Live-action three-dimensional building model data multi-level LOD generation method

Abstract

The invention discloses a live-action three-dimensional building model data multistage LOD generation method, and relates to the technical field of three-dimensional model lightweight, and the method comprises the following steps: S1, building multistage LOD: LOD0, LOD1, LOD2 and LOD3 according to the hierarchical visual condition and lightweight degree of a building model, and defining the grading standard of each stage of LOD; s2, according to OSGB data of the building model, extracting a vector surface, namely LOD0, of a target building by adopting an object-oriented method; and S3, based on the vector surface data of the building model, combining with the building elevation data contained in the OSGB, the method is higher in automation degree and simpler to operate, the volume of three-dimensional model data can be reduced, the basic structure of the model under different LOD levels can be kept, conversion between building models can be completed under limited data, and the construction efficiency is improved. And three-dimensional building model monomers with different dimensions are constructed and applied to different levels and scenes, so that the requirements of urban space analysis and management are met.

Description

Technical Field

[0001] The present invention relates to the technical field of lightweight three-dimensional models, and in particular to a multi-level LOD generation method for real-scene three-dimensional building model data. Background Art

[0002] The construction of a real-life 3D China is an important measure to implement the strategy of Digital China and Digital Economy, and is also a specific deployment for the implementation of the national new infrastructure construction. It aims to coordinate at the national level, adopt a unified 3D space benchmark, and establish a high-precision, realistic, and authoritative 3D geospatial database covering the whole country to meet the needs of natural resource management, economic and social development, and other fields for a 3D space public base. The construction of a real-life 3D China, through the transformation and upgrading of traditional basic surveying and mapping services, adopts specialized technical means such as three-dimensional reconstruction, entity modeling, and realistic description, digitizes the relevant geospatial entities within the scope of national space, develops multi-scale, time-series, and related real-life 3D information products, and constructs a real-life 3D digital space to support the construction and application of Digital China and Safe China.

[0003] With the continuous development of real-life 3D construction, the volume of 3D building model data is increasing, which brings challenges to model storage and real-time loading and display. In different application scenarios, the texture details of building models are not necessary. Different application scenarios require different levels of detail of building model textures, so they prefer faster data transmission, smooth real-time loading and display, and lower memory usage. Therefore, for real-life 3D building model data, it is very important to build a multi-level LOD model. The traditional model LOD classification is based on the city geographic markup language (CityGML) for detail level classification. This method is more used in building information modeling (BIM). The disadvantage of this method is that the construction of BIM models requires a lot of manpower and material resources. The simplification of different LOD levels corresponding to different buildings requires a lot of prior knowledge. With the continuous increase of 3D model data, this method cannot meet the needs of subsequent 3D model batch processing. To this end, a multi-level LOD generation method for real-life 3D building model data is proposed. Summary of the invention

[0004] The purpose of the present invention is to solve the problems in the prior art and to propose a multi-level LOD generation method for real-scene three-dimensional building model data.

[0005] A method for generating multi-level LOD of real-scene three-dimensional building model data comprises the following steps:

[0006] S1: According to the visual conditions and lightweight degree of the building model, multiple levels of LOD are established. 0 、LOD 1 、LOD2 、LOD 3 , define the grading standards for each level of LOD;

[0007] S2: Based on the OSGB data of the building model, the vector surface of the target building, i.e. LOD, is extracted using an object-oriented method 0 ;

[0008] S3: Based on the vector surface data of the building model, combined with the building elevation data contained in OSGB, the block building, i.e. LOD, is obtained by vector stretching. 1 ;

[0009] S4: Based on the vector surface data of the building model, the building monomer model (LOD) is obtained by range clipping 3 ;

[0010] S5: Based on the building monomer model data, a topological reconstruction method is used to simplify the building monomer model so that the building monomer model meets the LOD 2 Grading standards.

[0011] Preferably, in step S1, LOD 0 The building structure corresponding to the graded building model is a two-dimensional plane, and the corresponding classification standard is the building vector surface; LOD 1 The building structure corresponding to the graded building model is a block, and the corresponding classification standard is the minimum external volume; LOD 2 The building structure corresponding to the graded building model is differentiated roofs and blocks, and the corresponding grading standard is the actual size of the building; LOD 3 The building structure corresponding to the graded building model includes a real roof, building ancillary structures and building textures, and the corresponding grading standard is that the building model includes external detailed structures.

