Building roof elevation extraction method and system based on city modeling

Abstract

The invention discloses a building roof elevation extraction method and system based on city modeling, and the method comprises the steps: S1, point cloud data and image data fusion registration: extracting data features, and carrying out alignment registration; s2, roof denoising and boundary optimization: identifying interference information, extracting a roof boundary, and generating a roof digital surface model through interpolation; s3, roof elevation extraction: for a flat roof, extracting the top elevation of the flat roof through a plane fitting method; for the slope top roof, extracting the top elevation of the slope top roof through an extreme point detection method; and extracting the top elevation of the roofs except the flat roofs and the slope roofs through a point cloud clustering method. According to the scheme, the roof elevation can be rapidly and accurately extracted, and high-precision three-dimensional data support is provided for urban planning, energy assessment and disaster emergency; the problems that in the prior art, roof elevation extraction efficiency is low, complex roof structure recognition precision is insufficient, and multi-source data fusion is difficult are solved.

Description

Technical Field

[0001] The present invention relates to the technical fields of geographic information systems and 3D modeling technologies, and particularly relates to a method and a system for extracting building roof elevations. Background Art

[0002] Urban modeling is to simulate and represent the urban geographical space, buildings, infrastructure, etc. through 3D digital technologies, and is widely used in fields such as urban planning, disaster management, and smart city construction. During the implementation of operations, it usually has steps such as data collection, data processing and modeling, quality inspection and release. For urban modeling, in the steps of data processing and modeling, the extraction of the elevation of the building roof is an important work content.

[0003] In traditional urban modeling, the extraction of the elevation of the building roof mainly relies on manual or semi-automatic tools, with low efficiency and high costs. Some existing methods have poor adaptability to complex roof structures (such as pitched roofs, multi-level roofs, etc.) based on elevation extraction of LiDAR or photogrammetry, and it is difficult to fuse multi-source data (such as remote sensing images, point cloud data, vector maps, etc.), resulting in insufficient model accuracy. In addition, there is a lack of a dynamic update mechanism for roof elevation extraction, making it difficult to reflect the real-time changes of urban roofs in a timely manner. Summary of the Invention

[0004] The purpose of the present invention is to provide a method and a system for extracting building roof elevations based on urban modeling, which can at least solve the problem of insufficient model accuracy.

[0005] According to the first aspect of the present invention, there is provided a method for extracting building roof elevations based on urban modeling, which includes: S1. Fusion registration of point cloud data and image data: Extract corner points and edge features in the point cloud data; extract texture features of the image data; optimize the rigid body transformation parameters based on singular value decomposition to align the corner points, edge features and texture features, so as to align and register the point cloud data and the image data; combine the image data with the point cloud data to extract multi-modal features; S2. Roof denoising and boundary optimization: Identify and remove interference information from the point cloud data; based on a convolutional neural network, construct a point cloud map and apply a graph convolutional operator to capture the dependency relationships and underlying connection patterns between points, and extract the roof boundary; perform interpolation processing on the point cloud data based on an adaptive interpolation algorithm to generate a continuous roof digital surface model; S3. Roof elevation extraction: For flat roofs, extract their top elevations through a plane fitting method; for pitched roofs, extract their top elevations through an extreme point detection method; for roofs other than flat roofs and pitched roofs, extract their top elevations through a point cloud clustering method.

[0006] According to the building roof elevation extraction method based on urban modeling, in S1, corner points and edge features in the point cloud data are extracted through the SIFT feature extraction algorithm, and texture features of the image data are extracted using local binary patterns.

[0007] According to the building roof elevation extraction method based on urban modeling, the steps of extracting corner points and edge features in the point cloud data through the SIFT feature extraction algorithm are as follows: Define a three-dimensional Gaussian kernel function: where, is the point coordinate in three-dimensional space, is the scale parameter (controlling the width of the Gaussian kernel); Calculate the difference of Gaussian in three dimensions: Calculate the difference between Gaussian-blurred point clouds at adjacent scales: where, is the scaling factor between adjacent scales; Find the extreme points in the scale space: Find local extreme points in the three-dimensional scale space. For each point and scale , compare its value with the values of all adjacent points in its neighborhood; If is a local maximum or minimum, it is marked as a corner point and an edge feature.

