Urban building lightweight body frame model generation method based on three-dimensional model

Abstract

The invention discloses an urban building lightweight body frame model generation method based on a three-dimensional model, and the method comprises the following operation steps: collecting urban original point cloud data, and carrying out the rasterization of the urban original point cloud data to generate a DSM image; cutting the DSM image by using a digital line drawing to obtain a local DSM image and an outer contour of the local DSM image; performing pixel resampling on the local DSM image to obtain a top surface point cloud of a local building; preprocessing the top surface point cloud of the local building to obtain a plurality of local parts; constructing a body frame model of each building component through the plurality of local components; and combining the body frame models of all the building components to generate a complete body frame model of the building.

Description

Technical Field

[0001] The invention relates to the field of building models, and in particular to a method for generating a lightweight frame model of an urban building based on a three-dimensional model. Background Art

[0002] Building individualization is a very important part of the concept of new basic surveying and mapping, and is also the basis for the semanticization of urban information.

[0003] There are multiple forms of expression of urban building monomerization. There is the cutting monomerization that directly cuts the building in the urban tilted model using the model cutting method through the building's outer contour, and there is also manual monomerization that produces the monomer model of the building through manual operation, three-dimensional editing software, and combined with the urban three-dimensional model; however, the above methods all have obvious disadvantages: cutting monomerization has high operating efficiency and can be fully automated and produced in large-scale parallel, but the result of cutting monomerization essentially has the same resolution as the tilted model, resulting in a relatively large amount of data for a single building. When applied to city-level scenes, the amount of monomerized building data in the entire city will often reach TB or even PB levels. Such a data level is a huge load for both storage management and loading demonstration; manual monomerization keeps the details intact, and the geometric structure can be simplified to the greatest extent to compress the data volume, and is widely used in various display platforms, but due to the low degree of automation of manual monomerization, it will generate a large amount of labor cost burden, often requiring a large number of practitioners to invest for a long time, resulting in a prolonged production cycle.

[0004] According to the requirements of LOD1.0, LOD1.3, and LOD2.0 building monomerization results proposed in the "Work Plan for Rapid Construction of Urban 3D Models" of the Department of Land Surveying and Mapping, the generation method of lightweight body frame models (LOD1.0, LOD1.3, and LOD2.0) of urban buildings has gradually become a focus technology in the field of surveying and mapping. The main problems of traditional model simplification methods are that the simplification intensity is difficult to control (QEM, Quadric Error Metrics) and the simplified features are lost (Laplace smoothing simplification algorithm). Summary of the invention

[0005] The purpose of the present invention is to provide a method for generating a lightweight frame model of an urban building based on a three-dimensional model, which solves the above-mentioned technical problems pointed out in the prior art.

[0006] The present invention provides a method for generating a lightweight frame model of an urban building based on a three-dimensional model, comprising the following steps:

[0007] Collecting original point cloud data of the city, and generating a DSM image by rasterizing the original point cloud data of the city;

[0008] The DSM image is cropped using a digital line drawing to obtain a local DSM image and an outer contour of the local DSM image;

[0009] The local DSM image is pixel-resampled to obtain a top surface point cloud of a local building; the top surface point cloud of the local building is preprocessed to obtain a plurality of local components;

[0010] constructing a body frame model of each building component by using the plurality of local components;

[0011] The body frame model of all building components is combined to generate a complete body frame model of the building.

[0012] Preferably, the original point cloud data of the city is collected, and the DSM image is generated by rasterizing the original point cloud data of the city. The specific operation steps are as follows:

[0013] Collecting original point cloud data of the city, preprocessing the original point cloud data of the city, and generating a three-dimensional model of the city;

[0014] Extracting elevation data of the three-dimensional city model through a grid to extract the coordinate position of each vertex point cloud in the three-dimensional city model; performing two-dimensional mapping of the coordinate position of each vertex point cloud extracted by the grid to obtain a two-dimensional grid space;

[0015] Assigning an elevation value to each grid in the two-dimensional grid space, and composing the elevation value of each grid into DSM data;

[0016] The DSM data is saved in an image format to obtain a DSM image.

[0017] Preferably, the DSM image is cropped using a digital line drawing to obtain a local DSM image and an outer contour of the local DSM image. The specific operation steps are as follows:

[0018] Determine a plurality of polygonal buildings using a digital line drawing on the DSM image, and clip the local DSM image of each polygonal building to obtain a clipped local DSM image; and determine the outer contour of the local DSM image of each clipped polygonal building, and clip the outer contour of the local DSM image to obtain the outer contour of the clipped local DSM image;

[0019] The local DSM image of each polygonal building after clipping is matched with the outer contour of the local DSM image after clipping, and an engineering-organized local engineering file organization is formed and saved;

[0020] Extracting basic information of each building from the local engineering file organization;

[0021] The basic information of each building includes: horizontal area, maximum elevation, and minimum elevation.

[0022] Preferably, the local DSM image is pixel resampled to obtain a top surface point cloud of a local building; the top surface point cloud of the local building is preprocessed to obtain a plurality of local components, and the specific operation steps are as follows:

[0023] Resampling pixels of the local DSM image of each polygonal building, extracting top surface elevation data of the local DSM image of each polygonal building, and converting the data into a top surface point cloud of each local building;

[0024] Using a point cloud outlier filtering method and a point cloud bilateral filtering method to remove isolated points and denoise the top surface point cloud;

[0025] The preprocessed top surface point cloud of each local building is segmented using a dual clustering method to obtain multiple local components;

[0026] The dual clustering method includes: an elevation clustering method and a spatial clustering method.

[0027] Preferably, the pre-processed top surface point cloud of each local building is segmented by using a dual clustering method to obtain multiple local components. The specific operation steps are as follows:

[0028] Computing elevation differences of all point clouds in the top surface point cloud of the local building by means of an elevation clustering method, clustering them according to the elevation differences and the elevation difference threshold, and obtaining first clustered point clouds of multiple elevation segments;

[0029] The Euclidean distance between the first cluster point clouds of each elevation segment is calculated according to the spatial clustering method, and clustering is performed by judging the Euclidean distance and the spatial distance threshold to obtain a plurality of second cluster point clouds as a plurality of local building components.

[0030] Preferably, the body frame model of each building component is constructed by using the multiple local components, and the specific operation steps are as follows:

[0031] Determine the outer contour of the local building component according to the point cloud of the local component; compare the area of ​​the point cloud of the local component with the outer contour of the local building component to obtain the base of the local building component; determine the outer contour of the base of the local building component according to the base of the local building component;

[0032] Sampling the outer contour of the base of the local building component to obtain virtual facade points of the local building; calculating the energy function of the virtual facade points of the local building to construct a mixed linear model;

[0033] constructing a body-frame model of each building component by means of the mixed linear model;

[0034] Preferably, the body frame model of all building components is merged to generate a body frame model of a complete building. The specific operation steps are as follows: all local building components are calculated using spatial Boolean operation and merged to generate a body frame model of a complete building.

