A point cloud building contour regularization generation method and system based on adaptive volume projection

Abstract

This application provides a method and system for regularizing point cloud building outlines based on adaptive volume projection, relating to the fields of computer graphics and remote sensing data processing. The method includes acquiring the original 3D point cloud of the target building and performing preprocessing; conducting vertical structural feature analysis on the preprocessed point cloud to obtain Z-axis point cloud density features; dividing the building vertically into structural segments such as the podium segment, standard segment, and tower crown segment based on the density features; determining the projection sampling range for each segment type using a differentiated adaptive volume projection strategy of bidirectional shrinkage, upward point borrowing, and downward fusion; projecting the point cloud within each segment along the Z-axis into a 2D projected point cloud and extracting the initial building outline; then obtaining the main wall outline through penetrating morphological filtering using strong erosion and springback expansion; and generating the regularized final building vector outline through orthogonal shaping. This method improves the accuracy of building outline extraction, ensuring that the outline conforms to the actual geometric features of the building.

Description

Technical Field

[0001] This application relates to the fields of computer graphics and remote sensing data processing technology, and in particular to a method and system for regularizing point cloud building outlines based on adaptive volume projection. Background Technology

[0002] With the advancement of real-world 3D construction, extracting building vector contours from UAV oblique photogrammetry models has become a crucial requirement. In applications such as urban planning, building information modeling (BIM), and UAV facade inspection, it is necessary to extract regularized vector contours of buildings from point cloud data. However, point clouds generated by oblique photogrammetry suffer from issues such as concave corner adhesion, interference from ancillary facilities, and bottom occlusion. How to automatically extract regularized contours that conform to the actual geometric characteristics of buildings is a pressing technical challenge that needs to be addressed.

[0003] Existing techniques typically employ a fixed-height slicing method combined with contour extraction algorithms. This method selects a fixed-height layer in the building point cloud for horizontal slicing, obtaining a two-dimensional point set at that height. Then, a boundary extraction algorithm is used to construct the contour line of the point set. Finally, the contour line is simplified and rectangularized to obtain a regularized building vector contour. For multi-story buildings, the final contour is generated by repeating the slicing operation at multiple fixed heights or by simply averaging the results of multiple slices.

[0004] However, the technical problem with the above-mentioned existing technology is that the fixed height slicing strategy is difficult to perceive the structural changes in the vertical direction of the building. When the bottom of the building is blocked by trees, resulting in missing point clouds, a broken contour will be generated. When the exterior facade is attached with auxiliary facilities such as air conditioning units and pipes, the algorithm will include these facilities in the contour boundary, causing the extracted contour lines to deviate from the actual wall position and contain a lot of jitter, which makes it difficult to meet the accuracy requirements of fine mapping and close flight applications. Summary of the Invention

[0005] This application provides a method and system for generating regularized point cloud building contours based on adaptive volume projection, which is used to accurately extract regularized vector contours of buildings and improve the accuracy of contour extraction.

[0006] Firstly, this application provides a method for regularizing point cloud building contours based on adaptive volume projection, applied to a point cloud building contour regularization generation system. The method includes: acquiring the original 3D point cloud of the target building and preprocessing the original 3D point cloud; performing vertical structural feature analysis on the preprocessed original 3D point cloud to obtain point cloud density features along the Z-axis; based on the point cloud density features, dividing the target building into multiple structural segments in the vertical direction and determining the matching segment type for each structural segment; applying differentiated adaptive volume projection strategies to determine the projection sampling range of each structural segment according to the segment type; projecting the original 3D point cloud within each structural segment along the Z-axis to map it into a corresponding 2D projected point cloud, and extracting the initial building contour using a contour-grabbing algorithm on the 2D projected point cloud; sequentially performing penetrating morphological filtering processing of the initial building contour using strong erosion and springback expansion operations to obtain the filtered main wall contour; and performing orthogonal shaping processing on the main wall contour to generate a regularized final building vector contour.

[0007] By employing the above technical solution, the original 3D point cloud is first preprocessed to remove noise. Then, the Z-axis point cloud density features are obtained through vertical structural feature analysis. This allows for the division of different structural segments and matching of corresponding types, achieving accurate perception of vertical structural changes in the building. A differentiated adaptive volume projection strategy is used for different segments to specifically avoid point cloud defects in each segment. After projecting the point cloud into a 2D point cloud, the initial contour is extracted. Then, through-penetration morphological filtering is used to remove interference from ancillary facilities and fill contour breaks. Finally, orthogonal shaping generates a regular contour. Each step is interconnected and works in tandem, effectively solving the technical challenges of fixed slices being unable to adapt to changes in building structure, contours being prone to breakage, and being susceptible to interference from ancillary facilities. This significantly improves the accuracy, completeness, and regularity of building contour extraction, ensuring that the contour conforms to the building's true geometric features.

[0008] In conjunction with some embodiments of the first aspect, in some embodiments, based on the point cloud density characteristics, the step of dividing the target building into multiple structural segments in the vertical direction and determining the segment type matching each structural segment specifically includes: according to the point cloud density characteristics, counting the number of point clouds in each height layer along the Z-axis with a preset step size, or the projected area of ​​the point clouds in each height layer projected onto the horizontal plane, generating a Z-axis density histogram; based on the density abrupt change points in the Z-axis density histogram, dividing the target building into structural segments of the segment type that at least include a podium segment, a standard segment, and a tower crown segment.

[0009] By employing the aforementioned technical solution, a Z-axis density histogram is generated by statistically analyzing the number of point clouds or their projected area along the Z-axis based on point cloud density characteristics. This histogram visually presents the differences in point cloud distribution in the vertical direction, providing accurate data support for structural segmentation. Structural segments are defined using density abrupt change points in the density histogram, clearly classifying the podium, standard, and tower sections. This allows for precise identification of the structural characteristics of different functional areas of the building, avoiding the treatment of the entire building as a single structure. This segmentation method aligns with the actual structural patterns of buildings, laying the foundation for subsequent applications of differentiated projection strategies and effectively improving the targeting of contour extraction.

[0010] In conjunction with some embodiments of the first aspect, in some embodiments, the step of dividing the target building into structural segments, including at least a podium segment, a standard segment, and a crown segment, based on density abrupt change points in the Z-axis density histogram, specifically includes: calculating the point cloud change rate between adjacent height layers in the Z-axis density histogram, the point cloud change rate including the point cloud quantity change rate or the projected area change rate; when the absolute value of the point cloud change rate exceeds a preset abrupt change threshold, determining the matching location as a density abrupt change point; identifying a first abrupt change point and a second abrupt change point, wherein the first Z-axis height of the first abrupt change point is lower than the second Z-axis height of the second abrupt change point; dividing the height range from the bottom of the building to the first abrupt change point into a podium segment, the podium segment... The characteristic is that the corresponding projected area or point cloud quantity is the largest among all segments; the height range between the first mutation point and the second mutation point is divided into a standard segment, the characteristic of which is that the change rate of point cloud quantity or the change rate of projected area is less than a preset stability threshold; the height range from the second mutation point to the top of the target building is divided into a tower crown segment, the characteristic of which is that the corresponding projected area or point cloud quantity is reduced by more than a preset reduction ratio relative to the standard segment; within each divided height range, the contour similarity of multiple height layer slices is calculated; when the contour similarity between consecutive preset number of height layers is higher than a preset similarity threshold, it is determined that the target interval corresponding to the consecutive preset number of height layers has geometric stability, and the target interval is aggregated into a structural segment.