[0008] According to the building roof elevation extraction method based on urban modeling, the steps of extracting texture features of the image data using local binary patterns are as follows: Calculate the binary encoding: where: is the number of neighborhood pixels (such as ), is the neighborhood radius (such as ), is the sequence number of the pixel in the image; Count the frequencies of the LBP values of all pixels in the image to generate a histogram: where, is the number of pixels with the LBP value of , is the sequence number of the pixel in the image; Normalize the histogram to obtain the texture feature vector.

[0009] According to the method for extracting the elevation of building roofs based on urban modeling, in S1, the alignment of corner points, edge features, and texture features by optimizing the rigid body transformation parameters based on singular value decomposition includes: Performing rough registration of the corner points, edge features of the point cloud data, and the texture features of the image data through rigid body transformation; optimizing the rigid body transformation parameters through singular value decomposition to perform fine registration of the corner points, edge features of the point cloud data, and the texture features of the image data; Among them, the steps for optimizing the rigid body transformation parameters through singular value decomposition are as follows: Calculating the centroid of the source point cloud and the target point cloud : Subtracting each point by the corresponding centroid to obtain the centered coordinates: Calculating the covariance matrix: Performing singular value decomposition: Among them, and are orthogonal matrices, is a diagonal matrix; The optimal rotation matrix is: The optimal translation vector is: By iterating the above steps, gradually optimize the rigid body transformation parameters and , until converging to the optimal solution.

[0010] According to the method for extracting the elevation of building roofs based on urban modeling, in S2, identifying and removing interference information from the point cloud data includes: Regarding the interference information as noise points, calculating the edge vectors between the points in the point cloud data and their neighboring points through the edge convolution module in the deep learning model, extracting local features, and removing the noise points; Edge convolution formula: Among them, is the updated feature of point , is a feature of its neighboring points, is a point and a point is the edge vector between them, is the weight matrix, is a learnable vector, is the activation function.

[0011] According to the above-mentioned method for extracting the building roof elevation based on urban modeling, when constructing a point cloud map and applying a graph convolutional operator based on a convolutional neural network to capture the dependency relationship and underlying connection pattern between points and extract the roof boundary, the convolutional neural network formula is as follows: Dynamic edge feature: Among them, is a multi-layer perceptron, is the feature vector, including coordinates, color, and intensity; Neighbor aggregation: Among them, is the number of nearest neighbor nodes, is the updated feature vector; Multi-layer graph convolution: Among them, is the adjacency matrix, is the degree matrix, , is the weight matrix of the th layer, is the non-linear activation function; Through dynamic graph construction and feature propagation, the convolutional neural network can capture the local geometric model and global structure dependency in the roof point cloud and improve the boundary recognition accuracy.

[0012] According to the above-mentioned method for extracting the building roof elevation based on urban modeling, when interpolating the point cloud data based on the adaptive interpolation algorithm, the deep learning model U-Net is also used to optimize the interpolation process; Among them, U-Net can capture elevation information and generate a high-quality DSM through skip connections and an encoder-decoder structure; The adaptive interpolation formula is as follows: Among them, is the interpolation elevation value of the point , is the elevation value of its neighboring points, is the point input feature, is the interpolation weight, is the interpolation function; The U-Net interpolation formula is as follows: where, is the interpolation elevation value of the point is the interpolation elevation value, is the point input feature, is the low-level feature map.