[0035] Compared with the prior art, the embodiments of the present invention have at least the following technical advantages:

[0036] By analyzing the above-mentioned method for generating a lightweight body frame model of an urban building based on a three-dimensional model provided by the present invention, it can be known that in specific applications, the original point data of the city can be collected from multiple angles through the method of oblique photography, and the original point data of the city can be pre-processed by denoising, filtering, etc. to establish a three-dimensional model of the city; the three-dimensional model of the city is rasterized, the coordinate position of each vertex point cloud in the three-dimensional model of the city is extracted, mapped to a two-dimensional grid space, divided into grids, each grid is assigned an elevation value, and different grids can be compared and analyzed through the elevation values, thereby forming DSM data, and the DSM data is saved in image format to obtain a DSM image;

[0037] Furthermore, the DSM image uses a digital line drawing to determine multiple polygonal buildings, and the local DSM image of each polygonal building is cut out, and each local DSM image can be analyzed separately to improve the data processing speed; and the outer contour corresponding to each local DSM image is cut out; the local DSM image and the outer contour corresponding to each local DSM image can be analyzed and calculated separately, and operated independently to improve the speed of subsequent operation steps;

[0038] Furthermore, the local DSM image is pixel resampled, and the local DSM image top surface elevation data of each polygonal building is extracted and converted into the top surface point cloud of each local building. The top surface point cloud excludes the point cloud data of the facade points through the elevation clustering method and the spatial clustering method. Only the top surface point cloud can be operated separately to obtain multiple clustered elevation segments, which are confirmed as multiple local building components through multiple elevation segments. Each local building component can be operated separately to improve the calculation speed and efficiency.

[0039] Further, multiple local building components are sorted from low to high according to the elevation segment, and the overall horizontal area of ​​the building is calculated by the horizontal projection area and the outer contour of each local building component to determine which local building components are used as the base, and the main direction of the overall building is determined by the base; at the same time, the outer contour of the base of the local building component is determined by the base of the local building component, and the virtual facade of the local building component is determined by the outer contour of the base of the local building component. After the top surface point cloud is operated, the virtual facade of the local building component is operated, thereby reducing the operation time and independent calculations do not interfere with each other; the virtual facade of the local building component is regionally grown by the RegionGrow algorithm to obtain a rough initial result, and the rough initial result is used to fit the plane equation to merge, and the new plane equation is obtained by the calculation of the RegionGrow algorithm, which is a new segmentation scheme, improving the efficiency and accuracy of subsequent processing; the new plane equation is used as a space segmentation scheme to calculate the energy function, and a mixed linear model is constructed. Solving the mixed linear model can achieve global optimal optimization to determine the optimal spatial segmentation that best fits the original building shape, thereby obtaining a body frame model of each building component;

[0040] Furthermore, through the spatial Boolean operation in three-dimensional space, the body frame models of each building component are merged to eliminate any overlap or internal gaps, thereby generating a complete body frame model of the building. For buildings with more complex structural features, the use of a complete body frame model of the building can fully retain the basic features of the top of the building for the overall view of the urban building, making the display more consistent with the original data and reducing distortion; at the same time, it also reduces the involvement of manpower and saves labor costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0042] Figure 1 An overall flow chart of a method for generating a lightweight frame model of an urban building based on a three-dimensional model provided in the first embodiment of the present invention;

[0043] Figure 2 A flow chart of constructing a body frame model of a method for generating a lightweight body frame model of an urban building based on a three-dimensional model provided in the first embodiment of the present invention;

[0044] Figure 3A schematic diagram of a body frame model (left) compared with an original inclined mesh model (right) of a method for generating a lightweight body frame model of an urban building based on a three-dimensional model provided in Example 1 of the present invention;

[0045] Figure 4 A schematic diagram of a comparison between a frame model of multiple buildings (left) and an original inclined mesh model (right) of a method for generating a lightweight frame model of an urban building based on a three-dimensional model provided in Example 1 of the present invention;

[0046] Figure 5 A schematic diagram of a complete body frame model in a large scene of a method for generating a lightweight body frame model of an urban building based on a three-dimensional model provided in the first embodiment of the present invention;

[0047] Figure 6 A schematic diagram comparing a body frame model of a method for generating a lightweight body frame model of an urban building based on a three-dimensional model provided in Embodiment 1 of the present invention with a traditional model;

[0048] Figure 7 A flowchart of a method for generating a lightweight frame model of an urban building based on a three-dimensional model provided in Embodiment 2 of the present invention. DETAILED DESCRIPTION

[0049] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] The present invention is further described in detail below through specific embodiments in conjunction with the accompanying drawings.

[0051] Embodiment 1

[0052] like Figure 1 As shown, the present invention proposes a method for generating a lightweight frame model of an urban building based on a three-dimensional model, comprising the following steps:

[0053] S10: collecting original point cloud data of the city, and generating a DSM image by rasterizing the original point cloud data of the city;

[0054] It should be noted that the original urban point cloud data is collected by oblique photography and rasterized. The rasterized data can be converted into discrete point clouds. The X and Y coordinates of the converted point clouds are arranged regularly. In subsequent operations, whether it is outlier filtering or point cloud clustering, the rasterized X and Y coordinates can improve the calculation speed during the calculation process;

[0055] S20: cropping the DSM image using a digital line drawing to obtain a local DSM image and an outer contour of the local DSM image;

[0056] It should be noted that the local DSM of the building is cut out from the original scene. In the city-level geographical scene, for the convenience and optimization of processing, it is necessary to cut the local DSM image of each building out from the overall DSM image using a digital image. The local DSM image of each building and its own outer contour will not interfere with each other when executing subsequent operation steps, thereby improving the computational efficiency of the steps.

[0057] S30: resampling the local DSM image pixels to obtain a top surface point cloud of a local building; preprocessing the top surface point cloud of the local building to obtain a plurality of local components (i.e., local buildings);

[0058] It should be noted that the pixel resampling method is used to convert DSM data into discrete point cloud data. At this time, the point cloud data obtained is the data of most facade points filtered out. The reason for converting it into discrete point cloud is that the DSM data with images as carriers has insufficient utilization of three-dimensional information. It needs to be converted into a three-dimensional point cloud to increase the speed of obtaining multiple local components.

[0059] S40: constructing a body frame model of each building component (the building component here can also be understood as being equivalent to the aforementioned local component) through the multiple local components;

[0060] It should be noted that by clustering and optimizing the point clouds of multiple local components, a body frame model is generated for each individual local component, which greatly reduces manual participation and retains the top surface features of urban buildings as much as possible;

[0061] S50: The body frame models of all building components are combined to generate a complete body frame model of the building.