[0011] By employing the aforementioned technical solution, calculating the rate of change of point clouds at adjacent height levels and setting abrupt change thresholds, density abrupt change points can be accurately located, avoiding misjudgments or omissions. By distinguishing between the first and second abrupt change points and combining the differences in point cloud features of each segment, the podium segment, standard segment, and tower crown segment can be clearly delineated, ensuring accurate segment boundaries. Simultaneously, by calculating the similarity of the height layer slice outlines, geometrically stable intervals are aggregated into a single structural segment, effectively avoiding the problem of overly fine or coarse segmentation caused by local point cloud fluctuations. This process, through the coordinated use of data statistics, threshold judgment, and similarity verification, achieves precise and rational structural segmentation, further improving the accuracy of subsequent projection sampling and outline extraction, ensuring that the outlines of each segment closely match the actual structural height of the building.

[0012] In conjunction with some embodiments of the first aspect, in some embodiments, the step of determining the projection sampling range of each structural segment by applying a differentiated adaptive volume projection strategy according to the segment type specifically includes: if the segment type is a standard segment, a bidirectional shrinkage strategy is executed on the standard segment, the bidirectional shrinkage strategy including: removing point cloud data corresponding to a first preset ratio at the top and bottom of the standard segment along the Z-axis direction, and extracting the retained middle area point cloud as the first projection sampling range; if the segment type is a podium segment, an upward borrowing point strategy is executed on the podium segment, the upward borrowing point strategy... The method includes: extending the top sampling boundary of the podium segment upwards along the Z-axis by a second preset proportion into the adjacent upper structure segment, and defining the point cloud contained between the bottom boundary of the podium segment and the extended top boundary as the second projection sampling range; if the segment type is a tower crown segment, performing a downward fusion strategy on the tower crown segment, the downward fusion strategy including: extending the bottom sampling boundary of the tower crown segment downwards along the Z-axis by a third preset proportion into the adjacent lower structure segment, and defining the point cloud contained between the extended bottom boundary and the top boundary of the tower crown segment as the third projection sampling range.

[0013] By adopting the above technical solutions, a bidirectional shrinkage strategy is implemented for the standard segment to remove easily disturbed point clouds at the top and bottom, retaining the stable middle region and ensuring the accuracy of projection sampling. For the podium segment, an upward point borrowing strategy is implemented to compensate for potential gaps in the top point cloud, preventing contour breaks. For the tower crown segment, a downward fusion strategy is implemented to supplement the bottom point cloud, solving the contour distortion problem caused by sparse point clouds in the tower crown segment. Different segments employ projection strategies adapted to their structural characteristics, achieving adaptive adjustment of the projection sampling range. This effectively avoids the shortcomings of fixed projection ranges, which cannot adapt to the point cloud characteristics of different segments, ensuring the integrity and accuracy of the two-dimensional projected point clouds of each segment and providing high-quality data support for subsequent initial contour extraction.

[0014] In conjunction with some embodiments of the first aspect, in some embodiments, the step of performing a penetrating morphological filtering process of strong erosion and springback expansion operations sequentially on the initial building outline to obtain the filtered main wall outline specifically includes: determining a first contraction distance, which is greater than a preset building ancillary facility thickness threshold; performing an inward strong erosion operation on the initial building outline using the first contraction distance to shrink inward and eliminate protruding ancillary facility outlines with a thickness less than the building ancillary facility thickness threshold, and generating an inwardly offset core wall outline; determining a first expansion distance matching the first contraction distance; performing an outward springback expansion operation on the core wall outline using the first expansion distance to drive the core wall outline to spring back outward to the main building entity scale position before the ancillary facilities are stripped, obtaining a transitional main outline; and performing a buffer offset process on the transitional main outline using a preset micro-compensation offset to compensate for the wall thickness and corner reduction errors caused by the strong erosion operation, generating the main wall outline that fits the boundary of the real wall point cloud.

[0015] By employing the aforementioned technical solution, a first contraction distance exceeding the thickness threshold of building ancillary facilities is set, and a powerful corrosion operation is performed. This completely removes the protruding contours of ancillary facilities such as air conditioning units and pipes, accurately extracting the core wall contour. Then, a rebound expansion operation is performed using a matching first expansion distance, causing the core wall contour to spring back to the dimensions of the main building. Combined with micro-compensation offset processing, this compensates for the wall thickness and corner reduction errors caused by the corrosion operation. The synergistic effect of corrosion, expansion, and compensation operations effectively eliminates interference from ancillary facilities, avoids contour distortion, and fills in contour breaks caused by missing point clouds. Ultimately, a main wall contour that closely matches the boundary of the real wall point cloud, without redundant interference or significant distortion, is obtained, improving the purity and accuracy of the contour.

[0016] In conjunction with some embodiments of the first aspect, in some embodiments, the step of orthogonally shaping the main wall profile to generate a regularized final building vector profile specifically includes: calculating the minimum bounding rectangle of the main wall profile and determining the building principal axis direction of the building profile based on the orientation of the minimum bounding rectangle; rotating and aligning the main wall profile as a whole to the principal axis coordinate system defined by the building principal axis direction; in the principal axis coordinate system, identifying line segments whose direction and the angle between the coordinate axis are within a preset tolerance angle range, and forcibly projecting and straightening the identified line segments into horizontal or vertical orthogonal line segments; traversing the geometric features of the straightened main wall profile, and when the size of a local geometric feature is less than a preset first length threshold, performing noise reduction and smoothing operations on local micro-jitter features; when the size of a structural groove formed by a local geometric feature is greater than a preset second length threshold, retaining the structural groove and performing orthogonal normalization processing; performing an inverse rotation transformation on the profile after feature filtering and orthogonal normalization processing to restore it to the original coordinate system of the initial input point cloud, generating the regularized final building vector profile.

[0017] By adopting the above technical solution, the main axis direction of the building is first determined and the outline of the main wall is aligned to the main axis coordinate system to ensure the consistency of the outline shaping benchmark. Line segments that meet the tolerance angle are forcibly straightened into orthogonal line segments to conform to the right-angle geometric features of the building. Local minor vibrations are smoothed and noise is reduced to eliminate redundant interference. Structural grooves that meet the requirements are retained and standardized, balancing outline regularity with the actual building structure. Finally, the outline is restored to the original coordinate system through inverse rotation transformation, achieving orthogonality and standardization. The entire process, through the synergistic effect of benchmark alignment, line segment correction, feature selection, and coordinate restoration, effectively solves the problems of skewness, vibration, and non-orthogonality of the main wall outline. The generated regularized vector outline conforms to the building's geometric features, improving the regularity and standardization level of the outline.

[0018] In conjunction with some embodiments of the first aspect, in some embodiments, after the step of orthogonalizing and shaping the main wall outline to generate a regularized final building vector outline, the method further includes: calculating the fitting deviation between the final building vector outline and the original three-dimensional point cloud; when the fitting deviation is less than a preset deviation threshold, outputting the final building vector outline as a vector format file; when the fitting deviation is greater than or equal to the preset deviation threshold, adjusting the preset tolerance angle range or the slight compensation offset, and re-performing the orthogonalization and shaping process.