[0013] According to the building roof elevation extraction method based on urban modeling described above, in S3, For a flat roof, the steps to extract its top elevation by the plane fitting method are as follows: Use the point cloud data for plane fitting and fit a plane equation by the least squares method; Extract the median from the fitted plane as the elevation of the flat roof; For a pitched roof, the steps to extract its top elevation by the extreme point detection method are as follows: Identify the extreme points of the roof by calculating the curvature or gradient of the point cloud data; Take the height of the extreme points as the top elevation of the pitched roof; For roofs other than flat roofs and pitched roofs, the steps to extract their top elevations by the point cloud clustering method are as follows: Use the deep learning clustering algorithm K-means to extract the feature points of the roof and optimize the clustering results by minimizing the within-cluster distance. Its objective function is: where, is the set of cluster centers, is the assignment matrix from the samples to the cluster centers, is the center of the th cluster, is the th sample; Cluster the feature points to identify different parts of the roof; according to the clustering results, extract the highest point of each part as the top elevation of the roof.

[0014] According to the second aspect of the present invention, a system is provided, which includes a processor and a memory. The memory stores multiple instructions; the processor loads the instructions from the memory to execute the building roof elevation extraction method based on urban modeling.

[0015] The above scheme has the following technical effects: 1. Through the above solution, the roof elevation can be extracted quickly and accurately, providing high-precision three-dimensional data support for urban planning, energy assessment (such as solar panel layout) and disaster emergency response; solving the problems of low efficiency of roof elevation extraction, insufficient accuracy of complex roof structure recognition, and difficulty in multi-source data fusion in the existing technology; 2. Point cloud data provides three-dimensional structure, including three-dimensional spatial coordinates (XYZ), intensity and other information, and image data provides two-dimensional spectral semantic information. By integrating three-dimensional geometry and two-dimensional spectral information, it solves the limitation that a single data source is difficult to meet the needs of complex scenes, and significantly improves the classification accuracy and anti-interference ability of roof structures. Through multi-source data fusion, images can correct the missing points in point clouds due to occlusion (such as tree coverage), and point clouds can make up for the missing elevation information in images. 3. Through roof denoising and boundary optimization, the interference information that affects elevation extraction can be accurately identified, and the interference information can be removed from the point cloud data as noise points; secondly, the roof boundary can be accurately identified, and the roof digital surface model can be generated in combination with the adaptive interpolation algorithm; 4. The above scheme can accurately identify different roof structures and obtain appropriate top elevation information for different roof types.

[0016] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The present invention is further described below in conjunction with the accompanying drawings and embodiments: Figure 1 The figure is a flow chart of an embodiment of the method of the present invention. DETAILED DESCRIPTION

[0018] This section will describe in detail the specific embodiments of the present invention. The preferred embodiments of the present invention are shown in the accompanying drawings. The purpose of the accompanying drawings is to supplement the description of the text part of the specification with graphics, so that people can intuitively and vividly understand each technical feature and the overall technical solution of the present invention, but it cannot be understood as a limitation on the scope of protection of the present invention.

[0019] Reference Figure 1 , the method for extracting building roof elevation based on urban modeling in an embodiment of the present invention comprises: S1. Fusion registration of point cloud data and image data: Extract corner points and edge features from point cloud data; extract texture features from image data; optimize rigid body transformation parameters based on singular value decomposition to align corner points, edge features and texture features to align and register point cloud data and image data; combine image data with point cloud data to extract multimodal features; S2. Roof Denoising and Boundary Optimization: Identify and remove interference information from the point cloud data; based on a convolutional neural network, construct a point cloud graph and apply graph convolutional operators to capture the dependencies and underlying connection patterns between points, and extract the roof boundary; perform interpolation processing on the point cloud data based on an adaptive interpolation algorithm to generate a continuous roof digital surface model; S3. Roof Elevation Extraction: For flat roofs, extract their top elevations through plane fitting methods; for pitched roofs, extract their top elevations through extreme point detection methods; for roofs other than flat roofs and pitched roofs, extract their top elevations through point cloud clustering methods.

[0020] Among them, in S1, the point cloud data and the image data are fused and registered. Its core goal is to align the point cloud data (three-dimensional spatial coordinate data) with the image data (two-dimensional spatial coordinate data), so as to achieve the fusion under the unified reference framework of multi-source data.

[0021] Specifically, the corner points and edge features in the point cloud data can be extracted through the SIFT feature extraction algorithm, and the texture features of the image data can be extracted using the local binary pattern.