[0062] Specifically, in step S10, original point cloud data of the city is collected, and a DSM image is generated by rasterizing the original point cloud data of the city. The specific operation steps are as follows:

[0063] S11: collecting original point cloud data of the city, preprocessing the original point cloud data of the city, and generating a three-dimensional model of the city;

[0064] rasterizing the three-dimensional urban model to generate a DSM image;

[0065] It should be noted that the original urban point data can be collected by oblique photography (i.e., oblique photography can capture data from multiple angles) (i.e., the original urban point cloud data is stored in the form of mesh data, especially mainly in the form of triangular mesh (TriMesh), and the point cloud coordinate information of the vertices of urban buildings and the surface information composed of vertex indexes are recorded in the original urban point cloud data, generally in the form of OBJ files, OSGB files, etc. as data carriers); the original urban point data is pre-processed by point cloud filtering, denoising, and classification (i.e., classification and identification of the ground, buildings, and vegetation), and the pre-processed original urban point cloud data is mainly used as an oblique three-dimensional model (i.e., a comprehensive urban three-dimensional model including the ground, buildings, and vegetation is constructed using the original urban point cloud data captured from multiple angles);

[0066] Rasterize the city 3D model (i.e., the grid DSM (Digital Surface Model) is a gridded digital surface model, which generally uses images such as tiff or GeoTiff to store elevation information as data carriers), extract the elevation information of the city 3D model, extract the coordinate position of each vertex point cloud in the city 3D model, and map it to a two-dimensional grid space (divide the three-dimensional space into a series of small squares (grids), and assign an elevation value to each grid. The elevation value can be the average, maximum, minimum, or some form of interpolation result of the elevation of all points in the grid);

[0067] Save the rasterized elevation data in an image format, such as TIFF or GeoTIFF. In summary, the city 3D model can be converted into a discrete point cloud after rasterization. The XY coordinates of the converted point cloud are arranged regularly. In subsequent operations, whether it is outlier filtering or point cloud clustering, the rasterized point cloud XY coordinates can improve the subsequent calculation speed;

[0068] Specifically, in step S11, the three-dimensional city model is rasterized to generate a DSM image. The specific operation steps are as follows:

[0069] S111: extracting elevation data (i.e., elevation data can be visualized to help intuitively understand terrain features and point cloud features) of the three-dimensional city model through a grid, extracting the coordinate position of each vertex point cloud in the three-dimensional city model; performing two-dimensional grid mapping on the coordinate position of each vertex point cloud extracted to obtain a two-dimensional grid space;

[0070] It should be noted that the three-dimensional model of the city is rasterized, the elevation data is extracted, the coordinate position of each vertex point cloud is obtained, and then the coordinate position of each vertex point cloud is mapped to a two-dimensional grid (that is, the three-dimensional model of the city in the three-dimensional space is mapped to a two-dimensional grid space of small squares (that is, the small squares are grids)), and a two-dimensional grid space is obtained. Discretizing the three-dimensional space of the three-dimensional model of the city into a grid form can significantly reduce the amount of data that needs to be processed;

[0071] S112: assigning an elevation value to each grid in the two-dimensional grid space, and composing DSM data with the elevation value of each grid;

[0072] The DSM data is saved in an image format to obtain a DSM image;

[0073] It should be noted that assigning an elevation value to each grid can standardize the elevation data in two dimensions, so that different grids can be compared and analyzed; the elevation value can be the average, maximum, minimum or interpolated value of the elevation of all points in the grid;

[0074] Specifically, in step S20, the DSM image is cropped using a digital line drawing to obtain a local DSM image and an outer contour of the local DSM image. The specific operation steps are as follows:

[0075] S21: using a digital line drawing to determine a plurality of polygonal buildings on the DSM image, and clipping the local DSM image of each polygonal building to obtain a clipped local DSM image; and determining the outer contour of the local DSM image of each clipped polygonal building, and clipping the outer contour of the local DSM image to obtain the outer contour of the clipped local DSM image;

[0076] It should be noted that in order to facilitate the subsequent analysis and processing of the DSM image, a digital line graph is used for each individual polygonal building in the DSM image (i.e., a digital line graph, DLG is a vector data format that uses points, lines, and polygons to represent geographic features such as roads, boundaries, buildings, etc.; DSM images use digital line graphs to find polygonal buildings, and can also be used to define the boundaries of each polygonal building);

[0077] After the local DSM image of each polygonal building is cut out, each individual polygonal building can be processed and analyzed independently, which can significantly improve the speed and efficiency of data processing; at the same time, the outer contour of the local DSM image of each building cut out can also be used for subsequent independent operations, and the processing of a certain building will not affect other buildings, thereby ensuring the accuracy and consistency of data processing; and the local DSM image of each polygonal building and the outer contour of the local DSM image will not interfere with each other in subsequent operations, and can be operated independently, thereby improving the speed of subsequent operations;

[0078] S22: matching the clipped local DSM image of each polygonal building with the outer contour of the clipped local DSM image, and forming an engineering-organized local engineering file organization, and saving it;

[0079] Extracting basic information of each building from the local engineering file organization;

[0080] The basic information of each building includes: horizontal area, maximum elevation, minimum elevation;

[0081] It should be noted that in step S21, the outer contour of the local DSM image of each polygonal building is clipped to obtain the outer contour of each local DSM image; the local DSM image of each polygonal building is matched with the outer contour of the local DSM image, and the local DSM image of each corresponding polygonal building and the outer contour of the local DSM image are formed into a local engineering file organization by engineering arrangement (that is, the local engineering file organization is to create a separate folder or data set for each building, which contains the local DSM image (such as GeoTIFF file) and local outer contour (such as Shapefile file) of the building, and the files are named so that the buildings to which they belong can be quickly identified), so that the image data of each building is independently stored and managed, which is convenient for subsequent access and processing;

[0082] At the same time, the basic information of each building, such as horizontal area, maximum elevation, and minimum elevation, can be further extracted through local engineering file organization; basic information is crucial for subsequent analysis and processing (such as calculating the volume of the building, analyzing the terrain adaptability of the building, etc.);

[0083] Specifically, in step S30, the local DSM image is pixel-resampled to obtain a top surface point cloud of a local building; the top surface point cloud of the local building is preprocessed to obtain a plurality of local components. The specific operation steps are as follows:

[0084] S31: resampling pixels of the local DSM image of each polygonal building, extracting top surface elevation data of the local DSM image of each polygonal building, and converting the data into a top surface point cloud of each local building;

[0085] It should be noted that the DSM image is obtained by saving the DSM data in an image format, so the local DSM image can obtain the local DSM data; each grid in the local DSM data is a potential three-dimensional point cloud (that is, the reason for converting into a discrete point cloud is that the local DSM data with the local DSM image as the carrier has insufficient utilization of the information of the three-dimensional model, and it needs to be converted into a three-dimensional point cloud), and the top surface point cloud is created according to its X, Y coordinates and the stored elevation value (that is, the generated top surface point cloud is the top surface point cloud that has been removed of duplicate points, filtered (that is, the point cloud data of most of the building's facade points have been filtered out) (such as Gaussian filtering, median filtering) to reduce noise, and the point cloud density has been adjusted).