[0019] By employing the above technical solution, the fitting deviation between the final building vector contour and the original 3D point cloud is calculated, allowing for precise judgment of whether the contour extraction accuracy meets the standards. When the deviation is less than a preset threshold, a vector file is output to ensure the accuracy of the results meets the requirements. When the deviation exceeds the standard, the preset tolerance angle range or a small amount of compensation offset is adjusted, and orthogonal shaping is re-executed, forming a closed-loop optimization mechanism. This process effectively controls the contour accuracy through deviation verification and specifically addresses potential deviation problems during orthogonal shaping through parameter adjustment, avoiding contour distortion caused by unreasonable parameters. This further improves the accuracy and reliability of the final building vector contour, ensuring a high degree of fit between the contour and the original point cloud, meeting the needs of refined applications.

[0020] In a second aspect, this application provides a point cloud building outline regularization generation system, which includes: one or more processors and a memory; the memory is coupled to the one or more processors, and the memory is used to store computer program code, which includes computer instructions, and the one or more processors call the computer instructions to cause the point cloud building outline regularization generation system to perform the method described in the first aspect and any possible implementation thereof.

[0021] Thirdly, this application provides a computer-readable storage medium including instructions that, when executed on a point cloud building outline regularization generation system, cause the point cloud building outline regularization generation system to perform the method described in the first aspect and any possible implementation thereof.

[0022] Fourthly, this application provides a computer program product, including a computer program that, when run on a point cloud building outline regularization generation system, causes the point cloud building outline regularization generation system to perform the method described in the first aspect and any possible implementation thereof.

[0023] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0024] 1. By employing core technologies such as segmenting the building's vertical structure based on point cloud density features, applying differentiated adaptive volume projection strategies to different segments, and combining penetrating morphological filtering to remove interference from ancillary facilities and orthogonal shaping, this technology effectively solves the technical problems in existing technologies, such as fixed-height slices failing to perceive changes in the building's vertical structure, contours being prone to breakage due to occlusion, and being susceptible to offset and jitter caused by interference from ancillary facilities. This results in a significant improvement in the accuracy, completeness, and regularity of building contour extraction, ensuring that the extracted contours accurately match the building's true geometric features and generating a regularized vector contour that meets the requirements without manual intervention.

[0025] 2. By employing techniques such as calculating the rate of change of point cloud to locate density abrupt change points, combining the features of each segmented point cloud to distinguish between the podium segment, standard segment, and tower crown segment, and using contour similarity verification to aggregate geometrically stable intervals, this technology effectively solves the technical problems in existing technologies, such as blurred structural segment boundaries, excessively fine / coarse segmentation, and inability to accurately match the actual building structure. This achieves more precise and rational vertical structural segmentation of the building, accurately identifies structural differences in different functional areas of the building, provides a reliable basis for subsequent differentiated projection sampling, and further improves the targeting and accuracy of contour extraction.

[0026] 3. By employing a differentiated adaptive volume projection strategy that applies bidirectional inward contraction, upward borrowing, and downward fusion to the standard section, podium section, and tower crown section respectively, and by precisely adjusting the projection sampling range of each segment, the technical problem of the fixed projection range in the existing technology being unable to adapt to the point cloud characteristics of different building segments and being prone to distortion of the projected point cloud due to missing or interference point clouds is effectively solved. This achieves precise adaptation of the projection sampling range of each segment to its own structural features, ensuring the integrity and accuracy of the two-dimensional projected point cloud, providing high-quality data support for subsequent initial contour extraction, and reducing contour deviations caused by the projection process. Attached Figure Description

[0027] Figure 1 This is a scene diagram of the point cloud building outline regularization generation method based on adaptive volume projection in the embodiments of this application;

[0028] Figure 2 This is a flowchart illustrating a point cloud building outline regularization generation method based on adaptive volume projection in an embodiment of this application.

[0029] Figure 3 This is a schematic diagram of the vertical structure analysis of a building in an embodiment of this application;

[0030] Figure 4 This is a schematic diagram comparing the effects of contour filtering and shaping on the standard floor and roof floor of the building in the embodiments of this application;

[0031] Figure 5 This is a schematic diagram comparing the effects of extracting contours using traditional single-layer slicing and adaptive volume projection strategies in the embodiments of this application;

[0032] Figure 6 This is a schematic diagram of a physical device structure of a point cloud building outline regularization generation system in this application embodiment. Detailed Implementation

[0033] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. The singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to and includes any or all possible combinations of one or more of the listed items.

[0034] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0035] To facilitate understanding, the method provided in this implementation is described in a scenario below. Please refer to [link / reference]. Figure 1 This is a scene diagram of a point cloud building outline regularization generation method based on adaptive volume projection in an embodiment of this application.

[0036] Figure 1 This demonstrates a complete automated processing scenario from raw 3D point clouds to regularized building vector contours, intuitively showcasing the core technical path of this method:

[0037] The input box on the left displays the original 3D point cloud of the target building. This is high-density point cloud data acquired by UAV oblique photography or laser scanning, containing real-world scene information such as the building itself, trees, and ancillary facilities. After preprocessing the original point cloud, the point cloud density or projected area is statistically analyzed along the Z-axis (building height direction) to generate a Z-axis density histogram. The vertical structural changes of the building are identified through density abrupt change features, providing a basis for subsequent segmented processing.

[0038] Based on the density analysis results, the building is automatically divided into multiple vertical structural segments:

[0039] Segment X (e.g., roof / tower crown segment): Adaptive projection strategy A is adopted to fuse the lower layer point cloud data downwards, determine the projection sampling range, and solve the problem of sparse point cloud at the top;

[0040] Segment Y (e.g., standard layer / main body segment): Adaptive projection strategy B is adopted to bidirectionally shrink and remove the top and bottom interference layers, retaining the middle stable area as the sampling range to avoid interlayer adhesion;

[0041] Segment Z (e.g., base / podium segment): Adaptive projection strategy C is adopted to extend the sampling range upwards by borrowing points, compensating for point cloud gaps caused by bottom occlusion. The sampling ranges of different segments are adaptively matched through a differentiated strategy to ensure that each segment can obtain clean and complete projection data.

[0042] The 3D point cloud within each segment is projected along the Z-axis and mapped to a 2D point set. An initial building contour is extracted using a contour-grabbing algorithm. At this point, the contour still contains protrusions and jitters caused by auxiliary facilities such as air conditioning units and pipes. Strong erosion and springback expansion operations are sequentially performed on the initial contour, and the wall thickness reduction error caused by erosion is further corrected to obtain the main wall contour that fits the real wall surface. Then, the main wall contour is orthogonalized: the main axis of the building is aligned, orthogonal line segments are straightened, minor jitters are smoothed out, and structural grooves are preserved. Finally, a regularized building vector contour is generated, and a vector file that meets the requirements of surveying and mapping specifications and engineering applications is output.

[0043] The following describes the process of the method provided in this implementation, using the above scenario as an example. Please refer to [link / reference]. Figure 2 This is a flowchart illustrating a point cloud building outline regularization generation method based on adaptive volume projection in an embodiment of this application.

[0044] S101. Obtain the original 3D point cloud of the target building and preprocess the original 3D point cloud;

[0045] The target building refers to the single building whose regularized vector contour is to be extracted. It may include urban building objects such as commercial complexes, residential buildings, and office buildings. It is used to represent the core processing object of this contour generation task. The original 3D point cloud refers to the set of 3D discrete points containing the target building and its surrounding environment, which are collected by UAV oblique photography, ground laser scanning or airborne LiDAR and other equipment. It is used to represent the unprocessed raw measurement data.

