Building facade extraction method, device and system
By fusing point clouds with images, the building facade is extracted, which solves the problem of inaccurate extraction of building facades in complex environments, and achieves higher extraction accuracy.
Patent Information
- Application Number
- CN202410910734.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-08
- Publication Date
- 2025-06-13
AI Technical Summary
In situations where the built environment is more complex, such as when the building is surrounded by trees, it is difficult to correctly extract the building facade, especially because the tree and the point clouds of the building may be removed together.
By obtaining the initial point cloud information and generating a plane feature image based on the gridded point cloud information. Then, the non-building image and the building facade image are subtracted to obtain the remaining building image, converted into the remaining building point cloud, and merged with the directly extracted building facade point cloud to obtain the complete building facade point cloud.
This method can accurately extract building facades in complex urban environments, avoiding the inaccuracy problems of traditional methods when extracting building facades and improving the accuracy of building facades.
Smart Images
Figure CN120147577A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the technical field of surveying and mapping information, and particularly relates to a method, device, and system for extracting building facades. Background Art
[0002] The point cloud building facade extraction technology uses three-dimensional laser scanning or photogrammetry technology to obtain the point cloud data of the building surface. Through computer vision and geometric processing methods, the automatic extraction of the building facade is realized. Finally, through three-dimensional modeling software or algorithms, the segmented facade point cloud data is converted into a three-dimensional model to achieve the digital reconstruction of the building facade. The point cloud building facade extraction technology is widely used in multiple fields such as building information modeling (BIM), urban planning, and cultural heritage protection.
[0003] Currently, a relatively common method for extracting building facades is the extraction method based on handheld or backpack Lidar (Light Detection And Ranging, an abbreviation in English referring to lidar): First, the ground points are removed from the input point cloud, and then feature extraction is performed on the overall point cloud through methods such as RANSAC (Random Sample Consensus), so as to obtain the building facade information.
[0004] However, in a scenario where the building environment is relatively complex, for example, when the building is surrounded by trees, the point clouds of the trees and the building may be removed together, making it difficult to correctly extract the building facade. Summary of the Invention
[0005] In view of the above, it is necessary to provide a method, device, and system for extracting building facades that can accurately extract building facades.
[0006] The first aspect of this application provides a method for extracting building facades, including: obtaining initial point cloud information; based on the gridded initial point cloud information, obtaining a planar feature image corresponding to a reference plane, where the planar feature image includes multiple pixel regions, and the gray value of each pixel region corresponds to the number of point clouds in the initial point cloud information that are mapped to the reference plane corresponding to the grid; based on the gridded initial point cloud information, obtaining non-building point cloud information and building point cloud information; mapping the non-building point cloud information to the reference plane to obtain a non-building image; mapping the building point cloud information to the reference plane to obtain a building facade image; based on the non-building image and the building facade image, performing subtraction processing on the planar feature image to obtain a first remaining building image; based on the first remaining building image, obtaining remaining building point cloud information; and merging the remaining building point cloud information with the building point cloud information to obtain target building point cloud information.
[0007] In the building facade extraction method provided by the first aspect of the present application, the point cloud is mapped into multiple images according to the point cloud information, and the images are processed to obtain the remaining building facade images in relatively complex scenarios in the city, such as when the building is surrounded by trees, etc. Then, the remaining building images are converted into remaining building point clouds and merged with the building facade point clouds obtained by directly extracting the building facade from the point cloud, thereby obtaining a complete and accurate building facade. In other words, the building facade extraction method of the present application extracts the building facade in the point cloud space by fusing the point cloud with the image, obtains the remaining building images in the image domain, and continues to obtain the remaining building facade in the point cloud space based on the remaining building images, thereby effectively solving the problem of inaccuracy in building facade extraction by traditional methods.
[0008] In some embodiments, the planar feature image includes a quantity feature image and a normal vector feature image.
[0009] In some embodiments, based on the gridded initial point cloud information, a planar feature image corresponding to the reference plane is obtained, including: mapping the gridded initial point cloud information to the reference plane, determining the pixel value of the corresponding pixel area based on the quantity of the point cloud mapped to the reference plane for each grid, and creating a quantity feature image; calculating the normal vector of each point cloud in the gridded initial point cloud information based on a preset scale to obtain the normal vector value corresponding to each point cloud, determining the point clouds that meet the preset normal vector angle condition as valid point clouds, determining the pixel value of the corresponding pixel area based on the quantity of the valid point clouds mapped to the reference plane for each grid, and creating a normal vector feature image; merging the quantity feature image and the normal vector feature image to obtain a planar feature image.
[0010] In some embodiments, merging the quantity feature image and the normal vector feature image to obtain a planar feature image includes: adjusting the pixel value of each pixel area in the quantity feature image based on a first weight, and adjusting the pixel value of each pixel area in the normal vector feature image based on a second weight, and merging the adjusted quantity feature image and the normal vector feature image to obtain a planar feature image.
[0011] In some embodiments, based on the first remaining building image, the remaining building point cloud information is obtained, including: determining the pixel areas that meet the building screening conditions in the first remaining building image as target pixel areas, and determining a second remaining building image based on the multiple target pixel areas; obtaining the remaining building point cloud information based on the second remaining building image.
[0012] In some embodiments, the pixel area meets the building screening conditions including: the pixel value of the pixel area reaches a preset pixel value range; and / or, the pixel area and other adjacent pixel areas form a continuous area, and the area of the continuous area reaches a preset area range.
[0013] In some embodiments, based on the gridded initial point cloud information, building point cloud information is obtained, including: removing non-building point cloud information from the gridded initial point cloud information to obtain remaining point cloud information, dividing the multiple grids where the remaining point cloud information is located into multiple grid layers and multiple grid columns, wherein the multiple grid layers are distributed along the Z-axis, each grid layer includes multiple grids distributed parallel to the reference plane, each grid column includes multiple grids distributed along the Z-axis, and the Z-axis is perpendicular to the reference plane; analyzing each grid layer in sequence from top to bottom along the Z-axis, and determining, from the multiple first grids in the grid layer, the first grids that meet the building analysis conditions as second grids, wherein the first grids and the determined second grids are located in different grid columns; obtaining the building point cloud information according to all the determined second grids.
[0014] In some embodiments, the first grid meeting the building analysis conditions includes: the maximum height of the point cloud in the first grid reaches the corresponding building height range, wherein the building height range is associated with the height of the grid layer where the first grid is located.
[0015] In some embodiments, the first grid meeting the building analysis conditions includes: calculating the normal vector values of each point cloud in the first grid based on a preset scale to obtain the normal vector values corresponding to each point cloud; the normal vector values corresponding to each point cloud in the first grid reach the corresponding building angle conditions.