[0086] S32: removing isolated points and denoising the top surface point cloud by using a point cloud outlier filtering method and a point cloud bilateral filtering method;

[0087] S33: performing component segmentation on the preprocessed top surface point cloud of each local building using a dual clustering method to obtain multiple local components;

[0088] The dual clustering method includes: elevation clustering method and spatial clustering method;

[0089] It should be noted that the pre-processed top surface point cloud of each local building is clustered by using the elevation clustering method and the spatial clustering method, thereby splitting the local building components of the polygonal building top surface;

[0090] The elevation clustering method is the process of clustering elevation data (i.e., the height or altitude data of a point above the ground or sea level); dividing the data into different groups or clusters based on the similarity or distance between data objects, so that the data points in the same group are more similar or related;

[0091] Spatial clustering methods are used to group or cluster objects or point clouds in geographic space according to their spatial distribution relationships. The objects within these groups (or "clusters") show high similarity in spatial location, while objects between different groups are as different as possible.

[0092] Specifically, in step S32, the top surface point cloud is subjected to the removal of isolated points and denoising by using the point cloud outlier filtering method and the point cloud bilateral filtering method. The specific operation steps are as follows:

[0093] S321: removing isolated points in the top surface point cloud by using a point cloud outlier filtering method;

[0094] S322: Preset a neighborhood range K, calculate the distance between each point cloud in the top surface point cloud and the other point clouds in the neighborhood range K; calculate the average distance through the distance of each point cloud; calculate the square of the difference between the distance of each point cloud and the average distance to obtain the distribution information of the top surface point cloud;

[0095] It should be noted that the preset neighborhood range K (i.e., the neighborhood range K is usually K=10 by default) is a distance range value, and the distance between each point cloud and the other point clouds within the neighborhood range K is calculated; then the average distance is calculated, and the average distance can reflect the density of the surrounding environment of each point cloud; then the square of the difference between the distance of each point to its K neighboring points and the average distance is calculated, and the square of the difference can understand the distribution information of the point cloud;

[0096] S323: Calculate the mean error of the distribution information of the top surface point cloud; preset an isolated point threshold e, and determine whether the mean error is greater than the threshold;

[0097] If so, the point cloud is marked as an isolated point and filtered and deleted;

[0098] If not, the point cloud is retained;

[0099] It should be noted that the mean error (i.e., the mean error refers to a statistic that measures the difference between the median and the median of a data set and is used to evaluate the degree of dispersion of the data) is calculated based on the distribution information of the point cloud (i.e., the square of the difference of the point cloud). In this embodiment, the median absolute deviation (or similar statistic) of the square of the distance difference of all points is calculated, and the isolated point threshold is preset to distinguish between normal points and isolated points.

[0100] S324: Calculate the average distance d0 of each point cloud in the top surface point cloud using the preset neighborhood range K, calculate the normal vector of each point cloud, and use the point cloud within the neighborhood range K of each point cloud to fit the plane S 2 , the calculation formula using the point cloud bilateral filtering method is:

[0101]

[0102] Where: N(P) is the neighborhood of the neighborhood range K of the point cloud P, P i ∈N(P), P' is the point cloud P in S 2 projection on; is the normal vector of point cloud P, It is the point cloud P i The normal vector, W c and W s They are different standard deviations σc and σ s Gaussian kernel function, representing the spatial domain weight and feature domain weight respectively;

[0103] S325: Preset a noise threshold r, and calculate whether the d value of each point cloud is greater than the noise threshold r;

[0104] If yes, the point cloud is marked as noise and the noise is deleted;

[0105] If not, the point cloud is retained;

[0106] It should be noted that the distance or error d value is calculated for each point cloud P according to bilateral filtering; the point cloud d is compared with the preset noise threshold r (generally set to 0.15 by default). If the d value of the point cloud is greater than the threshold, the point cloud is considered to be a noise point and is deleted from the point cloud.

[0107] Bilateral filtering can effectively remove noise while maintaining edge and detail information in the point cloud by considering both spatial proximity and feature similarity.

[0108] Specifically, in step S33, the pre-processed top surface point cloud of each local building is segmented using a dual clustering method to obtain multiple local components. The specific operation steps are as follows:

[0109] Computing elevation differences of all point clouds in the top surface point cloud of the local building by means of an elevation clustering method, clustering them according to the elevation differences and the elevation difference threshold, and obtaining first clustered point clouds of multiple elevation segments;

[0110] The Euclidean distance between the first cluster point clouds of each elevation segment is calculated according to the spatial clustering method, and clustering is performed by judging the Euclidean distance and the spatial distance threshold to obtain multiple second cluster point clouds as multiple local building components. The specific operation steps are as follows:

[0111] S331: randomly selecting a point cloud from the top surface point cloud of the local building as a seed point by using the elevation clustering method;

[0112] Traverse the elevation difference between each point cloud and the current seed point in the top surface point cloud of the local building;

[0113] S332: Preset an elevation difference threshold t, and determine whether the elevation difference of the current seed point is less than the elevation difference threshold t;

[0114] If not, the point cloud will not be processed;

[0115] If yes, cluster the point cloud with the current seed point to obtain different first cluster point clouds of the elevation segment;

[0116] Iterate the above process until the elevation difference of the current seed point is greater than or less than the elevation difference threshold t (generally set to 2.0 by default), and then stop clustering;

[0117] It should be noted that, in this embodiment, the elevation segment can be understood as an area or distance with a specific height range or elevation difference in the vertical direction; elevation refers to the height or altitude of a certain point above the ground or sea level.

[0118] S333: randomly selecting seed points for the remaining point clouds of the top surface point clouds of the local buildings, and repeating steps S331 and S332 until all point clouds are clustered;

[0119] It should be noted that the elevation clustering method is a process of classifying the preprocessed polygonal building top surface point cloud. This process is mainly based on the elevation of each point in the point cloud (i.e., their vertical position) to cluster together points with similar elevations to distinguish the top surfaces of different parts of the building.