[0012] Preferably, in step S2, extracting the vector surface of the target building includes the following steps:

[0013] (1) Generate digital surface model (DSM) data and digital orthophoto (DOM) data based on OSGB data of building models;

[0014] (2) Input digital DSM data and DOM data;

[0015] (3) Based on the height information contained in the DSM data, the building objects are separated by threshold segmentation;

[0016] (4) Based on the building objects obtained in step (3), the spectral characteristics of the vegetation are used to separate them, thereby further optimizing the building extraction results;

[0017] (5) Merge all building classification objects;

[0018] (6) Based on the building objects obtained in step (5), non-building objects are eliminated by threshold segmentation according to the size of the building, thereby completing the classification of the building;

[0019] (7) Manually optimize the building extraction results.

[0020] Preferably, in step S3, LOD is generated by vector stretching method. 1 The hierarchical building monomer model includes the following steps:

[0021] (1) performing zoning statistics based on the vector data extracted in step S2 to obtain elevation data of the corresponding building in the DSM data;

[0022] (2) Assign building elevation information corresponding to the vector surface to generate block buildings.

[0023] Preferably, in step S4, the city three-dimensional model data is range-cut to obtain a single building model of LOD3 level.

[0024] Preferably, in step S5, a topology reconstruction method includes the following steps:

[0025] (1) Set the sharp edge angle threshold and convert the angle threshold into radians;

[0026] (2) Set the target side length and store it;

[0027] (3) Set the target number of triangles and store it;

[0028] (4) Build an axis-aligned bounding box tree, create an axis-aligned bounding box for each triangle, and build a bounding box tree;

[0029] (5) Obtain the average edge length of the initial grid;

[0030] (6) Reconstructing the mesh: The mesh reconstruction process is performed according to the specified number of iterations, including the steps of splitting long edges, folding short edges, flipping edges, moving vertices, and projecting vertices;

[0031] (7) Split the long edges, traverse each edge of each face, and if the edge length exceeds the square of the specified threshold, perform the edge splitting operation;

[0032] (8) Fold the short edges. Traverse each edge of each face. If the edge length is less than the square of the minimum threshold and both end points of the edge are not feature points, then perform the edge folding operation.

[0033] (9) Flip the edge. Traverse each edge of each face. If there is an opposite edge, try to flip the edge.

[0034] (10) Move the vertex, update the vertex degree, triangle normal and vertex normal, and adjust the position of each vertex;

[0035] (11) Projecting vertices: For each non-feature vertex, perform ray projection along a bounding box according to its normal direction, find the triangle that intersects with it, and project the vertex to the nearest intersection point.

[0036] Compared with the prior art, the present invention has the following advantages:

[0037] The method of the present invention has a higher degree of automation and is simpler to operate. It can not only reduce the volume of three-dimensional model data, but also maintain the basic structure of the model at different LOD levels. It can complete the conversion between building models under limited data, construct three-dimensional building model monomers of different dimensions, and apply them to different levels and scenes to meet the needs of urban space analysis and management. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a schematic diagram of the process of the present invention.

[0039] Figure 2 This is a schematic diagram of the classification of models at different LOD levels in the present invention.

[0040] Figure 3 LOD in the present invention 0 Schematic diagram of the hierarchical model.

[0041] Figure 4 LOD in the present invention 1 Schematic diagram of the hierarchical model.

[0042] Figure 5 LOD in the present invention 2 Schematic diagram of the hierarchical model.

[0043] Figure 6 LOD in the present invention 3 Schematic diagram of the hierarchical model.

[0044] Figure 7 Schematic diagram of different LOD level models in the present invention. DETAILED DESCRIPTION

[0045] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the present invention is further explained below in conjunction with specific implementation methods.