[0016] In some embodiments, the preset scale includes a first scale and a second scale, and the normal vector values include a first normal vector value corresponding to the first scale and a second normal vector value corresponding to the second scale;
[0017] In some embodiments, the normal vector values corresponding to each point cloud in the first grid reaching the corresponding building angle conditions includes: the average value of the sum of the first normal vector value and the second normal vector value corresponding to each point cloud in the first grid is within a preset angle threshold range.
[0018] In some embodiments, based on the gridded initial point cloud information, non-building point cloud information is obtained, including: determining non-building grids according to the absolute height and relative height of the point cloud in each grid, and obtaining the non-building point cloud information according to the non-building grids, the absolute height being the maximum height of the point cloud on the Z-axis, and the relative height being the difference between the maximum height and the minimum height of the point cloud on the Z-axis.
[0019] In some embodiments, before obtaining the remaining building point cloud information based on the first remaining building image, it further includes: performing image processing on the second remaining building image, and the image processing includes at least one of dilation, erosion, opening and closing operations, and filtering.
[0020] The second aspect of the present application provides a building facade extraction device, including: a collector and a processor. The collector is used to collect point cloud data, and the processor is used to process the point cloud data and perform building facade extraction by applying any of the building facade extraction methods in the first aspect of the present application.
[0021] The third aspect of the present application provides a building facade extraction system, which is a system for performing building facade extraction by applying any of the building facade extraction methods in the first aspect of the present application. The system includes: a collection module, a point cloud processing module, and an image processing module. The collection module is used to obtain point cloud data. The point cloud processing module is used to perform at least one of the operations of screening, extracting, mapping, and merging on the point cloud data. The image processing module is used to perform at least one of the operations of pixel superposition, pixel subtraction, dilation, erosion, opening and closing operations, and filtering on the image obtained by the point cloud processing module through mapping the point cloud data. Description of the Drawings
[0022] Figure 1 It is a flowchart of the building facade extraction method according to an embodiment of the present application.
[0023] Figure 2 It is a sub-flowchart of step S200 in the building facade extraction method according to an embodiment of the present application.
[0024] Figure 3 It is a sub-flowchart of step S300 in the building facade extraction method according to an embodiment of the present application.
[0025] Figure 4 It is a sub-flowchart of step S400 in the building facade extraction method according to an embodiment of the present application.
[0026] Figure 5 It is a structural block diagram of the building facade extraction device according to an embodiment of the present application.
[0027] Figure 6 It is a structural block diagram of the building facade extraction system according to an embodiment of the present application.
[0028] Figure 7 It is a flowchart of a specific operation process for extracting the building facade in the building facade extraction method according to an embodiment of the present application.
[0029] Figure 8 is the present application Figure 7 A schematic diagram of the initial point cloud information in the building facade extraction method according to an embodiment of the present application.
[0030] Figure 9 is the present application Figure 7 A schematic diagram of the building image without image processing obtained by the building facade extraction method according to an embodiment of the present application.
[0031] Figure 10 is the present applicationFigure 7 Schematic diagram of the processed building image obtained by the building facade extraction method of the embodiment.
[0032] Figure 11 This application Figure 7 Schematic diagram of the quantitative feature image obtained by the building facade extraction method of the embodiment.
[0033] Figure 12 This application Figure 7 Schematic diagram of the normal vector feature image obtained by the building facade extraction method of the embodiment.
[0034] Figure 13 This application Figure 7 Schematic diagram of the non-building point cloud obtained by the building facade extraction method of the embodiment.
[0035] Figure 14 This application Figure 7 Schematic diagram of the remaining point cloud obtained by the building facade extraction method of the embodiment.
[0036] Figure 15 This application Figure 7 Schematic diagram of the non-building image obtained by the building facade extraction method of the embodiment.
[0037] Figure 16 This application Figure 7 Schematic diagram of the building facade point cloud obtained by the building facade extraction method of the embodiment.
[0038] Figure 17 This application Figure 7 Schematic diagram of the remaining unprocessed building image obtained by the building facade extraction method of the embodiment.
[0039] Figure 18 This application Figure 7 Schematic diagram of the remaining processed building image obtained by the building facade extraction method of the embodiment.
[0040] Figure 19 This application Figure 7 Schematic diagram of the remaining building facade point cloud obtained by the building facade extraction method of the embodiment.
[0041] Figure 20 This application Figure 7 Schematic diagram of the complete building facade point cloud obtained by the building facade extraction method of the embodiment. Detailed implementation manners
[0042] In the description of the embodiments of the present application, words such as "exemplary", "or", "for example", etc. are used to give examples, illustrations or explanations. Any embodiment or design solution described as "exemplary" or "for example" in the embodiments of the present application should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Rather, the use of words such as "exemplary", "or", "for example", etc. is intended to present related concepts in a specific manner.
[0043] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which this application belongs. The terms used in the description of this application are only for the purpose of describing specific embodiments and are not intended to limit this application. It should be understood that unless otherwise specified in this application, " / " means "or". For example, A / B may represent A or B. The "and / or" in this application is only a description of the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B may represent: A exists alone, A and B exist simultaneously, and B exists alone, these three situations. "At least one" means one or more. "Multiple" means two or more than two. For example, at least one of a, b or c may represent: a, b, c, a and b, a and c, b and c, a, b and c, these seven situations.
[0044] In addition, it should be noted that the terms "first" and "second" in the description, claims and drawings of this application are used to distinguish similar objects and are not used to describe a specific order or sequence. For the methods disclosed in the embodiments of this application or shown in the flowcharts, including one or more steps for implementing the method, without departing from the scope of the claims, the execution order of multiple steps can be interchanged with each other, and some steps can also be deleted.
[0045] In addition, it should be noted that the terms "X", "Y", "Z" and "reference plane" in the description, claims and drawings of this application refer to the X-axis, Y-axis, Z-axis and XOY plane of the coordinate system set in the digital point cloud space. The terms "X value", "Y value" and "Z value" refer to the coordinate values of the point cloud on the X-axis, Y-axis and Z-axis. The term "image domain" refers to the digital image processing space that maps the point cloud into a grayscale image and performs image processing such as pixel, expansion and denoising, including the time domain, frequency domain and spatial domain. In addition, the term "point" also refers to "point cloud".
[0046] Some embodiments will be described below with reference to the drawings. Without conflict, the following embodiments and the features in the embodiments can be combined with each other.
[0047] The point cloud building facade extraction technology uses 3D laser scanning or photogrammetry technology to obtain the point cloud data of the building surface. Through computer vision and geometric processing methods, it realizes the automatic extraction of the building facade. Finally, through 3D modeling software or algorithms, the segmented facade point cloud data is converted into a 3D model to achieve the digital reconstruction of the building facade. The point cloud building facade extraction technology is widely used in multiple fields such as building information modeling (BIM), urban planning, and cultural heritage protection.