[0046] The system first reads the original 3D point cloud file of the target building through a data interface. This file is typically stored in formats such as LAS / LAZ, PLY, or PCD, and contains the 3D coordinates (X, Y, Z), reflection intensity, color, and other attributes of each point. Then, a preprocessing workflow is executed, including:

[0047] Denoising: Statistical filtering or radius filtering algorithms are used to identify and remove isolated noise points (such as floating objects in the air or sensor error points) that deviate from the main point cloud, so as to avoid noise interference with subsequent density analysis and contour extraction.

[0048] Ground point removal: Based on cloth simulation filtering or slope analysis algorithms, ground points and non-ground points are separated, and only non-ground point clouds belonging to the building body are retained, eliminating irrelevant interference such as ground vegetation and roads;

[0049] Voxel downsampling: The remaining point cloud is downsampled with a preset voxel size (e.g., 0.05m×0.05m×0.05m) to reduce the amount of data while maintaining the geometric features of the building and improve the efficiency of subsequent calculations;

[0050] Coordinate centering: The point cloud coordinates are translated to a local coordinate system with the building's centroid as the origin, eliminating the impact of global coordinate offset on subsequent vertical structure analysis.

[0051] By preprocessing to eliminate noise, ground interference, and redundant data in the original point cloud, a clean and well-organized point cloud of the building structure is obtained, providing a reliable data foundation for subsequent vertical structural feature analysis and avoiding structural segmentation errors or contour extraction deviations caused by data defects.

[0052] S102. Perform vertical structural feature analysis on the preprocessed original 3D point cloud to obtain the point cloud density features along the Z-axis.

[0053] The Z-axis direction refers to the vertical direction perpendicular to the building's ground level, and is used to represent the building's vertical dimension.

[0054] This step is executed immediately after the S101 preprocessing is completed. It is suitable for high-rise buildings or complex building scenarios with vertical structural changes (such as podium-standard floor-tower crown) and is a prerequisite for achieving adaptive volumetric projection. Specifically, the system divides the preprocessed point cloud into continuous height-layer slices along the Z-axis with a preset step size (such as 0.2m or 0.5m), and performs two optional statistics on each height-layer slice:

[0055] One method is point cloud quantity statistics: counting the total number of point clouds contained in the current height layer slice, which directly reflects the density of point clouds at that height;

[0056] Another method is projected area statistics: the point cloud within the current height layer slice is projected onto a horizontal plane (XY plane), and the area enclosed by the projected point set is calculated using a convex hull algorithm, indirectly reflecting the building's plan shape at that height. The statistical results for each height layer are arranged in order of Z-axis height to generate a Z-axis density histogram. The horizontal axis of this histogram represents the Z-axis height, and the vertical axis represents the number of point clouds or projected area, visually presenting the density change trend in the vertical direction of the building, such as the larger projected area of ​​the podium section, the stable density of the standard section, and the sharp decrease in density of the tower crown section.

[0057] Detailed vertical structure analysis diagrams can be found later. Figure 3 This step simplifies the 3D point cloud into a one-dimensional density distribution through vertical structural feature analysis, intuitively revealing the layering pattern of the building's vertical structure. This provides a quantitative basis for subsequent structural segmentation and solves the problem that traditional fixed-height slicing cannot perceive changes in the building's vertical structure.

[0058] S103. Based on the point cloud density characteristics, the target building is divided into multiple structural segments in the vertical direction, and the segment type matching each structural segment is determined.

[0059] This step is executed immediately after the system completes the S101 original point cloud preprocessing and the S102 vertical structure feature analysis. It is applicable to all urban building point cloud contour extraction scenarios obtained by UAV oblique photography, especially to complex building scenarios such as high-rise buildings with three-segment structures such as podiums, standard floors, and tower crowns, commercial and residential complexes, and large public buildings. It is also applicable to low-quality point cloud scenarios where the bottom is obscured by trees, there is adhesion between floors, and the top point cloud is sparse. It is a core preliminary step to achieve adaptive volume projection and solve the problem that traditional fixed slicing cannot perceive changes in building structure.

[0060] Specifically, this step is performed after the system completes the preprocessing of the original 3D point cloud and obtains the point cloud density features along the Z-axis, and before the adaptive volume projection strategy is executed. It is the core link to realize intelligent recognition and accurate layering of the building's vertical structure, and directly determines the rationality of the subsequent projection sampling range and the contour extraction accuracy.

[0061] The system first reads the pre-processed clean building point cloud. Based on the configured preset step size (e.g., typically 0.5m, 1.0m, or 1.5m, which can be adaptively adjusted according to building height and accuracy requirements), it slices the target building upwards along the positive Z-axis, creating continuous, equally spaced height layers. This ensures that each height layer is a horizontal, thin slice that does not overlap and completely covers the entire building height. Within each height layer, the system performs two statistical methods in parallel: first, it counts the number of point clouds contained within the current height layer, i.e., it counts all 3D points falling within the space of that thin layer to obtain the point cloud density value for that layer; second, it projects all the point clouds within that height layer vertically along the Z-axis onto a horizontal plane, calculates the projected area of ​​the enclosed region formed after projection, and obtains the planar coverage value for that layer. The system arranges the number of point clouds or projected area corresponding to all height layers in order of height, and generates a continuous and smooth Z-axis density histogram with height as the horizontal axis and statistical value as the vertical axis. This histogram can intuitively reflect the volume changes, density changes and structural boundary characteristics of the building from bottom to top. For example, the podium section presents a high amplitude platform, the standard section presents a stable straight line, and the tower crown section presents a rapidly descending curve.

[0062] After generating the Z-axis density histogram, the system enters the density abrupt change point detection stage. Starting from the first height layer, it sequentially calculates the point cloud change rate between every two adjacent height layers. The change rate calculation formula is (current layer value - previous layer value) / previous layer value. It can be calculated based on the number of point clouds or the projected area, or both methods can be calculated in parallel and the results weighted to improve detection robustness. The system compares the absolute value of the calculated change rate with a preset abrupt change threshold (e.g., set to 20%, 30%, or 40%, adaptively adjusted according to the building type). When the absolute value of the change rate is greater than the threshold, the height position between the current two layers is determined as a density abrupt change point, indicating that the building undergoes a sudden change in cross-sectional area or structural form at this height, such as the tapering of the podium, the start of the tower, or the capping of the tower—key structural boundaries. Within the entire height range, the system identifies at least two key boundary positions: the first abrupt change point and the second abrupt change point. It strictly verifies the height order, ensuring that the Z-axis height corresponding to the first abrupt change point is lower than the Z-axis height corresponding to the second abrupt change point, avoiding a reversal of the segmentation order.

[0063] After locating the mutation point, the system performs a three-segment division based on the height range: the interval from the bottom of the building to the first mutation point is divided into the podium segment, which naturally has the largest projected area and the most point cloud, corresponding to the large commercial base, lobby, and multi-story podium structure at the bottom of the building. For example, in a common shopping mall + tower complex, the cross-section of the podium on floors 1-5 is much larger than that of the upper tower. The interval from the first mutation point to the second mutation point is divided into the standard segment, in which the rate of change of the number of point clouds or the rate of change of the projected area is consistently less than the preset stable threshold (e.g., 5%-10%), showing stable density and almost unchanged outline shape, corresponding to the standard residential and office floors in the middle of the building, which is the most regular area of ​​the building's main structure and facade. The interval from the second mutation point to the top of the building is divided into the crown segment, in which the projected area or the number of point clouds in this segment is reduced by more than the preset reduction ratio (e.g., 30%, 50%) compared to the standard segment, showing rapid narrowing of the cross-section and sparse point clouds, corresponding to the top of the building's machine room, decorative crown, and parapet wall structure.