[0022] Among them, the SIFT feature extraction algorithm is an algorithm for extracting corner points and edge features in images. Its main steps include defining a three-dimensional Gaussian kernel function, calculating the three-dimensional Gaussian difference, and finding the extreme points in the scale space. The following are the detailed calculation formulas and steps: Define the three-dimensional Gaussian kernel function: Among them, are the point coordinates in three-dimensional space, is the scale parameter (controlling the width of the Gaussian kernel); Calculate the three-dimensional Gaussian difference: Calculate the difference between the Gaussian blurred point clouds at adjacent scales: Among them, is the scaling factor between adjacent scales (usually taking ); Find the extreme points in the scale space: Find the local extreme points in the three-dimensional scale space. For each point and scale , compare its value with the values of all adjacent points in its neighborhood; If is a local maximum or minimum, it is marked as a corner point and an edge feature.

[0023] In the steps of extracting texture features of image data using local binary patterns, for each pixel in the image, the local binary pattern (LBP) (the central pixel), around it neighboring pixels (usually evenly distributed on the circumference with a radius ), calculate the binary code: Where: is the number of neighboring pixels (such as ), is the neighborhood radius (such as ), is the sequence number of the pixel in the image; Count the frequencies of the LBP values of all pixels in the image to generate a histogram: Where, is the number of pixels with the LBP value of , is the sequence number of the pixel in the image.

[0024] Normalize the histogram (such as L1 or L2 norm) to obtain the final texture feature vector: In S1, based on singular value decomposition (SVD), optimize the rigid transformation ICP parameters to align the corner, edge features and texture features, which specifically includes: coarsely register the corner, edge features of the point cloud data and the texture features of the image data through rigid transformation; optimize the rigid transformation parameters through singular value decomposition and perform fine registration of the corner, edge features of the point cloud data and the texture features of the image data.

[0025] Among them, the steps of optimizing the rigid transformation parameters through singular value decomposition are as follows: Calculate the centroids of the source point cloud and the target point cloud : Subtract each point from the corresponding centroid to obtain the centered coordinates: Calculate the covariance matrix: Singular value decomposition: Where, and is an orthogonal matrix, is a diagonal matrix; The optimal rotation matrix is: The optimal translation vector is: By iterating the above steps, the rigid body transformation parameters and are gradually optimized until convergence to the optimal solution.

[0026] In the above solution, the goal of coarse registration is to quickly align the point cloud data and the image data, and the goal of fine registration is to improve the registration accuracy. The rigid body transformation ICP stage is coarse registration, and optimizing the coarse registration accuracy based on singular value decomposition (SVD) is the fine registration stage. By combining coarse and fine registration, the corner points and edge features in the point cloud data and the texture features in the image data are aligned, so that the point cloud data and the image data can be quickly aligned and registered with high registration accuracy.

[0027] In the step of combining the image data and the point cloud data to extract multi-modal features, the RGB image data and the point cloud data are combined, and multi-modal features are extracted through the feature fusion module to realize the fusion of the image data and the point cloud data and enhance the accuracy of semantic segmentation.

[0028] In S2, interference information is identified and removed from the point cloud data, which specifically includes: regarding the interference information as noise points, calculating the edge vectors between the points in the point cloud data and their neighboring points through the edge convolution module in the deep learning model, extracting local features and removing the noise points. Among them, the interference information may include trees, parapets, and temporary accumulations, etc., which will affect the accuracy of elevation extraction. Through the deep learning model, these interference information can be regarded as noise points and filtered.

[0029] The edge convolution formula is as follows: Among them, is the updated feature of point , is the feature of its neighboring point, is the edge vector between point and point , is the weight matrix, is the learnable vector, is the activation function.

[0030] In S2, the identification of the roof boundary is a crucial step in semantic segmentation. Through the Graph Convolutional Network (GCN), by combining point cloud data and image data, constructing a point cloud graph, and applying graph convolutional operators, it is possible to capture the dependency relationships and underlying connection patterns between points and accurately extract the roof boundary.