[0048] Currently, the more common building facade extraction methods include the extraction method based on handheld or backpack Lidar scanning and the extraction method based on images: The extraction method based on handheld or backpack Lidar scanning first removes the ground points from the input point cloud, and then extracts the plane features of the overall point cloud through methods such as RANSAC to obtain the building facade information. However, in a complex building environment, such as when the building is surrounded by trees, it is difficult to correctly extract the building facade.
[0049] Specifically, when the building environment is complex, the point cloud collected by the radar includes non-building point cloud, the first building point cloud, and the second building point cloud. Among them, the non-building point cloud can include the point cloud of non-buildings such as trees, flower beds, or street lamps. The first building point cloud can include the point cloud of the first buildings such as buildings, stations, factories, or gas stations. The second building point cloud can include the point cloud of the second buildings such as the vertical walls close to non-buildings or small buildings obscured by non-buildings. The above existing methods can easily distinguish the non-building point cloud and the building point cloud through plane feature extraction. However, due to the limitations of plane feature extraction (such as when the plane features are not obvious or the plane features are interfered by non-buildings), in the actual operation process, the building point cloud obtained by the above existing methods through plane feature extraction only includes the first building point cloud, that is, the building point cloud with obvious plane features such as buildings, stations, factories, or gas stations and not interfered by non-buildings such as trees. The obtained non-building point cloud may actually include the second building point cloud, that is, the point cloud of the vertical walls close to non-buildings or small buildings obscured by non-buildings. Therefore, it is difficult to correctly extract the building facade.
[0050] Therefore, the embodiments of the present application provide a building facade extraction method, device, and system, which can accurately extract the building facade.
[0051] Figure 1 It is a flowchart of the building facade extraction method of the embodiments of the present application. The building facade extraction method in the present application can be executed by the building facade extraction device or system involved in the present application. For example, the point cloud acquisition can be obtained by a point cloud collector such as a radar. The point cloud collected by the collector can be processed for the point cloud or the building facade extraction in a processor such as a computer.
[0052] AsFigure 1 As shown in Figure 1 , the building facade extraction method of the present application includes:
[0053] Step S100: Obtain initial point cloud information.
[0054] Among them, the initial point cloud information includes non-building point cloud, first building point cloud and second building point cloud.
[0055] Step S200: Obtain a plane feature image.
[0056] In step S200, specifically, based on the gridded initial point cloud information, a plane feature image corresponding to the reference plane is obtained. Thus, a pixel image reflecting the size, position or height of the object can be obtained in the image domain. The plane feature image may include multiple pixel regions, and the gray value of each pixel region corresponds to the number of point clouds in the initial point cloud information mapped to the reference plane of the corresponding grid.
[0057] Step S300: Obtain a non-building image. In step S300, specifically, based on the gridded initial point cloud information, non-building point cloud information is obtained, and the non-building point cloud information is mapped to the reference plane to obtain a non-building image.
[0058] Step S400: Obtain a building facade image.
[0059] In step S400, specifically, based on the gridded initial point cloud information, building point cloud information is obtained, and the building point cloud information is mapped to the reference plane to obtain a building facade image.
[0060] Based on the gridded initial point cloud information, it also includes screening the point cloud data to obtain non-building point cloud information and the remaining point cloud after removing the non-building point cloud information, and obtaining building point cloud information based on the remaining point cloud. Among them, the non-building point cloud information includes non-building point cloud and second building point cloud, and the building point cloud information includes first building point cloud. Thus, in the subsequent image domain, the building facade image and other related images can be processed to obtain the remaining building image, so as to accurately identify the complete building, that is, accurately identify all buildings including the main building, the enclosure wall and other relatively low and small building buildings.
[0061] Step S500: Obtain remaining building point cloud information.
[0062] In step S500, obtaining the remaining building point cloud information specifically includes:
[0063] Based on the non-building image and the building facade image, perform subtraction processing on the plane feature image to obtain a first remaining building image.
[0064] Based on the first remaining building image, obtain the remaining building point cloud information.
[0065] In this way, the building point cloud and the remaining building point cloud can be merged in the subsequent point cloud space to obtain a complete building point cloud. It can be understood that the complete building point cloud, also known as the target building point cloud information, includes the point cloud information of the main building, the vertical wall, and other relatively short and small buildings, etc.
[0066] In some embodiments, obtaining the remaining building point cloud information based on the first remaining building image in step S500 may include:
[0067] Determine the pixel area in the first remaining building image that meets the building screening condition as the target pixel area, and determine the second remaining building image based on the multiple target pixel areas.
[0068] Obtain the remaining building point cloud information based on the second remaining building image.
[0069] In some embodiments, the pixel area meets the building screening condition, including:
[0070] The pixel value of the pixel area reaches a preset pixel value range. Or, the pixel area and other adjacent pixel areas form a continuous area, and the area of the continuous area reaches a preset area range.
[0071] In other embodiments, the pixel area meets the building screening condition, including:
[0072] The pixel value of the pixel area reaches a preset pixel value range. And, the pixel area and other adjacent pixel areas form a continuous area, and the area of the continuous area reaches a preset area range.
[0073] It can be understood that whether the pixel area is a pixel of the remaining building is judged by the pixel value range and the area of the continuous area of the pixel area. For example, in a complex urban background, in a fence or a relatively short floor surrounded by trees, the pixel value of the trees will be lower than that of the fence or the relatively short floor, and the pixel areas of the trees are not all continuous, while the pixel areas of the fence or the relatively short floor are continuous and the continuous area is within a certain range.
[0074] In some embodiments, before obtaining the remaining building point cloud information based on the first remaining building image in step S500, it further includes:
[0075] Perform image processing on the second remaining building image.
[0076] It can be understood that the image processing includes at least one of dilation, erosion, opening and closing operations, and filtering.
[0077] It can be understood that by performing image processing on the second remaining building image, the quality of the remaining building image can be improved. For example, through dilation processing, the pixels near the remaining building are also regarded as the building facade, that is, image expansion. Also, for different application scenarios, image processing methods such as erosion, opening and closing operations, and filtering can be used to supplement image pixels and remove noise interference, etc.
[0078] Step S600: Merge point cloud information.
[0079] Specifically, in step S600, the remaining building point cloud information and the building point cloud information are merged to obtain the target building point cloud information.