[0064] To avoid false segmentation caused by local point cloud noise and minor ancillary structures, the system further performs geometric stability verification and segment aggregation operations based on the initial three-segment division: Within each divided height interval, multiple height layer slices are uniformly extracted, the two-dimensional contours of each slice are extracted, and the contour similarity between adjacent slices is calculated, which can be quantified by contour overlap area ratio, cosine similarity, etc. When the contour similarity between a consecutive preset number (e.g., 2 layers, 3 layers) of height layer slices is higher than a preset similarity threshold (e.g., 90%, 95%), the consecutive interval is determined to have geometric stability, that is, the contour shape is consistent and there is no obvious deformation, belonging to the same structural segment. The system aggregates all consecutive height layers that meet the geometric stability into an independent, complete, and non-redundant structural segment, and marks the corresponding segment type (podium segment, standard segment, tower crown segment), finally outputting the precise vertical segmentation result of the building, providing a unique, stable, and realistic layering basis for subsequent differentiated adaptive volume projection.

[0065] This step overcomes the shortcomings of traditional fixed-height slicing, which cannot detect changes in the vertical structure of buildings. It achieves adaptive recognition of building structures through Z-axis density histogram and mutation detection, accurately distinguishing three different morphological regions: podium, standard section, and tower crown. This solves the problems of contour breakage and distortion caused by bottom occlusion, interlayer adhesion, and top sparseness. At the same time, contour similarity verification ensures the geometric stability within the segments, avoiding false layering and over-segmentation. This lays the structural foundation for subsequent targeted repair of occlusion, filtering of interference, and contour closure, significantly improving the robustness and accuracy of complex building contour extraction.

[0066] S104. Based on the segmentation type, apply differentiated adaptive volume projection strategies to determine the projection sampling range of each structural segment.

[0067] Among them, the segment type refers to the vertical structural category of the building obtained by dividing it according to the Z-axis density histogram and density change points. Specifically, it can include three types: standard segment, podium segment, and tower crown segment. The standard segment refers to the main structural segment with a stable outline and dense point cloud in the middle of the building, but the top and bottom are prone to inter-layer adhesion and floor noise. The podium segment refers to the large cross-section structural segment at the bottom of the building, which is prone to point cloud loss and outline breakage caused by trees and obstructions. The tower crown segment refers to the end structural segment at the top of the building with sparse point cloud, narrow cross section, and easy to have an unclosed outline.

[0068] Specifically, this step is performed after the system completes the vertical structural segmentation of the building and determines whether each section belongs to the standard section, podium section, or tower crown section. It is executed before the three-dimensional point cloud is projected into a two-dimensional projected point cloud and the initial building outline is extracted. It is the core step in the whole method to achieve "adaptability" and "anti-interference". Its function is to tailor the optimal projection data range for the unique point cloud defects of different structural segments, and avoid outline distortion caused by occlusion, adhesion, and sparsity from the data source.

[0069] When the system enters this step, it already has the Z-axis start height, end height, segment type label, and pre-processed clean 3D point cloud for each structural segment. This step will judge and process each segment one by one, without using uniform fixed parameters, to ensure that each type of structure gets the most suitable projection sampling range.

[0070] When the system determines that the current structural segment being processed is a standard segment, it immediately initiates a bidirectional shrinkage strategy. Standard segments, as the main floors in the middle of a building, are characterized by a large number of floors, regular outlines, and high point cloud density. However, they also exhibit significant industry-wide problems: the boundaries between floors are susceptible to errors from oblique photogrammetry modeling, leading to inter-floor adhesion, mixed floor point clouds, ceiling noise, and concave corner distortion. Directly projecting the entire segment would introduce these interfering data into the 2D projection result, resulting in jagged edges, misalignments, and unevenness in the outline. Therefore, the system reads a pre-configured first preset ratio, which is typically set based on the building's floor height and point cloud quality, with typical values ​​of 20%–30% for the top and 20%–30% for the bottom, and can be adaptively adjusted according to the actual scene. The system uses the original Z-axis interval of the standard segment as a reference, and simultaneously cuts out a first preset proportion of point cloud data from top to bottom and from bottom to top, retaining only the core area point cloud at the vertical center of the segment. This central area point cloud is far from the floor boundaries, unaffected by floor adhesion and ceiling noise, and has a uniform point cloud distribution, flat walls, and stable geometric features, thus most realistically reflecting the exterior facade of the main building in the standard segment. The system formally defines the height range corresponding to this purified central area point cloud as the first projection sampling range, serving as the sole valid data source for subsequent standard segment projections, thereby eliminating contour distortion caused by inter-floor adhesion at its source.

[0071] When the system determines that the current structural segment being processed is a podium segment, it immediately initiates an upward point borrowing strategy. The podium segment is located at the very bottom of the building, and its typical problem is that street trees, billboards, fences, and terrain undulations cause significant gaps, voids, and breaks in the bottom point cloud. Directly projecting the original podium point cloud results in incomplete, open, and misaligned poor-quality outlines, failing to form a complete and closed building boundary. To solve this problem, the system maintains the original bottom boundary of the podium segment to ensure the correct positional reference of the bottom outline. Simultaneously, it extends the original top sampling boundary of the podium segment upwards along the positive Z-axis by a second preset percentage, entering the interior of the adjacent standard segment above. The second preset percentage can be set to 30%~50%, aiming to utilize the stable, clear, and unobstructed wall point cloud at the bottom of the standard segment to effectively supplement and repair the missing data at the bottom of the podium. After the extension is complete, the system integrates all point clouds contained between the original bottom boundary of the podium and the newly extended top boundary into a single second projection sampling range. This approach preserves the true bottom position of the podium while introducing high-quality point clouds from the upper layer to fill the occlusion holes in the lower layer, making the originally broken and incomplete podium outline continuous, complete, and closed, perfectly solving the problem of outline failure caused by bottom occlusion in existing technologies.

[0072] When the system determines that the currently processed structural segment is a tower crown segment, it immediately initiates a downward fusion strategy. The tower crown segment, located at the highest point of the building, includes structures such as parapet walls, machine rooms, decorative roofs, and equipment floors. Its point cloud typically suffers from sparseness, dispersion, cross-sectional shrinkage, and unclear edges, making direct projection prone to resulting in open outlines, missing sharp corners, and blurred boundaries. To achieve a complete closed top outline, the system maintains the boundary of the very top of the tower crown segment, ensuring no loss of tower crown shape and height information. Simultaneously, it extends the original bottom sampling boundary of the tower crown segment downwards along the negative Z-axis by a third preset ratio, penetrating deep into the adjacent standard segment below. This third preset ratio can typically be set to a value greater than the first and second preset ratios. The aim is to fully incorporate the high-density, high-continuity, and geometrically stable solid wall point cloud from the top of the standard segment, achieving deep fusion with the sparse point cloud of the tower crown itself. After the extension is complete, the system uniformly includes all point clouds from the newly extended bottom boundary to the original top boundary of the tower crown within the third projection sampling range. This approach strengthens the weak data foundation of the tower crown by merging downwards, enabling the originally sparse and easily broken top profile to achieve a stable closure, while preserving the tower crown's own tapering shape characteristics. This solves the technical shortcomings of traditional methods that cannot handle sparse tops and unclosed profiles.