[0120] When all points are marked and assigned to the corresponding clusters, the elevation clustering process ends; at this point, the top surface point cloud is divided into multiple clusters, each cluster represents the top surface of a local building component (i.e., polygonal building) on ​​an elevation segment;

[0121] By effectively segmenting the top surfaces of different components in the polygonal building top surface point cloud, it provides a basis for subsequent processing (such as Euclidean distance clustering in spatial clustering);

[0122] S334: Calculate the Euclidean distance between the first cluster point clouds of each elevation segment by a spatial clustering method, and preset a spatial distance threshold y (i.e., generally set to 0.5 by default);

[0123] Determine whether the Euclidean distance between the first cluster point clouds of each elevation segment is less than the spatial distance threshold y;

[0124] If not, the first cluster point cloud of the non-compliant elevation segment will not be processed;

[0125] If yes, cluster the first cluster point clouds of the matching elevation segments to obtain multiple second cluster point clouds;

[0126] treating the plurality of second clustered point clouds as a plurality of local building components;

[0127] It should be noted that the first cluster point cloud of the elevation segment after elevation clustering and then spatial clustering is divided into multiple second cluster point clouds, and each second cluster point cloud is regarded as a local building component. Thus, the building is divided into multiple components;

[0128] Specifically, Figure 2As shown, in step S40, a body frame model of each building component is constructed by using the multiple local components. The specific operation steps are as follows:

[0129] Determine the outer contour of the local building component according to the point cloud of the local component; compare the area of ​​the point cloud of the local component with the outer contour of the local building component to obtain the base of the local building component; determine the outer contour of the base of the local building component according to the base of the local building component;

[0130] Sampling the outer contour of the base of the local building component to obtain virtual facade points of the local building; calculating the energy function of the virtual facade points of the local building to construct a mixed linear model;

[0131] The body-frame model of each building component is constructed by the mixed linear model, and the specific operation steps are as follows:

[0132] S41: determining the outer contour of the local building component through the point cloud of the local building component determined by the second clustered point cloud;

[0133] The plurality of local building components are sorted from low to high according to the elevation segments; the horizontal projection area of ​​each local building component is calculated in a bottom-up order; and the overall horizontal area of ​​the building is calculated according to the outer contours of the local building components;

[0134] Calculate the area ratio of the overall horizontal area of ​​the building by calculating the horizontal projection area of ​​each local building component and the outer contour of the local building component; preset a ratio threshold u (i.e., generally set to 0.125 by default); and determine whether the area ratio is greater than or equal to the ratio threshold u;

[0135] If so, the local building components that meet the conditions are determined to be the base, and the other local building components that do not meet the conditions are not processed;

[0136] If not, all local building components do not meet the judgment conditions, then a virtual horizontal plane point cloud is constructed for the local building components in the lowest elevation segment as a base;

[0137] It should be noted that each building component must first be judged to determine whether the component is the base of the building. The base of a building is different from other components. It represents the main structural features of the building and the main direction of the building as a whole. Therefore, one component must be selected from multiple building components as the base of the entire building, and then the main directions of other building components are aligned to it.

[0138] The point cloud converted from DSM is regularly arranged on the XY axis and has no repeated superposition on the Z axis, so its horizontal projection area can be easily calculated through the point cloud and resolution. Compared with random discrete point clouds, which can only obtain the outer contour through topological algorithms and then calculate the horizontal area of ​​polygons, the computational efficiency of this method is greatly improved, thus reflecting the advantage of converting mesh models into DSM data and then into point clouds;

[0139] By calculating the area ratio of the overall horizontal area of ​​the building by the horizontal projection area of ​​each local building component and the outer contour of the local building component and the difference threshold, it is judged whether there is a base of the building; if there is no base, the local building component with the lowest elevation segment is taken as the base;

[0140] S42: using the outer contour of the partial building component as the outer contour of the base of the partial building component;

[0141] The local building components that do not meet the conditions as the base are horizontally projected through the point cloud, and the projections are arranged on the X and Y axes; the AlphaShape algorithm is performed on the horizontally projected point cloud to extract the boundary points of the horizontally projected point cloud, and the extracted boundary points are connected to form the outer contour of the local building components that do not meet the conditions as the base;

[0142] It should be noted that the outer contour of each local building component will be generated directly. Since the body frame model of the building component needs to construct a virtual facade, it is necessary to extract the outer contour of each local building component; instead of the base local building component, the point cloud of the local building component is first horizontally projected onto the XY axis, and then the AlphaShape algorithm is performed on the plane point cloud to extract the boundary points to form the outer contour of the local building component.

[0143] S43: determining the main direction of the building through the outer contour of the base of the local building component, and performing polygon optimization on the main direction of the building;

[0144] S44: uniformly differencing the outer contour of the base of the local building component into line segments; setting a sampling interval (i.e., the default setting in this embodiment is 0.25), vertically sampling the lowest elevation segment of the polygonal building from bottom to top (i.e., in step S41, the multiple local building components have been sorted from low to high according to the elevation segment, so the multiple local building components of the polygonal building are vertically sampled from bottom to top), and obtaining virtual facade points of the local building;

[0145] Merging the sampling points of the local building component (i.e., the virtual facade point cloud) with the top surface point cloud of the local building to obtain a virtual facade of the local building component;

[0146] It should be noted that in step S31, the facade points of the polygonal building have been filtered out, and the three-dimensional urban model can be quickly rasterized to increase the speed, and the calculation time for obtaining local building components of the polygonal building is also reduced;

[0147] Therefore, when it is necessary to construct a body frame model, it is necessary to complete the facade information of the local building components. The method adopted in this embodiment is to use the outer contour of the base of the local building components to vertically sample to obtain a virtual facade point cloud, and then merge it with the top surface point cloud of the polygonal building to form a complete polygonal building component point cloud (i.e., top surface plus virtual facade).

[0148] S45: performing region growth on the virtual facade of the local building component by using a RegionGrow algorithm to obtain an initial result; and optimizing an energy function of the initial result;

[0149] S46: fitting a plane equation using the point cloud in the area of ​​each plane area segmentation result to obtain a plurality of fitted plane equations; comparing the fitted plane equations in pairs to determine whether they are similar; if they are similar, merging the two fitted plane equations to obtain a new plane equation;

[0150] It should be noted that each region obtained by region segmentation is fitted into a plane equation, and the plane equations are compared pairwise to determine whether two plane equations are similar (that is, the similarity can be determined by calculating the distance difference between the two plane equations). All two similar plane equations are determined, and the regions represented by these plane equations are merged into a larger region. The merged region will correspond to a new plane equation, and a more concise and effective segmentation scheme is obtained, which improves the efficiency and accuracy of subsequent processing;

[0151] For example, there is a set of comparisons, point set a (i.e., the result of plane area segmentation) and plane A (i.e., the fitted plane equation), as well as point set b and plane B. The points in a are projected onto plane B, and the proportion Ra of points whose distance is less than ε (generally set to 0.1) is counted. Similarly, the proportion Rb of points in b whose distance to A is less than ε is counted. If both Ra and Rb are greater than the threshold (generally set to 80%), the two are merged.