[0080] Through the building facade extraction method provided by the embodiments of the present application, a plane feature image is obtained based on the initial point cloud information, and the distribution of the point cloud on the reference plane is mapped by multiple pixel regions of the plane feature image. Since the pixel regions of the non-building image mainly reflect the distribution of non-buildings, and the pixel regions of the building facade image mainly reflect the distribution of the first building, therefore, in the process of subtracting the plane feature image based on the non-building image and the building facade image, the pixel regions of the plane feature image that reflect non-buildings and the first building can be processed. The pixel regions of the first remaining building image obtained after the subtraction process can reflect the distribution of the second building. In this way, the remaining building point cloud information includes the point cloud that reflects the second building. By using the remaining building point cloud information and the building point cloud information for merging, the point clouds of the first building and the second building can be merged to obtain the complete point clouds of all buildings, improving the accuracy of building facade extraction.
[0081] In some embodiments, step S100 may further include preprocessing the initial point cloud information. The preprocessing may include downsampling (also known as subsampling) processing. Thus, the subsequent processing of the initial point cloud information can be accelerated.
[0082] Figure 2 It is a sub-flowchart of step S200 in the building facade extraction method of the embodiments of the present application. As Figure 2 shown, in some embodiments, step S200 may include:
[0083] Step S201: Grid the initial point cloud information.
[0084] In step S201, the gridding of the initial point cloud information is related to the sizes of the initial point cloud information in the X and Y directions.
[0085] Step S202: Create a quantity feature image.
[0086] In step S202, specifically, the gridded initial point cloud information is mapped to a reference plane, and the pixel value of the corresponding pixel area is determined based on the number of point clouds mapped to the reference plane for each grid, and a quantity feature image is created.
[0087] In some embodiments, determining the pixel value of the corresponding pixel area based on the number of point clouds mapped to the reference plane for each grid and creating a quantity feature image is specifically as follows: counting the number of point clouds in each grid, scaling the number of point clouds to a predetermined scale, such as [0, 255], thereby realizing the mapping from the number of point clouds to the quantity feature image, which is convenient for subsequent processing in the image domain.
[0088] Step S203, create a normal vector feature image. In step S203, specifically, based on a preset scale, the normal vectors of each point cloud in the gridded initial point cloud information are calculated to obtain the normal vector values corresponding to each point cloud, the point clouds that meet the preset normal vector angle condition are determined as valid point clouds, and the pixel value of the corresponding pixel area is determined based on the number of valid point clouds mapped to the reference plane for each grid, and a normal vector feature image is created.
[0089] In some embodiments, determining the pixel value of the corresponding pixel area based on the number of valid point clouds mapped to the reference plane for each grid and creating a normal vector feature image is specifically as follows: counting the number of point clouds with normal vectors at a specific angle in each grid, scaling the number of point clouds with normal vectors at the specific angle to a predetermined scale, such as [0, 255], thereby realizing the mapping from the point cloud normal vectors to the normal vector feature image, which is convenient for subsequent processing in the image domain.
[0090] In some embodiments, the normal vector calculation based on the preset scale can also be performed in advance in step S100, that is, the normal vector calculation can be performed before the gridding step, and there can be multiple preset scales. Performing multi-scale normal vector calculation on the initial point cloud information after voxel downsampling is convenient for subsequent gridding of the point cloud and multi-scale feature extraction.
[0091] Step S204, merge images.
[0092] In step S204, specifically, the quantity feature image and the normal vector feature image are merged to obtain a plane feature image.
[0093] Step S204 may include:
[0094] Adjust the pixel values of each pixel area in the quantity feature image based on the first weight.
[0095] Adjust the pixel values of each pixel area in the normal vector feature image based on the second weight.
[0096] Merge the adjusted quantitative feature image and the normal vector feature image to obtain a planar feature image.
[0097] It can be understood that by setting weights, the weights of the quantitative feature image or the normal vector feature image in the merging process can be set based on actual building features or building information, enhancing the building facade features, and thus the planar feature image can be obtained more accurately.
[0098] Figure 3 It is a sub - flowchart of step S300 in the building facade extraction method of the embodiment of the present application. Please refer to Figure 3 , in some embodiments, step S300 may include:
[0099] Step S301, determine non - building grids.
[0100] Step S302, obtain non - building point cloud information.
[0101] Step S303, obtain non - building images.
[0102] Among them, non - building grids are determined according to the absolute height and relative height of the point cloud in each grid. Non - building point cloud information is obtained based on the non - building grids. The absolute height is the maximum height of the point cloud on the Z - axis, and the relative height is the difference between the maximum height and the minimum height of the point cloud on the Z - axis. Through the restrictions of the absolute height and relative height, non - building grids are marked and removed, thereby being able to remove the influence of the ground, pedestrians, vehicles, and low - lying vegetation.
[0103] Figure 4 It is a sub - flowchart of step S400 in the building facade extraction method of the embodiment of the present application. Please refer to Figure 4 , in some embodiments, step S400 may include:
[0104] Step S401, obtain the remaining point cloud information.
[0105] In step S401, specifically, based on the grid - based initial point cloud information, the non - building point cloud information in the above step S300 is removed to obtain the remaining point cloud information. Thus, the building point cloud information obtained from the remaining point cloud information after removing the non - building point cloud information is more accurate than the building point cloud information directly obtained from the initial point cloud information. That is, after removing non - buildings such as trees or streetlights, the building model obtained in this case is closer to the actual building.
[0106] In the present application, step S401 can also be performed in step S300, that is, step S300 can include removing non - building point cloud information to obtain the remaining point cloud information.
[0107] Step S402: Divide the grid. Specifically, in step S402, the multiple grids where the remaining point cloud information is located are divided into multiple grid layers and multiple grid columns. Among them, the multiple grid layers are distributed along the Z-axis. Each grid layer includes multiple grids distributed parallel to the reference plane, and each grid column includes multiple grids distributed along the Z-axis. The Z-axis is perpendicular to the reference plane.
[0108] Step S403: Analyze the grid.
[0109] Specifically, in step S403, each grid layer is analyzed sequentially from top to bottom along the Z-axis. Among the multiple first grids in the grid layer, the first grid that meets the building analysis conditions is determined as the second grid. Among them, the first grid and the determined second grid are located in different grid columns.
[0110] Step S404: Obtain the building point cloud information. Specifically, in step S404, according to all the second grids determined in step S403, the building point cloud information is obtained.
[0111] Thus, through steps S402, S403, and S404, the grid is analyzed layer by layer to obtain the building point cloud information. And by preferentially analyzing the grids of high-rise buildings from top to bottom, the influence of other non-building point clouds (such as non-building point clouds reflected by building glass) can be reduced.
[0112] In addition, by determining the first grid that meets the building analysis conditions as the second grid, all the grids corresponding to the analyzed grid along the Z-axis can be omitted in the layer-by-layer analysis. Therefore, it can achieve accurate acquisition of the building point cloud information while reducing the analysis of a large amount of point cloud data.
[0113] In some embodiments, the first grid meets the building analysis conditions, including:
[0114] The maximum height of the point cloud in the first grid reaches the corresponding building height range, where the building height range is associated with the height of the grid layer where the first grid is located. Thus, it can be determined whether it belongs to the building point cloud through the height of the point cloud in each layer of the grid. For example, buildings and street lights can be judged and distinguished.