[0046] Reference Figure 1-7As shown, a method for generating multi-level LOD of real-scene 3D building model data comprises the following steps:

[0047] S1: According to the visual conditions and lightweight degree of the building model, multiple levels of LOD are established. 0 、LOD 1 、LOD 2 、LOD 3 , define the grading standards for each level of LOD;

[0048] S2: Based on the OSGB data of the building model, the vector surface of the target building, i.e. LOD, is extracted using an object-oriented method 0 ;

[0049] S3: Based on the vector surface data of the building model, combined with the building elevation data contained in OSGB, the block building, i.e. LOD, is obtained by vector stretching. 1 ;

[0050] S4: Based on the vector surface data of the building model, the building monomer model (LOD) is obtained by range clipping 3 ;

[0051] S5: Based on the building monomer model data, a topological reconstruction method is used to simplify the building monomer model so that the building monomer model meets the LOD 2 Grading standards.

[0052] In step S1, LOD 0 The building structure corresponding to the graded building model is a two-dimensional plane, and the corresponding classification standard is the building vector surface; LOD 1 The building structure corresponding to the graded building model is a block, and the corresponding classification standard is the minimum external volume; LOD 2 The building structure corresponding to the graded building model is differentiated roofs and blocks, and the corresponding grading standard is the actual size of the building; LOD 3 The building structure corresponding to the graded building model includes a real roof, building ancillary structures and building textures, and the corresponding grading standard is that the building model includes external detailed structures.

[0053] In step S2, extracting the vector surface of the target building includes the following steps:

[0054] (1) Generate digital surface model (DSM) data and digital orthophoto (DOM) data based on OSGB data of building models;

[0055] (2) Input digital DSM data and DOM data;

[0056] (3) Based on the height information contained in the DSM data, the building objects are separated by threshold segmentation;

[0057] (4) Based on the building objects obtained in step (3), the spectral characteristics of the vegetation are used to separate them, thereby further optimizing the building extraction results;

[0058] (5) Merge all building classification objects;

[0059] (6) Based on the building objects obtained in step (5), non-building objects are eliminated by threshold segmentation according to the size of the building, thereby completing the classification of the building;

[0060] (7) Manually optimize the building extraction results.

[0061] In step S3, LOD is generated by vector stretching method. 1 The hierarchical building monomer model includes the following steps:

[0062] (1) performing zoning statistics based on the vector data extracted in step S2 to obtain elevation data of the corresponding building in the DSM data;

[0063] (2) Assign building elevation information corresponding to the vector surface to generate block buildings.

[0064] In the step S4, the city three-dimensional model data is range-cut to obtain a single building model of LOD3 level.

[0065] In step S5, a topology reconstruction method includes the following steps:

[0066] (1) Set the sharp edge angle threshold and convert the angle threshold into radians;

[0067] (2) Set the target side length and store it;

[0068] (3) Set the target number of triangles and store it;

[0069] (4) Build an axis-aligned bounding box tree, create an axis-aligned bounding box for each triangle, and build a bounding box tree;

[0070] (5) Obtain the average edge length of the initial grid;

[0071] (6) Reconstructing the mesh: The mesh reconstruction process is performed according to the specified number of iterations, including the steps of splitting long edges, folding short edges, flipping edges, moving vertices, and projecting vertices;

[0072] (7) Split the long edges, traverse each edge of each face, and if the edge length exceeds the square of the specified threshold, perform the edge splitting operation;

[0073] (8) Fold the short edges. Traverse each edge of each face. If the edge length is less than the square of the minimum threshold and both end points of the edge are not feature points, then perform the edge folding operation.

[0074] (9) Flip the edge. Traverse each edge of each face. If there is an opposite edge, try to flip the edge.

[0075] (10) Move the vertex, update the vertex degree, triangle normal and vertex normal, and adjust the position of each vertex;

[0076] (11) Projecting vertices: For each non-feature vertex, perform ray projection along a bounding box according to its normal direction, find the triangle that intersects with it, and project the vertex to the nearest intersection point.

[0077] The working process and principle of the present invention are as follows:

[0078] A1: Perform LOD classification on the building model, namely LOD 0 、LOD 1 、LOD 2 、LOD 3 .