[0152] S47: Take each new plane equation as a space partitioning scheme, intersect the new plane equations in order, and obtain new candidate surfaces as candidate solutions; calculate the energy function of each candidate surface, and the calculation formula is:

[0153] E={ω c ·E c ,ω f ·E f,ω o ·E o}

[0154] ω c +ω s +ω o =1;

[0155] Where E is the energy function of the candidate surface, which is an array of three values, ω c ,ω f ,ω o There are three weight coefficients, representing the aggregation weight, coverage weight, and complexity weight respectively; ω c ,ω f ,ω o Corresponding to the respective type energy function E c 、E f 、E t ;

[0156] Energy function E c 、E f 、E t The calculation formulas are:

[0157]

[0158] Where n v is the number of vertices of the candidate face, z i is the z coordinate of the current vertex;

[0159]

[0160] Where r is the point cloud resolution, The area of ​​the current candidate surface;

[0161]

[0162] Where n e is the number of edges of the candidate face, l e is the length of the current edge, is the angle between the current edge and the next adjacent edge;

[0163] S48: Select the candidate face, select the edge of the candidate face for the candidate face, select the corresponding dihedral angle structure to build a model, and obtain a mixed linear model;

[0164] S49: solving the mixed linear model using an open source mathematical solver to obtain an optimal value of a binary unknown number;

[0165] Determining whether all candidate faces are to be retained according to the optimal value of the binary unknown number, performing topological intersection on the retained candidate faces, and obtaining a final body frame model of each building component;

[0166] It should be noted that the mixed linear model can be solved by the SCIP solver to obtain the optimal value of each binary unknown. The optimal value of the binary unknown represents the decision of whether each candidate face is retained in the mixed linear model. Therefore, the optimal value of the binary unknown can determine which candidate faces are necessary and which ones can be discarded when constructing the building component frame model.

[0167] Topological intersection is a geometric calculation method for determining the intersection between two or more geometric shapes. In this embodiment, the retained candidate faces are regarded as geometric shapes, and a topological intersection algorithm is used to calculate the common boundaries and regions between them.

[0168] Specifically, in step S43, the main direction of the building is determined by the outer contour of the base of the local building component, and polygon optimization is performed on the main direction of the building. The specific operation steps are as follows:

[0169] S431: splitting the outer contour of the base of the local building component into a plurality of segment segments, calculating the normal vector n of each segment segment, and clustering the normal vectors according to the same direction to obtain the distribution of the segment segments in the same direction; presetting a direction distribution threshold i (i.e., generally set to 30% by default); determining whether the distribution of the segment segments in the same direction is greater than the direction distribution threshold i;

[0170] If yes, the distribution of the same direction of the transverse line segment is determined as the main direction of the building;

[0171] If not, no processing is performed;

[0172] And calculated by the formula:

[0173]

[0174] <n i ,n j >≤ε∨| <n i ,n j > -π|≤ε;

[0175] Where ε is the angle threshold, which is generally set to 10; π is a constant; N is the set of normal vectors of the intercept segment; nj and ni represent the normal vectors of adjacent intercept segments i and j;

[0176] S432: Preset an angle threshold o (i.e., 15° is generally used by default), and determine whether the angle between the cut-off line segment and the vertical direction of the main direction of the building is less than the angle threshold o;

[0177] If yes, rotate the segment around the midpoint to the main direction of the building or its perpendicular direction; and intersect all the segment in the initial order to obtain a new regularized polygon (i.e., the edge optimization of the building);

[0178] It should be noted that the outer contour of the base of the local building component is regularized into a polygon according to the overall main direction, and as many segment segments as possible are made parallel or perpendicular to the main direction.

[0179] Specifically, in step S45, the virtual facade of the local building component is subjected to region growing by using the RegionGrow algorithm to obtain an initial result; the energy function of the initial result is optimized, and the specific operation steps are as follows:

[0180] S451: performing region growth on the virtual facade of the local building component by using the RegionGrow algorithm, randomly selecting a point cloud as a seed point for each virtual facade of the local building component; growing the neighborhood by using the seed point to obtain an initial result;

[0181] The initial test results are optimized through iterations to obtain multiple plane region segmentation results;

[0182] It should be noted that the initial results of the RegionGrow algorithm are often relatively rough and may contain some noise, over-segmentation or under-segmentation problems. The rough initial results are optimized and iterated to make the initial results closer to accurate and detailed planar region segmentation results.

[0183] S452: Using the energy function to determine whether the plane region segmentation result meets the optimization condition, the calculation formula is:

[0184] E(p)=ω d ·E d +ω f ·E f +ω t ·E t

[0185] ω d +ω f +ω t =1;

[0186] Where E(p) is the energy function of point cloud p; ω d ,ω f ,ω t They are three weight coefficients respectively; ω d ,ω f ,ω t The corresponding energy functions are E d 、E f 、Et ;

[0187] Energy function E d 、E f 、E t The calculation formulas are:

[0188]

[0189] Where N(p) is the point cloud set of the region to which point cloud p belongs; n is the number of point clouds in the region; S 2 It is the plane fitted by the point cloud in this area; is a point cloud p in the set i The corresponding probability density weight in Gaussian distribution;

[0190]

[0191] In the formula, is the distance between two point clouds; is the average point spacing of point cloud p; n is the number of point clouds in the region;

[0192]

[0193] In the formula, It is a point cloud i The normal vector of ; n is the number of point clouds in the region; It is a point cloud The normal vector of

[0194] When each iteration fails to reduce the energy function, the plane region segmentation result reaches the optimization condition:

[0195] S453: recalculate the normal vector of each point cloud; and calculate the moving distance of the point cloud on the normal vector according to the energy function of each point cloud, and the calculation formula is:

[0196]

[0197] Where E(p) is the energy function of point cloud p, S 2 is the plane fitted by the point cloud in the area; Δ is the moving distance;

[0198] Then use guided filtering to change the coordinates of the point cloud along the normal vector:

[0199]

[0200] In the formula, is the coordinate of the point cloud after moving on the normal vector; It is a point cloud The normal vector of ; Δ is the moving distance;

[0201] S454: voting on the plane region segmentation result, calculating the average energy function of all point clouds in the region of point cloud P in the above formula, and calculating the average energy function of the neighborhood region of point cloud P;

[0202] Compare the average energy function of the region of point cloud P with the average energy function of the neighboring region. When the average energy function of one region is lower than the average energy function of the other region, the average energy function of the lower region is attributed to the average energy function of the higher region, and the updated region is obtained.

[0203] It should be noted that the "voting" mentioned above is not a majority vote in the traditional sense, but refers to calculating a certain statistic or characteristic value of all points in the area; by calculating the average energy function value, the algorithm can obtain the overall fitting quality of the area; at the same time, by calculating the average energy function value of all point clouds in the regional neighborhood of point cloud P, the fitting quality of the neighborhood can be evaluated;

[0204] Compare the average energy function value of the area where point p is located with the average energy function value of its neighborhood. If the average energy function value of point p in a certain area in the neighborhood is lower (i.e. the fitting quality is better); if a point cloud has a better fitting effect in a certain area in the neighborhood, then it is more likely to belong to which area;

[0205] S455: re-iteratively calculating the energy function for all point clouds in the updated area; presetting an iteration threshold p (ie, generally 5-7 times by default);

[0206] When the number of iterations is greater than the iteration threshold p, the iteration is stopped and the process returns to step S452 to iterate again.