[0031] Among them, the formula of the graph convolutional network is as follows: Dynamic edge features: Among them, is a multi-layer perceptron, is the feature vector, including coordinates, color, and intensity; Neighbor aggregation: Among them, is the k nearest neighbor nodes, is the updated feature vector; Multi-layer graph convolution: Among them, is the adjacency matrix, is the degree matrix, , is the weight matrix of the l-th layer, is the non-linear activation function; Through dynamic graph construction and feature propagation, the GCN can capture the local geometric model and global structure dependencies in the roof point cloud and improve the boundary recognition accuracy.

[0032] In S2, after extracting the roof boundary, an adaptive interpolation algorithm is combined to generate a roof digital surface model. This algorithm interpolates the point cloud data to generate a continuous elevation surface, thereby improving the segmentation accuracy.

[0033] The adaptive interpolation formula is as follows: Among them, is the interpolated elevation value of point , is the elevation value of its neighboring points, is the input feature of point is the interpolation weight, is the interpolation function.

[0034] During the interpolation process of point cloud data based on the adaptive interpolation algorithm, the deep learning model U-Net is also used to optimize the interpolation process. Through skip connections and the encoder-decoder structure, U-Net can effectively capture elevation information and generate high-quality DSMs.

[0035] The U-Net interpolation formula is as follows: Where, is the interpolated elevation value of point , is the input feature of point , and is the low-level feature map.

[0036] In S3, for flat roofs, the top elevation is extracted by the plane fitting method, and the specific steps are as follows: Plane fitting: Use the point cloud data for plane fitting, and fit a plane equation through the least squares method or other optimization algorithms; Median extraction: Extract the median from the fitted plane as the elevation of the flat roof. This method is simple and efficient and is applicable to regular flat roofs.

[0037] For pitched roofs, the top elevation is extracted by the extreme point detection method, and the specific steps are as follows: Extreme point detection: Identify the extreme points (usually the highest points) of the roof by calculating the curvature or gradient of the point cloud data; Elevation extraction: Take the height of the extreme points as the top elevation of the pitched roof.

[0038] For roofs other than flat roofs and pitched roofs, these roofs are usually more complex (for example, some roofs are pyramid-shaped, polygonal, etc.), and the point cloud clustering method needs to be used to extract their top elevations. The specific steps are as follows: Feature point extraction: Use the deep learning clustering algorithm K-means to extract the feature points of the roof, and optimize the clustering results by minimizing the within-cluster distance. The objective function is: Where, is the set of cluster centers, is the assignment matrix of samples to cluster centers, is the center of the th cluster, is the th sample; Cluster analysis: Cluster the feature points to identify different parts of the roof, such as flat areas and changing areas; Elevation extraction: According to the clustering results, extract the highest point of each part as the top elevation of the roof.

[0039] In summary, the roof structures are roughly divided into three categories, and each has a different elevation extraction method. Flat roof: Take the median value by plane fitting; pitched roof: Extreme point detection; other complex roofs: Feature point cloud clustering analysis. Obtain appropriate top elevation information for different roof types; among them, extract the height of the flat roof position for flat roofs, and obtain the highest point for other non-flat roof types.

[0040] The present invention also discloses a system, which includes a processor and a memory, and the memory stores multiple instructions; the processor loads the instructions from the memory to execute the method for extracting the elevation of the building roof based on urban modeling as described above.