[0079] A2: Prepare oblique 3D model (OSGB) data, load tiles, select tiles by range, and output DOM and DSM data;

[0080] A3: Import DOM and DSM data generated by the oblique photography model;

[0081] A4: Based on the DSM and DOM RGB three-channel data, multi-scale segmentation is performed on the pixel layer, with the scale set to 95, and the shape and compactness weights set to 0.4 and 0.5 respectively;

[0082] A5: Set the elevation mean range to be greater than or equal to 30, and classify the building objects through the algorithm, that is, set the threshold condition to the classification algorithm of Mean DSM>=30, and set it to the building category in the category;

[0083] A6: Optimize the classification objects and separate the vegetation targets using their unique spectral characteristics. The characteristics of vegetation in the green channel are amplified by constructing the green channel ratio feature, i.e., green_ratio = mean of green / (mean of red + mean of green + mean of blue). The rule: "mean of green / (mean of red + mean of green + mean of blue)>=0.36" is used to remove vegetation objects that meet this feature from the building classification objects.

[0084] A7: Building objects obtained by merging segmentation classifications through algorithms;

[0085] A8: Compare the building objects obtained after step A7, set the optimal threshold value>=5200 pixels according to the size of the building area, and perform the classification algorithm to separate the non-building objects with smaller area from the building category;

[0086] A9: Manually edit some incompletely classified building objects and remove some non-building objects that are mistakenly classified as building objects, and finally get the LOD 0 Hierarchical building vector faces;

[0087] A10: Import building vector data and DSM raster data;

[0088] A11: Open the Zonal Statistics tool, input the building vector surface, summary field and DSM raster data, execute to get the zonal statistics table, associate the zonal statistics results with the building vector surface, and add the average elevation field to the building vector surface;

[0089] A12: Import building vector surface data with average elevation field added, perform stretching operation, and generate LOD 1 Graded building blocks;

[0090] A13: Based on the building vector surface data obtained in A9, the OSGB oblique photography data is range-cut to obtain the LOD 3 Level building single model data;

[0091] A14: Calculate the included angle of the edge;

[0092] A15: Set the scaling factor for converting angle to radians;

[0093] A16: Converts the angle value into radians for subsequent judgment and calculation;

[0094] A17: Set the target triangle side length for use in the subsequent mesh reconstruction process;

[0095] A18: Set the target number of triangles for use in the subsequent mesh reconstruction process;

[0096] A19: Create a vector that stores the axis-aligned bounding box. Its size is the same as the number of target triangles. Each element is an object representing the axis-aligned bounding box.

[0097] A20: Traverse each triangle, add the triangle data to the corresponding axis-aligned bounding box, and update the center point of the bounding box;

[0098] A21: Create a vector to store face indices, add the index of each triangle to the vector, the size of which is the same as the number of target triangles;

[0099] A22: Create a total axis-aligned bounding box to store the total axis-aligned bounding box of all triangles combined;

[0100] A23: Updates the center point of the axis-aligned bounding box, deletes the old axis-aligned bounding box tree, creates a new axis-aligned bounding box tree, and passes in the axis-aligned bounding box of each triangle, the triangle index vector, and the total axis-aligned bounding box as parameters;

[0101] A24: Define a constructor that accepts two parameters, a pointer to a vertex vector and a pointer to a pointer to a triangle index.

[0102] A25: Initialize the pointer to the vertex vector to the parameter value of the constructor;

[0103] A26: Initialize the pointer to the triangle index vector to the constructor parameter value;

[0104] A27: Initializes the mesh instance using the two parameters in A25 and A26, which point to the actual data of the vertex and triangle index respectively;

[0105] A28: Calculate and store the initial average edge length for subsequent use in the mesh reconstruction process;

[0106] A29: Get the average side length of the initial grid;

[0107] A30: Delete the old triangle normal vector data, create a new triangle normal vector data storage space, calculate the normal vector of each triangle and store it;

[0108] A31: If the target number of triangles is set, calculate the total area of ​​the current mesh, calculate the area of ​​each target triangle, and calculate the target side length based on the target triangle area;

[0109] A32: Calculate the minimum and maximum target side lengths, and the squares of the minimum and maximum target side lengths;

[0110] A33: Create an axis-aligned bounding box tree for the current vertices and triangles;

[0111] A34: If the sharp edge threshold is set, split the long edge, update the triangle normal vector, and mark the sharp edge. Otherwise, split the field edge and mark the boundary feature.

[0112] A35: Loop through steps A36-A40 according to the specified number of iterations.