[0207] When the sum of the energy functions of all point clouds in the updated area is greater than the sum of the energy functions before iteration, the iteration is stopped and the process returns to step S452 to iterate again;

[0208] It should be noted that the purpose of setting the iteration stop condition is to prevent the algorithm from looping indefinitely and to ensure that the algorithm gives results within a reasonable time. The sum of all point energy functions after the update is greater than the sum of the energy functions before the iteration. This is a more flexible stopping criterion, which allows the iteration to be stopped when the energy function is no longer significantly reduced, helping to avoid unnecessary calculations.

[0209] Specifically, in step S48, the candidate face is selected, the edge of the candidate face is selected for the candidate face, and the corresponding dihedral angle structure is selected to construct a model to obtain a mixed linear model. The specific operation steps are as follows:

[0210] S481: Select the candidate surface (i.e., the selection principle is to use the above equation to solve the best candidate surface), select the edge of the candidate surface for the candidate surface, and select the corresponding dihedral angle structure; calculate the unknown solution through the candidate surface, the edge of the candidate surface and the corresponding dihedral angle structure;

[0211] It should be noted that these three components have strong constraints on each other (called hard constrains in the mixed linear model). For example, if alternative surface a is selected in the final solution, then each edge of the alternative surface should be selected. If two adjacent alternative surfaces a and b are selected, then the dihedral structure formed by them should also be selected. The reason for establishing so many strong constraints is to make the final settlement result of the mixed linear model watertight (water tide, that is, the model is sealed). The consequence of the lack of strong constraints is that although the energy function of the final result is the lowest value, the final result is fragmented. Therefore, in this embodiment, three types of unknown solutions are established at the same time.

[0212] S482: Let a binary array X of type bool be:

[0213]

[0214] in, represents the binary solution of the candidate surface, is the binary solution of the edge of the candidate face, is the binary solution of the dihedral angle, n f is the number of candidate faces, n e is the number of adjacent edges;

[0215] S483: Create an adjacency list that expresses the adjacency relationship between each candidate face and the common edges of adjacent faces; first traverse the adjacency relationship and add the coefficient corresponding to each adjacent dihedral angle. coefficient The average of the third term of the energy function of the two candidate faces corresponding to the dihedral angle; add a row to the mixed linear model:

[0216]

[0217] Then traverse each candidate face and add the candidate face coefficient Add the second term of the energy function for each candidate face:

[0218]

[0219] Traverse each candidate face again, add the candidate face coefficient, and add the first term of the energy function of each face:

[0220]

[0221] Add the above three rows of coefficients as floating-point items to the model, and then add constraints;

[0222] Traverse the adjacency relationship and add a row of coefficients as constraint coefficients:

[0223] P=[1.01.0…1.01.000…0000…00];

[0224] Then traverse each adjacency relationship, and in each adjacency relationship, traverse each candidate face, and construct a row of constraint coefficients for each candidate face, in the form of the above formula, and make a judgment on each candidate face and the adjacency relationship;

[0225] If the candidate face involves the adjacency relationship, a coefficient is assigned (i.e., a coefficient of 2.0 is generally assigned by default);

[0226] If not, a coefficient is assigned (i.e., a coefficient of -1.0 is generally assigned by default);

[0227] This iteration generates a total of n f ·n e Constraint coefficients;

[0228] At this point, the construction of the mixed linear model is completed;

[0229] It should be noted that, in this embodiment, solving the mixed linear model can achieve global optimal optimization to determine the optimal spatial segmentation that best fits the original building shape, thereby obtaining a body frame model of each building component, which is clearer than the traditional original inclined mesh model and reduces distortion, such as Figure 3 As shown, Figure 4 As shown;

[0230] Specifically, Figure 5 As shown, in step S50, the body frame models of all building components are merged to generate a complete body frame model of the building. The specific operation steps are as follows:

[0231] All local building parts are calculated using spatial Boolean union operations to merge into a complete building;

[0232] Finally, a complete building frame model is generated;

[0233] It should be noted that, through the spatial Boolean operation in three-dimensional space, multiple geometric bodies (i.e., local building components) are merged into a single geometric body (i.e., a complete building). This new geometric body contains all the volumes and boundaries of the original geometric bodies, while eliminating any overlaps or internal gaps. The resulting body frame model of the complete building is less distorted than the traditional model, is clearer, and reduces labor costs, such as Figure 6 As shown;

[0234] Embodiment 2

[0235] See also Figure 7 , Example 2 of the present invention proposes a method for generating a lightweight frame model of an urban building based on a three-dimensional model, and its main overview processing logic refers to this figure.

[0236] Specifically, it includes: first, collecting the original point cloud data of the city for preprocessing to generate a three-dimensional model of the city;

[0237] Rasterize the city 3D model to generate DSM images;

[0238] The DSM image is cropped using a digital line drawing to obtain a local DSM image and an outer contour of the local DSM image;

[0239] Extract the top surface point cloud of each local building from the local DSM image;

[0240] The top surface point cloud of each local building is pre-processed to remove noise and isolated points, and multiple local components (i.e. building components) are obtained;

[0241] Determining whether a plurality of local components can be constructed into a base of a local building component;

[0242] If yes, then obtain the outer contour of the base of the constructed partial building component to obtain the outer contour of the base of the partial building component;

[0243] If not, construct a virtual horizontal plane point cloud as a base for the local building component, obtain the outer contour of the base of the local building component, determine the main direction of the building based on the outer contour of the base of the local building component, and perform polygon optimization on the main direction of the building;

[0244] Sampling the outer contour of the base of the local building component to obtain the virtual facade of the local building component; calculating the energy function of the virtual facade of the local building component to obtain the body frame model of each building component;

[0245] Merge the body frame models of all building components to finally obtain a complete body frame model of the building;

[0246] In summary, the method for generating a lightweight body frame model of an urban building based on a three-dimensional model proposed in the example of the present invention can be known that the original point data of the city can be collected from multiple angles by the method of oblique photography, and the original point data of the city can be pre-processed by denoising and filtering to establish a three-dimensional model of the city; the three-dimensional model of the city is rasterized to extract the coordinate position of each vertex point cloud in the three-dimensional model of the city, and mapped to a two-dimensional grid space, divided into grids, each grid is assigned an elevation value, and different grids can be compared and analyzed by the elevation values, thereby forming DSM data, and the DSM data is saved in an image format to obtain a DSM image;

[0247] Furthermore, the DSM image uses a digital line drawing to determine multiple polygonal buildings, and the local DSM image of each polygonal building is cut out, and each local DSM image can be analyzed separately to improve the data processing speed; and the outer contour corresponding to each local DSM image is cut out; the local DSM image and the outer contour corresponding to each local DSM image can be analyzed and calculated separately, and operated independently to improve the speed of subsequent operation steps;

[0248] Furthermore, the local DSM image is pixel resampled, and the local DSM image top surface elevation data of each polygonal building is extracted and converted into the top surface point cloud of each local building. The top surface point cloud excludes the point cloud data of the facade points through the elevation clustering method and the spatial clustering method. Only the top surface point cloud can be operated separately to obtain multiple clustered elevation segments, which are confirmed as multiple local building components through multiple elevation segments. Each local building component can be operated separately to improve the calculation speed and efficiency.