[0115] In some embodiments, the first grid meets the building analysis conditions, including:
[0116] Based on a preset scale, the normal vector of each point cloud in the first grid is calculated to obtain the normal vector value corresponding to each point cloud; the normal vector values corresponding to each point cloud in the first grid reach the corresponding building angle condition. Thus, it can be determined whether it belongs to the building point cloud through the normal vector value of the point cloud in each layer of the grid. For example, buildings and trees can be judged and distinguished.
[0117] In some embodiments, the preset scale includes a first scale and a second scale, and the normal vector values include a first normal vector value corresponding to the first scale and a second normal vector value corresponding to the second scale. Thus, the building point cloud can be judged by normal vector values at multiple scales, improving the accuracy.
[0118] In some embodiments, the normal vector values corresponding to the point clouds of each point in the first grid meet the corresponding building angle conditions, including:
[0119] The average value of the sum of the first normal vector value and the second normal vector value corresponding to the point clouds of each point in the first grid is within a preset angle threshold range.
[0120] Thus, it is possible to judge some buildings with inclined angles or errors caused by point cloud acquisition. For example, for an inclined building, since the normal vector directions of the building facades are the same, the average value of the normal vectors at multiple scales is still within the preset angle threshold range, while for trees or other discrete point clouds collected, the difference in the normal vector directions at multiple scales is relatively large, not meeting the building angle conditions.
[0121] Step S405: Obtain the building facade image.
[0122] It can be understood that before step S405, it further includes:
[0123] Perform image processing on the building facade image.
[0124] It can be understood that the image processing includes at least one of dilation, erosion, opening and closing operations, and filtering.
[0125] It can be understood that by performing image processing on the building facade image, the quality of the building facade image can be improved. For example, through dilation processing, the pixels near the building facade are also regarded as the building facade, and then through pixel adjustment (such as setting the pixel to 255), the point clouds of the corresponding grid are also regarded as building point clouds, that is, image expansion. Also, for different application scenarios, image processing methods such as erosion, opening and closing operations, and filtering can be used to supplement image pixels and remove noise interference, etc.
[0126] It can be understood that the building facade extraction method of the present application maps point clouds into multiple images according to point cloud information and processes the images to obtain the remaining building facade images in relatively complex urban scenarios, such as when a building is surrounded by trees, etc. Then, the remaining building images are converted into remaining building point clouds and merged with the building facade point clouds obtained by directly extracting the building facade from the point clouds, thereby obtaining a complete and accurate building facade. In other words, the building facade extraction method of the present application extracts the building facade in the point cloud space by fusing point clouds and images, obtains the remaining building images in the image domain, and continues to obtain the remaining building facades in the point cloud space based on the remaining building images, thereby effectively solving the problem of inaccuracy in building facade extraction by traditional methods.
[0127] Figure 5 It is the structural block diagram of the building facade extraction device according to an embodiment of the present application.
[0128] Please refer to Figure 5 , the building facade extraction device 1 of the present application includes a data receiving module 11, an image mapping module 12, a point cloud analysis module 13, an image processing module 14, a point cloud conversion module 15, and a point cloud merging module 16.
[0129] Among them, the data receiving module 11 can be used to obtain initial point cloud information.
[0130] The image mapping module 12 can be used to obtain a plane feature image corresponding to the reference plane based on the meshed initial point cloud information. Among them, the plane feature image includes multiple pixel regions, and the gray value of each pixel region corresponds to the number of point clouds in the initial point cloud information where the corresponding grid is mapped to the reference plane.
[0131] The point cloud analysis module 13 can be used to obtain non-building point cloud information and building point cloud information based on the meshed initial point cloud information, map the non-building point cloud information to the reference plane to obtain a non-building image, and map the building point cloud information to the reference plane to obtain a building facade image.
[0132] The image processing module 14 can be used to perform a subtraction process on the plane feature image based on the non-building image and the building facade image to obtain a first remaining building image.
[0133] The point cloud conversion module 15 can be used to obtain remaining building point cloud information based on the first remaining building image.
[0134] The point cloud merging module 16 can be used to merge the remaining building point cloud information with the building point cloud information to obtain target building point cloud information.
[0135] In the present application, the data receiving module 11 may include an airborne scanning radar, a handheld scanning radar, a backpack scanning radar, or a lidar. The image mapping module 12, the point cloud analysis module 13, the image processing module 14, the point cloud conversion module 15, and the point cloud merging module 16 may be integrated into a processor 22 with data processing capabilities such as a CPU, a computer, or a cloud processor 22.
[0136] Figure 6 is a structural block diagram of the building facade extraction system according to an embodiment of the present application.
[0137] Please refer to Figure 6 , the present application also provides a building facade extraction system 2, including: a collector 21 and a processor 22.
[0138] Among them, the collector 21 can be used to perform scanning to collect point cloud data. The processor 22 can be used to execute the above-mentioned building facade extraction method to extract the building facade from the point cloud data collected by the collector 21.
[0139] In the present application, the collector 21 may be an airborne scanning radar, a handheld scanning radar, a backpack scanning radar, or a lidar, and the processor 22 may include but are not limited to a CPU, a computer, a cloud processor 22, etc.
[0140] For the working principles and technical effects of the building facade extraction device and the building facade extraction system 2 provided by the embodiments of the present application, reference may be made to the relevant descriptions of the building facade extraction method in the foregoing embodiments, and the present application will not elaborate herein.
[0141] Hereinafter, in conjunction with Figures 7 to 20 , the building facade extraction method involved in the embodiments of the present application will be described in detail.
[0142] Please refer to Figure 7 , in step S701, obtaining the initial point cloud information includes obtaining the point cloud (i.e., the original point cloud) and preprocessing the point cloud. Among them, the preprocessing operation can be voxel downsampling (also known as voxel subsampling), with a voxel side length of 0.1 m. All input point clouds are voxelized, and the center of each voxel represents all the point clouds within that voxel. This effectively reduces the number of point clouds to be calculated and improves the subsequent segmentation efficiency. It can be understood that step S701 corresponds to step S100 of the above-mentioned building facade extraction method.
[0143] In some embodiments, the optimal value of the voxel side length can be determined based on the quality of the collected point cloud.
[0144] In step S701 or step S702, multi-scale normal vector calculation is performed on the initial point cloud information. Specifically, by calculating the normal vector of the downsampled point cloud, the normal vector of each point is estimated for a specific number of point clouds around each point cloud, such as 10 point clouds or 20 point clouds, to obtain the normal vector of each point at the corresponding scale. Thus, the normal vector feature can be used as the basis for generating an image from the point cloud. Moreover, since most building facades are planar features and can still maintain planar features at similar scales, while the point clouds of trees and the like are divergent and cannot maintain the same features at different scales, the multi-scale normal vectors are used for subsequent determination of building facade point clouds.
