarded as a defect region, and each defect region is marked on the component three-dimensional model;

[0099] When the accuracy of the component three-dimensional model cannot meet the expectation and further optimization is needed, by selecting a number of comparison surfaces and performing coincidence on the comparison surfaces, the integrity coefficient is obtained to judge the consistency between each sub-region and the actual component. If the consistency cannot meet the expectation, the defect region is screened out after the judgment is completed, which can narrow the optimization target when optimizing the target, and helps to realize targeted optimization and improve the model optimization efficiency;

[0100] Step 503, after obtaining the position and integrity coefficient of each defect region, the optimization priority is obtained, and the order of optimization is determined when each sub-region needs to be optimized, according to the optimization priority , each defect region is marked as follows:

[0101] ;

[0102] Among them, is the shortest distance from the i-th defect region to the j-th defect region, is the average value of the distance; the weight coefficient is: ; is the integrity coefficient of the i-th defect region, is the corresponding qualified standard value, is the variance of the integrity coefficient;

[0103] ​​​Step 504, when optimization of each defect area is needed, obtain model state data in each defect area, such as model accuracy, model size and smoothing state, etc., perform feature extraction on the model state data position in the defect area, and obtain corresponding model optimization features;

[0104] With building component model optimization as the target word, after deep retrieval and entity relationship building, a building component model optimization knowledge graph is constructed in advance; using the trained matching model, according to the model optimization features and the model optimization scheme in the defect area, the building component model optimization knowledge graph gives the corresponding model optimization scheme for the defect area, and according to the optimization priority, each defect area is optimized;

[0105] In use, in combination with the contents in steps 501 and 504:

[0106] After several defect areas have been screened out, the optimization priority of each defect area is analyzed and obtained , after sorting each defect area, the efficiency of optimization can be improved by optimizing each defect area in turn;

[0107] And as further content, after selecting the defect area, the corresponding model optimization scheme for the defect area is given by the building component model optimization knowledge graph after data collection and feature extraction in the defect area, and when optimizing each defect area, the pertinence is higher, and the optimization effect is further improved.

[0108] Analytic Hierarchy Process (AHP) is a decision-making method that decomposes elements related to decision-making into target, criteria, scheme and other levels, and conducts qualitative and quantitative analysis on this basis. It is particularly suitable for handling target systems with hierarchical and staggered evaluation indicators, and when the target value is difficult to quantify, AHP is an effective decision-making tool.

[0109] The core of AHP is to decompose the decision-making problem into multiple levels to form a hierarchical structure, which usually includes the target layer, the criterion layer, the sub-criterion layer and the scheme layer. By solving the characteristic vector of the judgment matrix, the priority weight of each element in a level to a certain element in the previous level is obtained, and the final weighted sum is used to merge the priority of each alternative scheme to the total target, so as to find the optimal scheme.

[0110] The construction method of building component model optimization knowledge graph can refer to the following contents:

[0111] Data Collection: Data source determination, collection of various types of data related to building component model optimization, including design drawings, construction drawings, construction specifications, material performance data, historical project data, etc. Data sources may include design units, construction units, material suppliers, research institutions, etc.

[0112] Data Filtering and Cleaning: Filtering of collected data to remove redundant, incorrect or irrelevant data. Data cleaning to ensure data accuracy and consistency.

[0113] Data Preprocessing: Entity recognition, using natural language processing (NLP) technology to extract key entities from text data, such as building components (beams, columns, plates, walls, etc.), material types, design parameters, etc. Identify attributes between entities, such as size, strength, weight, etc.

[0114] Relationship Extraction: Extract relationships between entities from data through relationship extraction technology, such as connection relationships between components, material and component correspondence, design parameters and component performance, etc.

[0115] Knowledge Modeling: Ontology definition, define the ontology of building component model optimization, including entities, relationships, attributes, etc. Entity definition: such as component ontology (beams, columns, plates, walls, etc.), material ontology, design parameter ontology, etc. Relationship definition: such as "composed of", "use material", "meet standard", etc. Attribute definition: such as component size, weight, strength, etc.

[0116] Knowledge Representation: Use structured triples (entity-relation-entity, entity-attribute-attribute value) to represent knowledge. Build the framework of knowledge graph, including nodes (entities) and edges (relationships).

[0117] Knowledge Fusion and Calculation: Knowledge fusion, fusion of data from different sources, eliminating contradictions and ambiguities. Through the integration of information from multiple data sources, a more complete and accurate knowledge system is formed.