[0113] A36: Split edges that are too long and merge edges that are too short;

[0114] A37: Flip edges that may improve mesh quality;

[0115] A38: Adjust vertex position;

[0116] A39: Project vertices inside axis-aligned bounding box;

[0117] A40: traverse every face in the mesh;

[0118] A41: Get the starting half of the current front edge;

[0119] A42: traverse each half edge in front of the current one;

[0120] A43: Get the next half edge of the current half edge;

[0121] A44: Use the vertex position information to calculate the square of the length of the current half edge and the next half edge;

[0122] A45: If the square of the length exceeds the square of the maximum allowed side length, split the side where the current half side is located;

[0123] A46: After the segmentation is completed, jump out of the current loop and continue to process the next half edge;

[0124] A47: Get the reconstructed half-edge mesh;

[0125] A48: Traverse each face in the mesh;

[0126] A49: If the face has been marked for removal, skip processing;

[0127] A50: Get the index of the current face and the starting half edge of the face;

[0128] A51: Traverse all half edges of the current face starting from the starting half edge;

[0129] A52: Get the next half edge and calculate the square of the distance between the starting vertex of the current half edge and the next half edge;

[0130] A53: If the length of the current half edge and the next half edge is less than the square of the minimum allowed length, enter the conditional judgment;

[0131] A54: Determine whether the starting vertex of the current half edge and the starting vertex of the next half edge are not feature vertices;

[0132] A55: If the condition of A54 is met, try to fold the current half edge, and if successful, exit the current loop;

[0133] A56: Continue to process the next half edge, loop until all the previous half edges are processed, and end the loop when returning to the starting half edge;

[0134] A57: Move to the next face;

[0135] A58: Traverse each face in the grid and get the starting half edge of the current face;

[0136] A59: Traverse all half edges of the current face starting from the starting half edge;

[0137] A60: Get the next half edge of the current half edge;

[0138] A61: Determine whether the current half edge can be flipped, that is, whether there is an opposite half edge;

[0139] A62: Check whether the flipping operation can be performed. If the flipping is successful, exit the current half-edge traversal;

[0140] A63: Move to the next half edge and loop until all the half edges of the current one have been checked;

[0141] A64: Move to the next face;

[0142] A65: Update the degree information of the vertex, update the normal vector of the triangle, and update the normal vector of the vertex;

[0143] A66: Loop condition, check if the vertex pointer is not empty, and adjust and optimize the position of the current vertex;

[0144] A67: The iteration ends, get the pointer of the next vertex, and loop until all vertices of the current face are processed;

[0145] A68: Traverse each vertex in the mesh;

[0146] A69: If the vertex is a feature point, skip it and get the first half edge of the vertex. If the vertex has no half edge, skip it.

[0147] A70: Create a bounding box and update its position to the position of the current vertex;

[0148] A71: Starting from the first half-edge of the current vertex, traverse the surrounding half-edges and update the position of the bounding box to include all relevant vertices;

[0149] A72: Create a bounding box tree;

[0150] A73: Test whether the bounding box of the current vertex intersects with the bounding boxes in the entire bounding box tree, and record the intersecting triangle index pairs;

[0151] A74: stores the intersection points and their distance information;

[0152] A75: Calculate the length of the ray segment of vertex movement;

[0153] A76: Traverse each pair of intersecting ray bounding boxes and triangle bounding boxes;

[0154] A77: Get the vertex position of the current triangle;

[0155] A78: Calculate the intersection of the ray and the triangle plane;

[0156] A79: Calculate the normal vector at the intersection point;

[0157] A80: Check the direction of the normal vector and select the intersection point that meets the conditions;

[0158] A81: If there is an intersection point that meets the conditions, select the intersection point with the smallest distance as the new position of the vertex;

[0159] A82: The loop ends and continues to process the next vertex.

[0160] In summary, compared with the LOD grading of BIM models, the method of the present invention has a higher degree of automation and is simpler to operate. It solves the defects of excessive consumption of human and material resources, complex implementation process, long time consumption and difficulty in batch processing in the LOD grading process of building model data in the prior art. It can not only reduce the volume of three-dimensional model data, but also maintain the basic structure of the model at different LOD levels. It can complete the conversion between building models under limited data, construct three-dimensional building model monomers of different dimensions, and apply them to different levels and scenes to meet the needs of urban space analysis and management.