[0073] After the projection sampling range of all three types of segments is defined, the system outputs the first projection sampling range corresponding to the standard segment, the second projection sampling range corresponding to the podium segment, and the third projection sampling range corresponding to the tower crown segment. All ranges are mutually exclusive, complementary in data, and cover the entire height of the target building. Each segment has been specifically optimized, with no obstruction, no overlap, and no sparseness or missing parts.

[0074] This step employs a differentiated adaptive volume projection strategy that applies bidirectional inward contraction, upward borrowing, and downward fusion to the standard section, podium section, and tower crown section respectively. This precise adjustment of the projection sampling range for each segment effectively solves the technical problems in existing technologies where fixed projection ranges cannot adapt to the point cloud characteristics of different building segments, and where point cloud distortion is easily caused by missing or interference points. This achieves precise adaptation of the projection sampling range of each segment to its own structural features, ensuring the integrity and accuracy of the two-dimensional projection point cloud. It provides high-quality data support for subsequent initial contour extraction and reduces contour deviations caused by the projection process.

[0075] S105. Project the original three-dimensional point cloud in each structural segment along the Z-axis to map it into the corresponding two-dimensional projected point cloud, and use the contour grabbing algorithm to extract the initial building contour from the two-dimensional projected point cloud.

[0076] This step is performed after the system has completed the adaptive projection sampling range definition for each structural segment, and before the contour is subjected to penetrating morphological filtering. It is a key step in converting 3D point cloud data into 2D contour base lines.

[0077] Upon entering this step, the system has already obtained the first projection sampling range corresponding to the standard section, the second projection sampling range corresponding to the podium section, and the third projection sampling range corresponding to the tower crown section. Each sampling range has undergone interference removal, missing data filling, and density optimization, providing a foundation for high-quality projection. The system first processes each independent structural segment in parallel, maintaining the independence and data integrity between segments and avoiding mutual interference between point clouds of different segments. For the already filtered valid original 3D point cloud within each structural segment, the system uses a unified Z-axis as the sole projection direction and performs strict orthogonal projection mapping. This means discarding the height coordinate Z-value of each 3D point (X, Y, Z), retaining only the planar coordinates (X, Y), and aggregating all points onto the same horizontal reference plane to form a 2D projection point cloud corresponding to that structural segment. During the projection process, the system maintains the relative planar position, distribution density, and boundary features of all points, ensuring that the planar form of the 3D building is completely preserved in 2D space.

[0078] After completing the projection of all segments, the system spatially stitches and merges the generated 2D projected point clouds from each segment to form a complete 2D projected point cloud covering the entire planar area of ​​the target building. Subsequently, the system calls a preset contour-grabbing algorithm to extract the boundaries of the merged 2D projected point cloud. The algorithm transforms the discrete point set into a continuous closed polyline by performing neighborhood search, density discrimination, outer edge point tracking, boundary point sorting, and connection on the point cloud. This contour-grabbing algorithm uses high-sensitivity parameter configuration, which can capture the real details of the building's exterior walls, such as concavities, corners, and grooves, while retaining minor protrusions and burrs caused by ancillary facilities such as air conditioning units, pipes, and railings. Finally, it generates an initial building contour that can fully reflect the building's planar shape and local interference features.

[0079] This step accurately transforms the adaptively optimized 3D point cloud into a 2D point cloud through segmented projection and overall fusion. It fully preserves the data integrity advantages brought by bidirectional shrinkage, upward point borrowing, and downward fusion, avoiding contour breaks and missing parts caused by traditional single-layer slicing. A high-sensitivity contour capture algorithm is used to obtain the initial contour, which can completely preserve the real structure of the building and interference information of the ancillary facilities, providing accurate processing objects for subsequent penetrating morphological filtering.

[0080] S106. Perform penetrating morphological filtering on the initial building outline by sequentially performing strong erosion and springback expansion operations to obtain the filtered main wall outline.

[0081] The "Strong Erosion" operation refers to a morphological geometric operation that uniformly shrinks the contour boundary inward along the normal direction to eliminate minor protrusions on the contour surface. The "Spring Expansion" operation refers to a morphological geometric operation that equally expands the eroded contour outward along the normal direction to restore the true dimensions of the main wall. The "First Shrinkage Distance" refers to the inward shrinkage dimension used when performing the "Strong Erosion" operation, which is a preset configurable parameter of the system. The "Building Accessory Thickness Threshold" refers to the critical thickness value used to distinguish between the main building structure and the external facade accessories. The "First Expansion Distance" refers to the outward expansion dimension used when performing the "Spring Expansion" operation, which is equal in value and opposite in direction to the "First Shrinkage Distance".

[0082] Specifically, this step is performed after the system completes the extraction of the two-dimensional projection point cloud and obtains the initial building outline, but before the orthogonal shaping of the main wall outline. It is the core and key step in the present invention to achieve "eliminating interference from ancillary facilities and restoring the real main wall", and directly determines whether the outline can meet the stringent requirements of scenarios such as fine-grained surveying, close-up detection of the facade by UAV, and high-precision vector modeling.

[0083] When the system enters this step, the initial building outline already carries the complete planar geometric information of the target building. However, it inevitably contains a large number of protrusions, burrs, and jagged distortions caused by facade accessories. These distortions originate from air conditioning units, drainage pipes, security windows, exterior railings, small decorative components, etc., and their thickness is usually small and irregularly distributed. If the system directly enters the orthogonal shaping stage, it will cause the overall outline to deviate from the real wall surface, resulting in geometric distortion, flight path jitter, and measurement deviations in subsequent applications. Therefore, the system follows a complete process of "first determining the shrinkage distance → strong corrosion → determining the expansion distance → rebound expansion → micro-compensation calibration," performing penetrating morphological filtering processing. The entire process is parameter adaptive, logically closed-loop, and requires no manual intervention, achieving automated deep noise reduction.

[0084] The system first performs a parameter determination step, reading and parsing a preset threshold for the thickness of building ancillary facilities. This threshold is calculated based on the dimensions of conventional external components in the construction industry, covering the maximum thickness of ancillary facilities on the facades of most civil and commercial buildings, effectively distinguishing between ancillary facilities and the main building structure. After determining the threshold, the system configures a first shrinkage distance, strictly ensuring that this first shrinkage distance is greater than the thickness threshold for building ancillary facilities. This ensures that the shrinkage dimension is sufficient to completely cover and eliminate all protrusions of external facilities with a thickness less than the threshold, avoiding incomplete filtering and residual equipment burrs due to insufficient shrinkage distance. Once the parameters are determined, the system uses the first shrinkage distance as the calculation benchmark and performs a strong inward erosion operation on the entire initial building outline. During the erosion process, each boundary point on the outline is offset inward at a uniform speed and in equal amount along its own normal direction. All protrusions, burrs, and external component outlines with a thickness less than the thickness threshold of building ancillary facilities are completely "melted" away, leaving no trace. However, the main building wall, due to its large structural dimensions and thickness far exceeding the threshold of ancillary facilities, only undergoes a uniform inward offset. Its core geometry, corner relationships, and concave-convex structures are completely preserved, thus generating a core wall outline that only represents the main building structure without any minor interference. At this point, although the core wall outline is pure, its overall dimensions have an inward shrinkage deviation compared to the actual building, making it unusable for direct output.