[0145] In some embodiments, the number of point clouds for calculating the normal vector and the optimal value of the number of scales can be determined based on the quality of the collected point cloud.
[0146] Thus, the initial point cloud information as shown in Figure 8 is obtained.
[0147] Please refer to Figure 7 , step S702 may specifically include: gridifying the point cloud features; based on the gridified initial point cloud information, obtaining a planar feature image including a quantity feature image and a normal vector feature image. Gridifying the initial point cloud information specifically means: traversing the entire preprocessed initial point cloud information, determining the maximum value maxPt and the minimum value minPt of the initial point cloud information on the X, Y, and Z axes, determining the grid step size. For example, in this application, the grid step size is set to 0.8m, and through the formula:
[0148]
[0149] determining the size of the grid, where maxPt x and maxPt y are the X value and Y value of the maximum value point maxPt; minPt x and minPt y are the X value and Y value of the minimum value point minPt; gridScale is the grid step size, which can be set to 0.8m in this application; col is the number of columns of the grid, and row is the number of rows of the grid.
[0150] In step S702, creating the quantity feature image specifically means: constructing an image with the same size as col and row of the grid obtained by gridifying, where the initial pixel value of the image is a preset value, such as the pixel value is preset to 0; according to the grid divided by the point cloud on the reference plane obtained by gridifying, traversing all points in the point cloud, and through the formula:
[0151]
[0152] determining the grid position where the point is located, where Ptx 、Pt y is the X value and Y value of the current point Pt being traversed; minPt x 、minPt y are the X value and Y value of the minimum point minPt; gridScale is the grid step size, for example, in this application, it can be set to 0.8m; colPt is the column number of the grid where the current point Pt is located, rowPt is the row number of the grid where the current point Pt is located, and the symbol is for ceiling to avoid the point being on the grid edge. After determining the grid positions of all points, by counting the number of point clouds in each grid, the maximum value grideNum of the number of point clouds in the grid is determined max and the minimum value grideNum min , through the formula:
[0153]
[0154] the number of point clouds is scaled to the preset scale in the pixel value space, for example, in this application, it is preset to the scale of [0, 255]. Where Grayvalue is the gray value of the pixel, grideNum ij is the number of point clouds in the grid at the i-th row and j-th column, grideNum max is the maximum value of the number of point clouds in all grids, grideNum min is the minimum value of the number of point clouds in all grids, and 255 is the maximum value of the grayscale image. Thus, the mapping between the number of point clouds in the point cloud grid and the image pixels is realized, and the quantity feature image as shown in Figure 11 is obtained. It should be noted that Figure 11 has been processed by color inversion. In the actual quantity feature image, the background is the filling color with pixel value 0, while the quantity feature image is the filling color with pixel value 255.
[0155] In this application, the quantity feature image has only one dimension, so the quantity feature image appears as a grayscale image, and its pixel value is the gray value.
[0156] Create a normal vector feature image, specifically: after determining the grid positions of all point clouds when creating the quantity feature image, by traversing the points in each grid, calculate the angle between the normal vector of each point and the Z-axis direction, count the number of points within the preset range of the angle (i.e., the angle limit range) in this grid. For example, in this application, the preset range is between [75°, 110°], and determine the maximum value grideAngle of the number of points that meet the angle limit in all grids max and the minimum value grideAngle min , through the formula:
[0157]
[0158] Scale the included angle quantity value to a preset scale in the pixel value space. For example, in this application, the preset scale is [0, 255]. Where Grayvalue is the gray value of the pixel, 9rideAngle ij is the number of point clouds that meet the included angle limit in the grid at the i-th row and j-th column, grideAngle max is the maximum number of point clouds that meet the included angle limit in all grids, grideAngle min is the minimum number of point clouds that meet the included angle limit in all grids, and 255 is the maximum value of the grayscale image. Thus, the mapping between the included angle quantity in the point cloud grid and the image pixels is realized, and a normal vector feature image as shown in Figure 12 is obtained. It should be noted that Figure 12 has been processed by color inversion. In the actual normal vector feature image, the background is the filling color with a pixel value of 0, while the normal vector feature image is the filling color with a pixel value of 255.
[0159] In this application, the normal vector feature image has only one dimension. Therefore, the normal vector feature image appears as a grayscale image, and its pixel value is the gray value.
[0160] In this application, the above-mentioned included angle limit range can be determined by collecting the specific features of the building.
[0161] In this application, multiple feature images can also be created according to the actual situation, such as the maximum-minimum value feature image, the PCA feature image, etc.
[0162] In this application, the merged image (also known as image fusion) in step S702 can be performed before step S705 or synchronously with step S705. For the specific process, refer to the content of step S705 below. It can be understood that step S702 corresponds to step S200 of the above-mentioned building facade extraction method.
[0163] Thus, a quantity feature image as shown in Figure 11 and a normal vector feature image as shown in Figure 12 are obtained.
[0164] Please refer to Figure 7 , step S703 may include removing non-building facade point clouds to obtain non-building point cloud information and non-building images. In step S303, to obtain the non-building image, specifically: traverse all grids, and for the point clouds in each grid, according to the formula:
[0165]
[0166] judge non-building grids, where grideHighMax ij is the maximum value of the Z value of the point cloud in the grid at the i-th row and j-th column, grideHighMinij is the minimum value of the Z value of the point cloud at the midpoint of the grid in the i-th row and j-th column. By restricting the absolute height within the grid to a preset value, for example, the preset value in this application is 3.0, non-building grids such as those belonging to the ground and pedestrians can be removed; by restricting the relative height to a preset value, for example, the preset value in this application is 1.0, non-building noise clutter points can be effectively removed. Mark the grids screened out above as non-elevation building grids, and all the point clouds contained therein are non-building point clouds (as Figure 13 shown), and remove the non-building point clouds from the initial point cloud information to obtain the remaining point cloud as shown in Figure 14 shown.
[0167] In this application, the limit parameters of the absolute height and relative height can be determined according to the target features to be removed.
[0168] Construct a non-building image of the same size according to col and row in step S702, where the initial pixel value of the non-building image is a preset value, for example, the preset value in this application is 0. According to the coordinates of the non-elevation building grids marked above, set the pixel value of the corresponding position of the image to the preset value, for example, the preset value in this application is 255, to obtain the non-building image as shown in Figure 15 shown.
[0169] Thus, the non-building image as shown in Figure 15 is obtained. It can be understood that step S703 corresponds to step S300 of the above building elevation extraction method.