[0118] Knowledge Calculation: Quality assessment of fused knowledge to ensure the quality of the knowledge base. Perform necessary reasoning and calculation, such as calculating the bearing capacity and stability of components based on design parameters.

[0119] Knowledge Graph Generation and Optimization: Knowledge graph generation, organize preprocessed and modeled data according to the structure of the knowledge graph, and generate the knowledge graph. Use graph database and other storage technologies to realize persistent storage and efficient query of knowledge.

[0120] Knowledge Graph Optimization: With the progress of the project and the update of data, regularly optimize and update the knowledge graph, collect and process new data, audit and update old data, optimize the structure of the knowledge graph, and ensure the accuracy and efficiency of the query.

[0121] Application and feedback: Apply the built building component model optimization knowledge graph to actual projects to provide support and help for design, construction, optimization, etc. Query and analyze the information in the knowledge graph to assist decision-making and problem solving. Collect user feedback to understand the effectiveness and problems of the knowledge graph in actual application; continuously optimize and improve the construction method and application mode of the knowledge graph according to the feedback.

[0122] The above embodiments can be realized wholly or partially by software, hardware, firmware or any combination thereof. When realized by software, the above embodiments can be realized wholly or partially in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are wholly or partially generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another, for example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center by wired (for example, infrared, wireless, microwave, etc.) mode. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. containing one or more available medium collections. The available medium can be a magnetic medium (for example, floppy disk, hard disk, magnetic tape), an optical medium (for example, DVD) or a semiconductor medium. The semiconductor medium can be a solid-state disk.

[0123] Those of ordinary skill in the art can realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0124] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described system, device and unit can refer to the corresponding processes in the foregoing method embodiments, which will not be described here.

[0125] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other manners. For example, the described device embodiments are merely schematic. For example, the division of the units is only a logical function division. There can be another division manner for the actual implementation, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections between the units can be indirect couplings or communication connections through some interfaces, devices or units, and can be in electrical, mechanical or other forms.

[0126] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, i.e., can be located in one place, or can be distributed on multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the embodiment.

[0127] In addition, each functional unit in the various embodiments of the present application can be integrated into a processing unit, or each unit can be a physically separate unit, or two or more units can be integrated into one unit.

[0128] If the functions are implemented in the form of software function units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application can be embodied in the form of a software product, and the computer software product is stored in a storage medium, and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0129] The above is merely specific embodiments of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method of converting a two-dimensional image of a building component into a three-dimensional model, characterized by: Comprise, Performing quality analysis on source images and obtaining quality data, generating image quality coefficients from image quality data , in accordance with the image quality coefficient The low-quality image is screened out from a plurality of source images, an optimization scheme is matched for the low-quality image, and the low-quality image is optimized by executing the optimization scheme; wherein the contrast and noise density Linear normalization is performed to map the corresponding data values to the interval Inside, the image quality coefficient is generated according to the following formula ： ； weighting factor, ， ； After the optimized source image is taken as the target image, the edges in the target image are recognized, the key points in the target image are extracted, and a semi-global matching algorithm is used to generate a depth map, the depth map is converted into point cloud data in a three-dimensional space, three-dimensional coordinates are obtained from the pixel coordinates and the corresponding depth values, and a point cloud is constructed; A model surface is generated from the point cloud data, the obtained model surface is converted into a polygon mesh model composed of triangular meshes, the noise of the polygon mesh is smoothed after repairing the holes in the polygon mesh, and the initial model of the component is adjusted to obtain a three-dimensional model of the building component; The three-dimensional model size of the component is compared with the size of the actual building component, error data between the two is obtained, and corresponding accuracy is generated from the model error data set , if the accuracy of the three-dimensional model of the component is No more than expected, send model optimization instructions to the outside; Performing coincidence analysis on the three-dimensional model of the component and the building component and obtaining a coincidence degree, generating a complete coefficient from the coincidence degree , according to the complete coefficient Screening out defect areas in the component three-dimensional model, and correlating to obtain an optimization priority of each defect area After the building component model optimization knowledge graph gives a model optimization scheme for the defect area, the optimization priority is used to optimize each defect area in turn.

2. The method of claim 1, wherein: Collecting source images related to two-dimensional drawings of building construction, performing quality analysis on the source images and obtaining quality data, and constructing a quality data set after summarizing; constructing an image quality coefficient from the image quality data set , if the image quality coefficient If the quality is lower than the preset quality threshold, the corresponding image quality is taken as a low-quality image; After the low-quality image is screened, the quality data of the low-quality image is obtained, the quality data is feature-extracted to obtain the corresponding image features, a plurality of image optimization schemes are obtained to generate an image optimization scheme library, and the optimized source image is obtained to send a three-dimensional reconstruction instruction to the outside.