[0085] The system then enters the rebound and restoration phase. Based on the determined first contraction distance, a first expansion distance is automatically generated. This distance is numerically equal to the first contraction distance but in the opposite direction, ensuring strict symmetry between the expansion and corrosion amplitudes and achieving precise dimensional rebound. Using the first expansion distance as a calculation parameter, the system performs an outward rebound expansion operation on the core wall contour, expanding the contour boundary outward at a uniform speed and in equal amount along the normal direction. This allows the contour dimensions to quickly return to the true physical scale of the main building before the removal of ancillary facilities, resulting in a transitional main contour. After the rebound expansion, the interference from ancillary facilities is permanently eliminated, and the shape, size, and position of the main wall are basically restored. However, due to the inherent geometric reduction effect of the strong corrosion operation, the transitional main contour may still have minor errors such as slight thinning of the wall thickness, slight inward shrinkage of corner positions, and blunting of edges. If directly used for high-precision mapping, there will still be a small deviation.

[0086] To completely eliminate the aforementioned residual errors, the system further performs refined buffer offset calibration, reading a preset micro-compensation offset. This offset is a tiny value much smaller than the first shrinkage distance, used only to correct the slight geometric loss caused by corrosion expansion. Using this micro-compensation offset as a parameter, the system performs outward buffer offset processing on the transition main contour, slightly shifting the entire contour outward along the normal direction. This precisely compensates for the wall thickness reduction, corner shrinkage, and edge blunting errors caused by the strong corrosion operation, ensuring that the final contour boundary completely coincides with and closely fits the outer edge of the real wall in the original 3D point cloud, without offset, shrinkage, or expansion. Ultimately, a pure, smooth, high-precision, interference-free main wall contour that perfectly fits the boundary of the real wall point cloud is generated, providing the highest quality basic contour data for subsequent orthogonal shaping and vector output.

[0087] This step effectively solves the technical problems in existing technologies, such as the inability to distinguish between building walls and external ancillary facilities, severe equipment interference on the contour, numerous protrusions and burrs, and deviation from the real wall surface. It achieves in-depth filtering of interference from ancillary facilities, accurate restoration of the main wall structure, and minor compensation for geometric errors, so that the generated main wall contour completely fits the real boundary of the original point cloud, significantly improving the contour accuracy and practicality, and providing a solid and reliable data foundation for subsequent orthogonal shaping and vector file output.

[0088] S107. Perform orthogonal shaping on the main wall outline to generate a regularized final building vector outline.

[0089] Among them, orthogonal shaping refers to the geometric regularization operation of correcting tilted, non-right-angled, and irregular contours into horizontal and vertical straight lines with standard 90° corners; the main axis direction of the building refers to the reference direction determined by the long side of the smallest circumscribed rectangle, which represents the overall extension trend of the building, also known as the main orientation of the building or the north-finding direction.

[0090] Specifically, this step is performed after the system completes penetrating morphological filtering and obtains a clean main wall outline, but before fitting deviation verification and vector file output. When the system enters this step, the main wall outline has eliminated all interference from ancillary facilities, and its overall shape is close to the actual building structure. However, it still has issues such as slight angle tilting, non-horizontal or non-vertical edges, minor local jitter, and some non-90° corners caused by point cloud acquisition, modeling errors, and projection calculations, making it unsuitable for direct use as a standard vector result. Therefore, the system performs refined orthogonal shaping processing according to the complete process of "determining the principal axis → rotation and alignment → line segment straightening → feature filtering → coordinate restoration," which is completed automatically throughout the entire process. This eliminates noise while preserving the actual building structure to the greatest extent, achieving a dual balance between regularity and fidelity.

[0091] The system first calculates the minimum bounding rectangle of the main wall outline. By traversing the coordinates of all vertices of the outline, it generates a rectangle that completely encloses the outline and has the smallest area. The long side of this rectangle stably reflects the overall extension trend of the building. Based on the angle of the long side of this minimum bounding rectangle, the system determines the main axis direction of the building, completing the building's "north-finding" and main orientation identification. This step is the foundation of orthogonalization and avoids distortion caused by forced straightening due to the overall tilt of the building. Subsequently, the system constructs a main axis coordinate system based on the main axis direction of the building and calculates the rotation angle required for the main wall outline to rotate from the original coordinate system to the main axis coordinate system. The main wall outline is rotated and transformed as a whole to ensure that the main extension direction of the outline is strictly parallel to the coordinate axes of the main axis coordinate system, providing a unified benchmark for subsequent orthogonal correction. In the aligned principal axis coordinate system, the system traverses all the edges of the contour segment by segment, calculates the angle between each line segment and the horizontal and vertical coordinate axes, and determines the line segments whose angles are within the preset tolerance angle range (e.g., 85° to 95°, 0° to 5°, 175° to 180°) as correctable line segments. Forced projection straightening is then performed on these line segments: line segments that are inclined to near the horizontal are forcibly corrected to be strictly horizontal, and line segments that are inclined to near the vertical are forcibly corrected to be strictly vertical, so that the contour edges conform to the typical geometric characteristics of horizontal and vertical lines of a building.

[0092] After straightening the line segments, the system traverses and identifies all geometric features of the contour, implementing differentiated processing. For minor local jitter features such as small protrusions, depressions, and short burrs remaining on the contour due to point cloud noise and calculation errors, the system compares their size with a first length threshold. When the size is less than the threshold, it is determined to be invalid noise, and noise reduction and smoothing operations are performed. Jitter is eliminated through node fusion and edge smoothing, making the contour smoother and more regular. At the same time, the system identifies structural grooves such as building recesses, courtyards, recessed balconies, and L-shaped structures that actually exist on the contour. The groove size is compared with a second length threshold. When the size is greater than the threshold, it is determined to be a valid building structure and is completely preserved. The edges and corners of the grooves are orthogonally normalized to ensure that the groove edges are horizontal and vertical and the corners are standard 90°, thus preserving the real building structure while meeting the requirements of vector regularization.

[0093] After completing all straightening, noise reduction, structural preservation, and normalization processes, the system needs to restore the contour to its original acquisition location. The system reads the angle parameters recorded during the previous rotation and alignment, performs a reverse rotation transformation on the normalized contour, and accurately restores the contour from the principal axis coordinate system to the original geodetic coordinate system of the initial input 3D point cloud, ensuring that the position coordinates do not drift, shift, or distort. Finally, the system generates a regularized final building vector contour, which features standard angles, straight edges, regular corners, noise reduction, preservation of the true structure, and accurate coordinates. It can be directly output in commonly used vector formats, meeting the needs of high-precision surveying and professional engineering applications.

[0094] This step effectively solves the technical problems of tilted outlines, non-right angles, local jitter, insufficient regularity, and inability to meet surveying standards in existing technologies. It then achieves standardized, orthogonal, and regularized output of building outlines, completely preserving the real building structure while eliminating noise. This results in a high-precision vector outline that fits the original point cloud and meets the technical requirements of building geometry and real-world 3D construction.