[0170] Please refer to Figure 7 , step S704 includes: obtaining the remaining point cloud, and performing point cloud building elevation extraction in the remaining point cloud, that is, obtaining the building elevation point cloud and obtaining the building elevation image. In step S704, to obtain the building elevation image, specifically: for the remaining point cloud obtained by removing the non-building point cloud in step S703 from the initial point cloud information, first perform point cloud building elevation extraction layer by layer in decreasing order from the highest point according to the overall point cloud height, and the overall point cloud height is the maximum height of the input point cloud. Among them, decreasing layer by layer from the highest point means starting from the maximum height and decreasing step by step according to the specified step size. For example, the step size in this application can be set to 1m. In this application, the step size can be determined by collecting the difference in height between the building and the surrounding point cloud.
[0171] In each layer of decreasing order, use the formula:
[0172]
[0173] to determine whether the grid participates in the building grid judgment in the current layer, where grideHigh ijk is the Z value of the k-th point in the grid in the i-th row and j-th column, and LayerHigh nis the floor height value of the nth layer that decreases step by step. By counting the number of points in the grid whose z value is greater than the current floor height (i.e., Num ij ), for example, this application determines whether it is greater than 100, and determines whether this grid participates in the building grid judgment at the current layer. Thus, through this step-by-step decreasing method, high-rise building grids can be preferentially judged, reducing the influence of other non-building point clouds. In this application, the limit parameter of the quantity can be determined by the point cloud quality, such as 200, 300, 400 or more.
[0174] Specifically, step S704 further includes: at each layer, for the grids that meet the quantity limit, perform building point cloud judgment on this layer. Through the multi-scale normal vectors calculated by step 1, determine the building grid through the angle between the normal vector and the Z axis and the multi-scale angle limit.
[0175] For the grids that meet the above building grid judgment requirements, traverse all the points in this grid. According to the multi-scale normal vectors calculated in step S701, obtain the angles a1 and a2 between the normal vector and the Z axis; where a1 is the angle between the normal vector of the surrounding point cloud at the first preset scale and the Z axis. For example, the first preset scale in this application can be 10; a2 is the angle between the normal vector of the surrounding point cloud at the second preset scale and the Z axis. For example, the second preset scale in this application can be 20. Through the formula:
[0176]
[0177] perform the screening of the building grid. Among them, averageAngle is the average value of a1 and a2. Since the building facade shows planar characteristics and can still maintain similar planar characteristics at different radius scales, the change in the angle between the normal vector and the Z axis at different scales is small, and its average value range is around 90 degrees, and the difference between the two should be similar. In this application, the average value range is set to [75°, 110°], and the difference is 30°. In this application, the best values of the average value range and the difference can be determined by the flatness of the building facade. For example, the average value range is set to [80°, 100°], and the difference is 20°, or the average value range is set to [65°, 115°], and the difference is 35°.
[0178] Specifically, please refer to Figure 7 , step S704 further includes: extracting the building grid point clouds that meet the above building grid screening conditions, and marking the building grid to obtain a building facade image as shown in Figure 9 . Expand the building facade image through image dilation operation, and regard the pixels around the building as the building facade to obtain as shown in Figure 10The building facade image shown; through the mapping relationship between the image and the point cloud, the corresponding grid point cloud is also divided into the building facade point cloud to achieve building facade extraction. The specific implementation is as follows: According to col and row in step S702, a building facade image of the same size is constructed, where the initial pixel value of the building facade image is a preset value. For example, in this application, the preset value is 0. According to the marked facade building grid coordinates above, the pixel value at the corresponding position of the image is set to the preset value. For example, in this application, the preset value is 255, and the building facade image shown in Figure 9 is obtained. Through the dilation operation in image processing, the pixels near the building facade are also regarded as the building facade, and the building facade image shown in Figure 10 is obtained. Its pixel value is set to 255. Therefore, the point cloud in the corresponding grid is also regarded as the building point cloud; since the image is obtained by mapping from the point cloud, the grid position with a pixel value of 255 is extracted from the image, and its mapping is the grid at the same position. The point cloud in the corresponding grid is extracted to obtain the building facade point cloud shown in Figure 16 .
[0179] In this application, for different application scenarios, the image processing methods include but are not limited to common image processing methods such as erosion, dilation, opening and closing operations, filtering, etc. for supplementing image pixels and removing noise interference, etc.
[0180] Thus, the building point cloud information and the building facade image are obtained. It can be understood that step S704 corresponds to step S400 of the above building facade extraction method.
[0181] In step S705, the remaining building point cloud information (also called image domain processing) is obtained. Specifically: The quantity feature image and the normal vector feature image in step S702 are pixel-overlaid according to the corresponding weights to enhance the building facade feature, subtract the non-building image obtained in step S703, and then subtract the extracted building facade image in step S704, that is, image fusion, to obtain the first remaining building image (that is, the remaining building image without image processing shown in Figure 17 ). Then, pixel screening is performed by setting a threshold (that is, pixel threshold limitation), and the pixel area in the first remaining building image is screened (that is, area area threshold limitation) to obtain the second remaining building image. Through image processing such as dilation operation, the pixels near the remaining building facade are also regarded as the building facade, and finally the remaining building image shown in Figure 18 is obtained (also called the remaining building facade image). The specific operation is as follows:
[0182]
[0183] where ImageBuild2 is the remaining building facade image, Image num is the quantity feature image, Image normalis the normal vector feature image, Image non is the non-building image in step S703, and ImageBuild1 is the building elevation image extracted in step S704. In this application, the above addition and subtraction refer to operations on pixel values.
[0184] In this application, the weight of the quantity feature image is set to a preset value. For example, in this application, it is preset to 0.5. The weight of the normal vector feature image is set to a preset value. For example, in this application, it is preset to 0.5. The pixel threshold is also set to a preset value. For example, in this application, it is preset to 200. The image area threshold is also set to a preset value. For example, in this application, it is preset to 100.
[0185] In this application, the weights of the quantity feature image and the normal vector feature image can be set according to specific building point cloud features. In addition, appropriate pixel thresholds and image area thresholds can be set according to the actual fused image to remove noise.
[0186] In this application, for different application scenarios, the image processing methods include but are not limited to common image processing methods such as erosion, dilation, opening and closing operations, and filtering. Thus, it can be used to supplement image pixels and remove noise interference, etc.
[0187] Specifically, step S705 further includes: extracting the remaining grid point cloud through the mapping relationship between the image and the point cloud grid, that is, as Figure 19 shown in the remaining building elevation point cloud (also called the remaining building point cloud information).
[0188] Thus, the remaining building point cloud information is obtained. It can be understood that step S705 corresponds to step S500 of the above building elevation extraction method.