[0041] Although the methods described above are illustrated and described as a series of acts for simplicity of explanation, it should be understood and appreciated that the methods are not limited by the order of the acts, since in accordance with one or more embodiments, some acts may occur in different orders and / or concurrently with other acts not illustrated and described herein or other acts that would be understood by those skilled in the art. Those skilled in the art will further appreciate that the various illustrative logical blocks, modules, circuits, and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or combinations of both. To clearly illustrate this interchangeability of hardware and software, the various illustrative components, blocks, modules, circuits, and steps are described above in terms of their functionality. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the overall system. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention. The various illustrative logical blocks, modules, and circuits described in connection with the embodiments disclosed herein may be implemented using a general purpose processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. A general purpose processor may be a microprocessor, but in the alternative, the processor may be any conventional processor, controller, microcontroller, or state machine. The processor may also be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, multiple microprocessors, one or more microprocessors in conjunction with a DSP core, or any other such configuration. The steps of a method or algorithm described in connection with the embodiments disclosed herein may be embodied directly in hardware, in a software module executed by a processor, or in a combination of both. The software module may reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor such that the processor can read from, and write to, the storage medium. In the alternative, the storage medium may be integral to the processor. The processor and the storage medium may reside in an ASIC. The ASIC may reside in a user terminal. In the alternative, the processor and the storage medium may reside as discrete components in a user terminal. In one or more exemplary embodiments, the described functionality may be implemented in hardware, software, firmware, or any combination thereof. If implemented in software as a computer program product, the functions may be stored on or transmitted via a computer readable medium as one or more instructions or code.Computer-readable media includes both computer storage media and communication media including any medium that facilitates transfer of a computer program from one place to another. Storage media can be any available media that can be accessed by a computer. By way of example, and not limitation, such computer-readable media can include RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer. Any connection is properly termed a computer-readable media. For example, if software is transferred from a web site, server, or other remote source using a coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave, then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are included in the definition of the medium. As used herein, disk and disc include compact disc (CD), laser disc, optical disc, digital versatile disc (DVD), floppy disk and Blu-ray disc, where disk generally reproduces data magnetically, while disc reproduces data optically with lasers. Combinations of the above should also be included within the scope of computer-readable media.

[0042] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments, and various changes can be made without departing from the spirit of the present invention within the knowledge of those of ordinary skill in the art.

Claims

1. A building roof elevation extraction method based on urban modeling, characterized in that: include: S1. Fusion registration of point cloud data and image data: Extract corner points and edge features from point cloud data; extract texture features from image data; Based on singular value decomposition, the rigid body transformation parameters are optimized to align corner points, edge features and texture features, so as to align and register point cloud data with image data; image data is combined with point cloud data to extract multimodal features; S2, roof denoising and boundary optimization: Identify and remove interference information from point cloud data; construct a point cloud map based on a convolutional neural network and apply a graph convolution operator to capture the dependencies between points and the underlying connection pattern to extract the roof boundary; interpolate the point cloud data based on an adaptive interpolation algorithm to generate a continuous digital surface model of the roof; S3. Roof elevation extraction: For flat roofs, the top elevation is extracted by plane fitting method; for sloping roofs, the top elevation is extracted by extreme point detection method; for roofs other than flat roofs and sloping roofs, the top elevation is extracted by point cloud clustering method.

2. The method for extracting building roof elevation based on urban modeling according to claim 1, characterized in that: In S1, the corner points and edge features in the point cloud data are extracted by the SIFT feature extraction algorithm, and the texture features of the image data are extracted using the local binary pattern.

3. The building roof elevation extraction method based on urban modeling according to claim 2 is characterized in that: The steps to extract corner points and edge features from point cloud data using the SIFT feature extraction algorithm are as follows: Define a three-dimensional Gaussian kernel function: in, are the coordinates of a point in three-dimensional space, is the scale parameter; Compute the difference of Gaussians in three dimensions: Compute differences between Gaussian blurred point clouds at adjacent scales: in, is the scaling factor of the adjacent scale; Find the scale space extreme point: Find local extreme points in the three-dimensional scale space, for each point and scale , compared with its The values ​​of all adjacent points in the neighborhood; like If it is a local maximum or minimum, it is marked as a corner point and edge feature.

4. The method for extracting building roof elevation based on urban modeling according to claim 2 is characterized in that: The steps to extract texture features of image data using local binary patterns are as follows: Calculate the binary encoding: in: is the number of neighborhood pixels, is the neighborhood radius, is the sequential number of the pixel in the image; The LBP value frequency of all pixels in the image is counted to generate a histogram: in, The LBP value is The number of pixels, is the sequential number of the pixel in the image; Normalize the histogram to obtain the texture feature vector.