[0161] It is known from common technical knowledge that the present invention can be implemented by other embodiments that do not deviate from its spirit or essential features. Therefore, the above disclosed embodiments are only illustrative in all respects and are not exclusive. All changes within the scope of the present invention or within the scope equivalent to the present invention are included in the present invention.

Claims

1. A multi-level LOD generation method for real-life 3D building model data. Features: The following steps are involved: S1: According to the visual conditions and lightweight degree of the building model, multiple levels of LOD are established. 0 、LOD 1 、LOD 2 、LOD 3 , define the grading standards for each level of LOD; S2: Based on the OSGB data of the building model, the vector surface of the target building, i.e. LOD, is extracted using an object-oriented method 0 ; S3: Based on the vector surface data of the building model, combined with the building elevation data contained in OSGB, the block building, i.e. LOD, is obtained by vector stretching. 1 ; S4: Based on the vector surface data of the building model, the building monomer model (LOD) is obtained by range clipping 3 ; S5: Based on the building monomer model data, a topological reconstruction method is used to simplify the building monomer model so that the building monomer model meets the LOD 2 Grading standards.

2. A method for generating multi-level LOD of real-life 3D building model data according to claim 1, Features: In step S1, LOD 0 The building structure corresponding to the graded building model is a two-dimensional plane, and the corresponding classification standard is the building vector surface; LOD 1 The building structure corresponding to the graded building model is a block, and the corresponding classification standard is the minimum external volume; LOD 2 The building structure corresponding to the graded building model is differentiated roofs and blocks, and the corresponding grading standard is the actual size of the building; LOD 3 The building structure corresponding to the graded building model includes a real roof, building ancillary structures and building textures, and the corresponding grading standard is that the building model includes external detailed structures.

3. A method for generating multi-level LOD of real-life 3D building model data according to claim 1, Features: In step S2, extracting the vector surface of the target building includes the following steps: (1) Generate digital surface model (DSM) data and digital orthophoto (DOM) data based on OSGB data of building models; (2) Input digital DSM data and DOM data; (3) Based on the height information contained in the DSM data, the building objects are separated by threshold segmentation; (4) Based on the building objects obtained in step (3), the spectral characteristics of the vegetation are used to separate them, thereby further optimizing the building extraction results; (5) Merge all building classification objects; (6) Based on the building objects obtained in step (5), non-building objects are eliminated by threshold segmentation according to the size of the building, thereby completing the classification of the building; (7) Manually optimize the building extraction results.

4. The method for generating multi-level LOD of real-scene three-dimensional building model data according to claim 1, Features: In step S3, LOD is generated by vector stretching method. 1 The hierarchical building monomer model includes the following steps: (1) performing zoning statistics based on the vector data extracted in step S2 to obtain elevation data of the corresponding building in the DSM data; (2) Assign building elevation information corresponding to the vector surface to generate block buildings.

5. The method for generating multi-level LOD of real-scene 3D building model data according to claim 1, Features: In the step S4, the city three-dimensional model data is range-cut to obtain a single building model of LOD3 level.

6. A method for generating multi-level LOD of real-life 3D building model data according to claim 1, Features: In step S5, a topology reconstruction method includes the following steps: (1) Set the sharp edge angle threshold and convert the angle threshold into radians; (2) Set the target side length and store it; (3) Set the target number of triangles and store it; (4) Build an axis-aligned bounding box tree, create an axis-aligned bounding box for each triangle, and build a bounding box tree; (5) Obtain the average edge length of the initial grid; (6) Reconstructing the mesh: The mesh reconstruction process is performed according to the specified number of iterations, including the steps of splitting long edges, folding short edges, flipping edges, moving vertices, and projecting vertices; (7) Split the long edges, traverse each edge of each face, and if the edge length exceeds the square of the specified threshold, perform the edge splitting operation; (8) Fold the short edges. Traverse each edge of each face. If the edge length is less than the square of the minimum threshold and both end points of the edge are not feature points, then perform the edge folding operation. (9) Flip the edge. Traverse each edge of each face. If there is an opposite edge, try to flip the edge. (10) Move the vertex, update the vertex degree, triangle normal and vertex normal, and adjust the position of each vertex; (11) Projecting vertices: For each non-feature vertex, perform ray projection along a bounding box according to its normal direction, find the triangle that intersects with it, and project the vertex to the nearest intersection point.

Images