[0189] Please refer to Figure 7 , in step S706, the point cloud information is merged (i.e., the point cloud is merged). Specifically: the building elevation point cloud obtained in step S704 and the remaining building elevation point cloud obtained in step S705 are merged to achieve the extraction of the building elevation point cloud, that is, to obtain the complete building elevation point cloud as Figure 20 shown.
[0190] Thus, the complete building elevation point cloud (also called the target building point cloud information) is obtained. It can be understood that step S706 corresponds to step S600 of the above building elevation extraction method.
[0191] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and not to limit them. Although the present application has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present application can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present application.
Claims
1. A building facade extraction method, characterized in that: include: Get initial point cloud information; Based on the gridded initial point cloud information, a plane feature image corresponding to the reference plane is obtained, wherein the plane feature image includes a plurality of pixel areas, and the grayscale value of each pixel area corresponds to the number of point clouds mapped to the reference plane by the corresponding grid in the initial point cloud information; Based on the gridded initial point cloud information, non-building point cloud information and building point cloud information are obtained; Mapping the non-building point cloud information to the reference plane to obtain a non-building image; Mapping the building point cloud information to the reference plane to obtain a building facade image; Based on the non-building image and the building facade image, subtract the plane feature image to obtain a first residual building image; Based on the first remaining building image, obtaining remaining building point cloud information; The remaining building point cloud information is merged with the building point cloud information to obtain target building point cloud information.
2. The building facade extraction method according to claim 1, characterized in that: The plane feature image includes a quantity feature image and a normal vector feature image; The step of obtaining the plane feature image corresponding to the reference plane based on the gridded initial point cloud information includes: Mapping the gridded initial point cloud information to the reference plane, determining the pixel value of the corresponding pixel area based on the number of point clouds of each grid mapped to the reference plane, and creating the quantity feature image; Based on a preset scale, normal vector calculation is performed on each point cloud of the initial point cloud information after gridding to obtain a normal vector value corresponding to each point cloud, a point cloud satisfying a preset normal vector angle condition is determined as a valid point cloud, a pixel value of a corresponding pixel area is determined based on the number of valid point clouds mapped from each grid to the reference plane, and the normal vector feature image is created; The quantity feature image and the normal vector feature image are combined to obtain the plane feature image.
3. The building facade extraction method according to claim 1, characterized in that: The step of merging the quantity feature image with the normal vector feature image to obtain the plane feature image includes: Based on the first weight, the pixel value of each pixel area in the quantity feature image is adjusted, and based on the second weight, the pixel value of each pixel area in the normal vector feature image is adjusted, and the adjusted quantity feature image and the normal vector feature image are merged to obtain the plane feature image.
4. The building facade extraction method according to claim 1, characterized in that: The obtaining the remaining building point cloud information based on the first remaining building image includes: Determine a pixel area in the first remaining building image that meets the building screening condition as a target pixel area, and determine a second remaining building image based on a plurality of the target pixel areas; Based on the second remaining building image, the remaining building point cloud information is obtained.
5. The building facade extraction method according to claim 4, characterized in that: The pixel area meets the building screening conditions including: The pixel value of the pixel area reaches a preset pixel value range; and / or, The pixel region and other adjacent pixel regions form a continuous region, and the area of the continuous region reaches a preset area range.
6. The building facade extraction method according to claim 1, characterized in that: Based on the gridded initial point cloud information, building point cloud information is obtained, including: Based on the gridded initial point cloud information, the non-building point cloud information is removed to obtain remaining point cloud information; Divide the multiple grids where the remaining point cloud information is located into multiple grid layers and multiple grid columns, wherein the multiple grid layers are distributed along the Z axis, each of the grid layers includes multiple grids distributed parallel to the reference plane, and each of the grid columns includes multiple grids distributed along the Z axis, and the Z axis is perpendicular to the reference plane; Analyze each of the grid layers from top to bottom along the Z axis, and determine, from a plurality of first grids in the grid layer, the first grid that meets the building analysis condition as the second grid, wherein the first grid and the determined second grid are located in different grid columns; The building point cloud information is obtained according to all the determined second grids.
7. The building facade extraction method according to claim 6, characterized in that: The first grid satisfies the building analysis condition including: The maximum height of the point cloud in the first grid reaches a corresponding building height range, wherein the building height range is associated with the height of the grid layer where the first grid is located.
8. The building facade extraction method according to claim 6, characterized in that: The first grid satisfies the building analysis condition including: Based on a preset scale, normal vectors are calculated for each point cloud of the first grid to obtain normal vector values corresponding to each point cloud; the normal vector values corresponding to each point cloud of the first grid meet corresponding building angle conditions.
9. The building facade extraction method according to claim 8, characterized in that: The preset scale includes a first scale and a second scale, and the normal vector value includes a first normal vector value corresponding to the first scale and a second normal vector value corresponding to the second scale.
10. The building facade extraction method according to claim 8, characterized in that: The normal vector value corresponding to each point cloud of the first grid meets the corresponding building angle condition, including: An average value of the sum of the first normal vector value and the second normal vector value corresponding to each point cloud of the first grid is within a preset angle threshold range.
11. The building facade extraction method according to claim 1, characterized in that: The non-building point cloud information is obtained based on the initial point cloud information after gridding, including: determining the non-building grid according to the absolute height and relative height of the point cloud in each grid, and obtaining the non-building point cloud information according to the non-building grid, the absolute height is the maximum height of the point cloud on the Z axis, and the relative height is the maximum height and minimum height of the point cloud on the Z axis.
12. The building facade extraction method according to claim 5, characterized in that: Before obtaining the remaining building point cloud information based on the first remaining building image, the method further includes: performing image processing on the second remaining building image, wherein the image processing includes at least one of dilation, erosion, opening and closing operations, and filtering.
13. A building facade extraction device, characterized in that: include: Data receiving module, used to obtain initial point cloud information; An image mapping module, configured to obtain a plane feature image corresponding to a reference plane based on the gridded initial point cloud information, wherein the plane feature image includes a plurality of pixel areas, and a grayscale value of each pixel area corresponds to the number of point clouds mapped to the reference plane by a corresponding grid in the initial point cloud information; A point cloud analysis module, for obtaining non-building point cloud information and building point cloud information based on the gridded initial point cloud information, mapping the non-building point cloud information to the reference plane to obtain a non-building image, and mapping the building point cloud information to the reference plane to obtain a building facade image; An image processing module, configured to perform subtraction processing on the plane feature image based on the non-building image and the building facade image to obtain a first residual building image; A point cloud conversion module, used for obtaining remaining building point cloud information based on the first remaining building image; The point cloud merging module is used to merge the remaining building point cloud information with the building point cloud information to obtain the target building point cloud information.
14. A building facade extraction system, characterized in that: include: A collector and a processor, wherein the collector is used to scan to collect point cloud data, and the processor is used to execute the building facade extraction method as described in any one of claims 1 to 12.