5. The building roof elevation extraction method based on urban modeling according to claim 1 is characterized in that: In S1, the optimization of rigid body transformation parameters based on singular value decomposition is used to align corner points, edge features and texture features, including: The corner points and edge features of the point cloud data are roughly aligned with the texture features of the image data through rigid body transformation; the corner points and edge features of the point cloud data are precisely aligned with the texture features of the image data through singular value decomposition to optimize the rigid body transformation parameters; The steps of optimizing rigid body transformation parameters by singular value decomposition are as follows: Calculate source point cloud and target point cloud Center of gravity: Subtract the corresponding center of gravity from each point to get the centralized coordinates: Compute the covariance matrix: Singular Value Decomposition: in, and is an orthogonal matrix, is a diagonal matrix; Optimal rotation matrix for: Optimal translation vector for: By iterating the above steps, the rigid body transformation parameters are gradually optimized. and , until it converges to the optimal solution.

6. The method for extracting building roof elevation based on urban modeling according to claim 1, characterized in that: In S2, identifying and removing interference information from point cloud data includes: The interference information is regarded as noise points, and the edge vector between the point in the point cloud data and its neighboring points is calculated through the edge convolution module in the deep learning model to extract local features and remove noise points. Edge convolution formula: in, Yes The updated features of is the characteristic of its neighboring points, Yes and Point The edge vector between is the weight matrix, is a learnable vector, is the activation function.

7. The method for extracting building roof elevation based on urban modeling according to claim 1, characterized in that: Based on the convolutional neural network, a point cloud map is constructed and the graph convolution operator is applied to capture the dependencies between points and the underlying connection pattern. In the step of extracting the roof boundary, the convolutional neural network formula is as follows: Dynamic edge features: in, is a multi-layer perceptron, is a feature vector, including coordinates, color, and intensity; Neighbor aggregation: in, yes nearest neighbor nodes, yes Updated feature vector; Multi-layer graph convolution: in, is the adjacency matrix, is the degree matrix, , It is The weight matrix of the layer, is a nonlinear activation function; Through dynamic graph construction and feature propagation, convolutional neural networks can capture local geometric models and global structural dependencies in roof point clouds and improve boundary recognition accuracy.

8. The method for extracting building roof elevation based on urban modeling according to claim 1, characterized in that: When interpolating point cloud data based on the adaptive interpolation algorithm, the deep learning model U-Net is also used to optimize the interpolation process; Among them, U-Net can capture elevation information and generate high-quality DSM through skip connections and encoder-decoder structure; The adaptive interpolation formula is as follows: in, Yes The interpolated elevation value of is the elevation value of its neighboring points, Yes The input features of is the interpolation weight, is the interpolation function; The U-Net interpolation formula is as follows: in, Yes The interpolated elevation value of Yes Input features, It is a low-level feature map.

9. The method for extracting building roof elevation based on urban modeling according to claim 1, characterized in that: In S3, For a flat roof, the steps to extract its top elevation by plane fitting method are as follows: Use point cloud data to fit a plane and fit a plane equation using the least squares method; The median value from the fitted plane is extracted as the elevation of the flat roof; For a sloped roof, the steps to extract its top elevation by using the extreme point detection method are as follows: By calculating the curvature or gradient of the point cloud data, the extreme points of the roof are identified; The height of the extreme point is taken as the top elevation of the sloping roof; For roofs other than flat roofs and pitched roofs, the steps to extract their top elevations through point cloud clustering are as follows: The deep learning clustering algorithm K-means is used to extract the feature points of the roof, and the clustering results are optimized by minimizing the intra-cluster distance. The objective function is: in, is the set of cluster centers, is the assignment matrix of samples to cluster centers, For the The center of the cluster, For the samples; The feature points are clustered to identify the different parts of the roof; based on the clustering results, the highest point of each part is extracted as the top elevation of the roof.

10. A system, characterized in that It comprises a processor and a memory, wherein the memory stores a plurality of instructions; the processor loads instructions from the memory to execute the building roof elevation extraction method based on urban modeling as described in any one of claims 1-9.

Images