3. The method of claim 2, wherein: After receiving the three-dimensional reconstruction instruction, the optimized source image is obtained and taken as a target image, a Canny edge detection algorithm is used to detect strong edges in the target image, a Sobel operator is applied to detect horizontal and vertical edges in the target image, the continuity of the edges of the target image is optimized after the edges are obtained, and a key point detection instruction is sent to the outside.

4. The method of claim 3, wherein: After receiving the key point detection instruction, a Harris corner detection algorithm is used to detect corner points in the target image, the corner points are taken as key points, a SURF algorithm is used to calculate the descriptors of the key points, a BFMatcher algorithm is used to identify and match the feature points in the left and right images, and a RANSAC algorithm is used to estimate the model parameters by iteratively selecting a random subset to identify and exclude outliers; A disparity map is calculated by finding the best matching pixel block in the left and right images, corresponding processing measures are taken according to the disparity map, a semi-global matching algorithm is used to generate a depth map, and an SGM algorithm is applied to aggregate costs in multiple directions.

5. The method of claim 4, wherein: A Poisson surface reconstruction algorithm is used to generate a smooth and continuous surface from the point cloud data, which is taken as a model surface; a Marching Cubes algorithm is used to extract triangular meshes from the voxel data of the model surface to generate a mesh model composed of triangular meshes, and an implicit surface is converted into an explicit polygon mesh to obtain an initial model of the component.

6. The method of claim 5, wherein: UV mapping is used to assign texture coordinates of the two-dimensional image to each vertex of the initial model of the building component, ensuring accurate alignment and minimal distortion of the texture map on the model surface; a mesh repair tool is used to repair holes in the polygon mesh, and a Taubin smoothing algorithm is applied to smooth the noise in the polygon mesh, and the initial model of the building component is adjusted according to the actual building size to obtain the three-dimensional model of the building component.

7. The method of claim 6, wherein: after obtaining the reconstructed three-dimensional model of the building component, the three-dimensional model is taken as a target model, and after dimension labeling is performed on the target model, labeled dimension data of the target model and dimension data of the actual building component are obtained, the labeled dimensions are compared with the dimensions of the actual building component, error proportion data between the two are obtained, and model error data set is constructed by summarizing.

8. The method of claim 7, wherein: generating a corresponding accuracy from a set of model error data wherein the error proportionality Linear normalization is performed to map the corresponding data values to the interval in the interior, according to the following formula: ； wherein the error proportion on the i-th size-labeled point, for the error proportion of the qualified target value, a standard deviation for a number of error proportions; a weight coefficient: ， , and 。 9. The method of claim 8, wherein: after receiving the model optimization instruction, the target model is divided into a plurality of sub-regions, a plurality of anchor points are selected in the sub-regions, and a random face in the vicinity of the anchor point is selected as a model face; a corresponding face is selected as a standard face at the corresponding position of the actual building component; and the coincidence degree between the standard face and the model face is obtained according to the dimension data between the standard face and the model face.

10. The method of claim 9, wherein: Generating complete coefficients from a degree of coincidence wherein, at the degree of polymerization under the dimensionless condition, according to the following manner: ； weighting factor: ， ； for the degree of coincidence between the i-th standard surface in the sub-region and the model surface, for the qualified standard value of the degree of polymerization, the number of coincidence degrees; if complete coefficient if the integral threshold is not exceeded, the corresponding sub-region is taken as a defect region, and each defect region is marked on the three-dimensional model of the building component.

11. The method of claim 10, wherein: In acquiring the position and the complete coefficient of each defect region Post-correlation acquisition optimization priority , in accordance with the optimization priority each defect region is marked in the following manner: ； wherein, is the first a defect region to a first shortest distance of a defect area, Distance from average; weight factor: ； complete_coefficient for the i-th defect region, for the respective pass standard value, the variance of the integral coefficient.

12. The method of claim 11, wherein: model state data in each defect region is obtained, feature extraction is performed on the position of the model state data in the defect region, and corresponding model optimization features are obtained; building component model optimization is taken as a target word, a building component model optimization knowledge graph is constructed in advance, and model optimization features and model optimization schemes in the defect region are obtained according to the model optimization features.

Images