[0249] Further, multiple local building components are sorted from low to high according to the elevation segment, and the overall horizontal area of ​​the building is calculated by the horizontal projection area and the outer contour of each local building component to determine which local building components are used as the base, and the main direction of the overall building is determined by the base; at the same time, the outer contour of the base of the local building component is determined by the base of the local building component, and the virtual facade of the local building component is determined by the outer contour of the base of the local building component. After the top surface point cloud is operated, the virtual facade of the local building component is operated, thereby reducing the operation time and independent calculations do not interfere with each other; the virtual facade of the local building component is regionally grown by the RegionGrow algorithm to obtain a rough initial result, and the rough initial result is used to fit the plane equation to merge, and the new plane equation is obtained by the calculation of the RegionGrow algorithm, which is a new segmentation scheme, improving the efficiency and accuracy of subsequent processing; the new plane equation is used as a space segmentation scheme to calculate the energy function, and a mixed linear model is constructed. Solving the mixed linear model can achieve global optimal optimization to determine the optimal spatial segmentation that best fits the original building shape, thereby obtaining a body frame model of each building component;

[0250] Furthermore, through the spatial Boolean operation in three-dimensional space, the body frame models of each building component are merged to eliminate any overlap or internal gaps, thereby generating a complete body frame model of the building. For buildings with more complex structural features, the use of a complete body frame model of the building can fully retain the basic features of the top of the building for the overall view of the urban building, and is more consistent with the original data in the display, with less distortion and clearer than traditional models; at the same time, it also reduces the involvement of manpower and saves labor costs.

[0251] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. A person skilled in the art may modify the technical solutions described in the above embodiments, or replace part or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for generating a lightweight frame model of an urban building based on a three-dimensional model, characterized in that: The steps are as follows: Collecting original point cloud data of the city, and generating a DSM image by rasterizing the original point cloud data of the city; The DSM image is cropped using a digital line drawing to obtain a local DSM image and an outer contour of the local DSM image; The local DSM image is pixel-resampled to obtain a top surface point cloud of a local building; the top surface point cloud of the local building is preprocessed to obtain a plurality of local components; constructing a body frame model of each building component by using the plurality of local components; The body frame model of all building components is combined to generate a complete body frame model of the building.

2. The method for generating a lightweight frame model of an urban building based on a three-dimensional model according to claim 1, characterized in that: The original point cloud data of the city is collected, and a DSM image is generated by rasterizing the original point cloud data of the city. The specific operation steps are as follows: Collecting original point cloud data of the city, preprocessing the original point cloud data of the city, and generating a three-dimensional model of the city; Extracting elevation data of the three-dimensional city model through a grid to extract the coordinate position of each vertex point cloud in the three-dimensional city model; Performing a two-dimensional grid mapping on the coordinate position of each vertex point cloud extracted to obtain a two-dimensional grid space; Assigning an elevation value to each grid in the two-dimensional grid space, and composing the elevation value of each grid into DSM data; The DSM data is saved in an image format to obtain a DSM image.

3. The method for generating a lightweight frame model of an urban building based on a three-dimensional model according to claim 2, characterized in that: The DSM image is cropped using a digital line drawing to obtain a local DSM image and an outer contour of the local DSM image. The specific operation steps are as follows: Determine a plurality of polygonal buildings using a digital line drawing on the DSM image, and crop the local DSM image of each polygonal building to obtain a cropped local DSM image; and determining the outer contour of the local DSM image of each clipped polygonal building, clipping the outer contour of the local DSM image to obtain the outer contour of the clipped local DSM image; The local DSM image of each polygonal building after clipping is matched with the outer contour of the local DSM image after clipping, and an engineering-organized local engineering file organization is formed and saved; Extracting basic information of each building from the local engineering file organization; The basic information of each building includes: horizontal area, maximum elevation, and minimum elevation.

4. The method for generating a lightweight frame model of an urban building based on a three-dimensional model according to claim 3, characterized in that: The local DSM image is pixel resampled to obtain a top surface point cloud of a local building; the top surface point cloud of the local building is preprocessed to obtain a plurality of local components. The specific operation steps are as follows: Resampling pixels of the local DSM image of each polygonal building, extracting top surface elevation data of the local DSM image of each polygonal building, and converting the data into a top surface point cloud of each local building; Using a point cloud outlier filtering method and a point cloud bilateral filtering method to remove isolated points and denoise the top surface point cloud; The preprocessed top surface point cloud of each local building is segmented using a dual clustering method to obtain multiple local components; The dual clustering method includes: an elevation clustering method and a spatial clustering method.

5. The method for generating a lightweight frame model of an urban building based on a three-dimensional model according to claim 4, characterized in that: The preprocessed top surface point cloud of each local building is segmented using the dual clustering method to obtain multiple local components. The specific steps are as follows: Computing elevation differences of all point clouds in the top surface point cloud of the local building by means of an elevation clustering method, clustering them according to the elevation differences and the elevation difference threshold, and obtaining first clustered point clouds of multiple elevation segments; The Euclidean distance between the first cluster point clouds of each elevation segment is calculated according to the spatial clustering method, and clustering is performed by judging the Euclidean distance and the spatial distance threshold to obtain a plurality of second cluster point clouds as a plurality of local building components.

6. The method for generating a lightweight frame model of an urban building based on a three-dimensional model according to claim 5, characterized in that: The body frame model of each building component is constructed by using the multiple local components. The specific operation steps are as follows: Determine the outer contour of the local building component according to the point cloud of the local component; compare the area of ​​the point cloud of the local component with the outer contour of the local building component to obtain the base of the local building component; determine the outer contour of the base of the local building component according to the base of the local building component; Sampling the outer contour of the base of the local building component to obtain virtual facade points of the local building; calculating the energy function of the virtual facade points of the local building to construct a mixed linear model; A body-frame model of each building component is constructed by the mixed linear model.

7. A method for generating a lightweight frame model of an urban building based on a three-dimensional model according to claim 6, characterized in that: The body frame model of all building components is merged to generate a complete body frame model of the building. The specific operation steps are as follows: all local building components are calculated using spatial Boolean operation and merged to generate a complete body frame model of the building.

Images