[0095] In some embodiments, in practical engineering applications, the final building vector contour obtained by a single orthogonal shaping may have problems such as local positional deviation, insufficient edge fit, corner offset, and large structural groove errors compared to the original 3D point cloud, due to factors such as point cloud acquisition quality, building complexity, concave corner occlusion, and initial parameter settings. If directly output, it will lead to consequences such as substandard accuracy of subsequent surveying results, deviation of UAV flight paths, and distortion of model construction. Therefore, after generating the final building vector contour, the system does not output it directly, but automatically enters the accuracy verification stage. By quantitatively calculating the fitting deviation, it achieves an adaptive closed-loop processing of "output if qualified, iterative optimization if unqualified", ensuring that each output meets the preset accuracy standard.

[0096] Specifically, the system first calculates the fitting deviation between the final building vector outline and the original 3D point cloud. Several mature quantification methods can be used: for example, projecting the final building vector outline onto a horizontal plane and calculating the average distance, maximum distance, and root mean square error from the outline to the nearest point in the point cloud; or calculating the overlap between the outline-enclosed area and the point cloud projection area; or using the average normal deviation between the outline edge and the point cloud edge as the fitting deviation. These calculation methods can be adaptively selected according to the application scenario, with the core purpose of using numerical values ​​to objectively reflect the degree of fit between the final outline and the original point cloud. After calculation, the system compares the obtained fitting deviation with a preset deviation threshold. This preset deviation threshold is configured according to industry surveying accuracy standards, project requirements, or user-defined settings, such as 0.1 meters or 0.08 meters, suitable for building outline extraction tasks of different accuracy levels.

[0097] When the fitting deviation is less than the preset deviation threshold, it means that the accuracy of the final building vector outline generated by the orthogonalization is high enough, and the outline position, shape and size are highly consistent with the original 3D point cloud, which meets the output requirements. The system directly encapsulates the final building vector outline into a standardized vector format file, which can be directly used for subsequent urban planning, GIS modeling, UAV flight path generation and real-scene 3D asset storage.

[0098] When the fitting deviation is greater than or equal to the preset deviation threshold, it indicates that the current contour deviates significantly from the original point cloud, and the accuracy does not meet the requirements. This may be due to an excessively wide preset tolerance angle range during the orthogonalization stage, leading to over-straightening; an unreasonable setting of the micro-compensation offset, causing an overall contour shift; or over-correction of local line segments and incorrect smoothing of structural grooves. In this case, the system will not directly output unqualified results, but will automatically activate the parameter adaptive adjustment and iterative optimization mechanism to correct the key parameters on which orthogonalization shaping depends: on the one hand, adjusting or expanding the preset tolerance angle range to avoid incorrect or insufficient correction of line segments; on the other hand, increasing or decreasing the micro-compensation offset to correct the overall positional deviation of the contour.

[0099] After the parameters are adjusted, the system automatically jumps back to the orthogonal shaping process of the main wall outline. Using the updated parameters, it re-executes the complete process of orthogonal correction, line straightening, minor jitter smoothing, structural groove standardization, and coordinate inverse restoration to generate a new round of optimized final building vector outline. It then recalculates the fitting deviation and repeats the above "calculation-judgment-adjustment-re-execution" process until the fitting deviation meets the preset deviation threshold requirement. Finally, it outputs a qualified vector format file, thereby achieving fully automated precision control and iterative optimization.

[0100] This step effectively solves the technical problems that may exist in single contour generation, such as insufficient accuracy, excessive deviation, and inability to self-correct. It avoids the distortion of the final contour caused by unreasonable initial parameter settings, local anomalies in the point cloud, and complex building structures. As a result, it achieves controllable accuracy, automatic correction, and iterative optimization throughout the entire building contour extraction process, ensuring that the output vector contour always maintains a high degree of fit with the original 3D point cloud. This improves the reliability of the results, the accuracy of surveying and mapping, and the practicality of engineering applications. It enables this method to stably adapt to various complex buildings, different point cloud qualities, and high-precision application needs in multiple scenarios.

[0101] In the above embodiment, the original 3D point cloud is preprocessed to remove noise and interference data. Based on the Z-axis point cloud density features, the building is adaptively divided into podium, standard, and crown segments for vertical structural analysis. Simultaneously, a differentiated adaptive volume projection strategy is implemented for each segment, employing bidirectional shrinkage, upward point borrowing, and downward fusion to determine the optimal projection sampling range. Then, the point clouds of each segment are projected along the Z-axis, and an initial building contour is extracted using a contour-grabbing algorithm. Subsequently, a penetrating morphological filter using strong erosion and springback expansion is used to remove interference from ancillary facilities, resulting in a clean main wall contour. Finally, the main wall undergoes orthogonal shaping, including principal axis alignment, line straightening, feature filtering, and coordinate restoration. The complete technical means of processing effectively solves the technical problems in existing technologies, such as the inability of fixed-height slices to perceive changes in the vertical structure of buildings, outline breaks caused by occlusion at the bottom of the building, outline protrusion and jitter caused by interference from external facilities, interlayer adhesion at concave corners, insufficient outline regularity, and difficulty in meeting the requirements of fine-grained surveying and close-range flight of UAVs. It then realizes the adaptive and accurate projection of building point clouds from three-dimensional to two-dimensional, automated depth filtering of external interference facilities, and orthogonalization and regularization shaping of building outlines. Finally, it generates building vector outlines that fit the real walls, are geometrically regular, have reliable accuracy, and comply with surveying and mapping standards, significantly improving the automation level and engineering practicality of building outline extraction.

[0102] To further illustrate the specific process and effects of vertical structural feature analysis and structural segmentation of the original 3D point cloud, please refer to [link / reference]. Figure 3 , Figure 3 This is a schematic diagram of the vertical structure analysis of a building in an embodiment of this application.

[0103] exist Figure 3 In the figure, the horizontal axis represents the approximate floor area (square meters) of each height level of the target building, and the vertical axis represents the building's height along the Z-axis. The blue solid line in the figure represents the actual curve of the cross-sectional area of ​​each height level (i.e., the bounding box area of ​​the point cloud within the height level projected onto the XY horizontal plane) changing with height; the red dashed line represents the shape changes extracted after system analysis and calculation (i.e., boundary abrupt change points and piecewise fitting lines).

[0104] This system does not adopt a crude, globally unified processing approach, but rather generates a refined version. Figure 3 The Z-axis area / density distribution characteristics are shown. By calculating the area change rate between adjacent height layers along the blue solid line, and locating abrupt change points when the absolute value of the change rate exceeds a preset abrupt change threshold, the system can clearly identify the vertical structural boundaries of the building. Specifically:

[0105] In the bottom section of the building (such as Figure 3 The height range is approximately -36 to -24 (the blue solid line indicates that the approximate floor area of ​​this region is the largest in each segment and is accompanied by local fluctuations). After the system identifies the first abrupt change point above it, it automatically divides the height range from the bottom to the first abrupt change point into the "podium segment".

[0106] Between the first mutation point and the second mutation point (e.g.) Figure 3 The height range is approximately -24 to +17. The blue solid line shows a long span of near verticality, indicating that the area change rate of each floor slice within this range is extremely small (less than the preset stability threshold), and it has a high degree of geometric stability. The system aggregates and divides this height range into "standard segments".

[0107] Crossing the second mutation point to the top of the building (e.g.) Figure 3 (For the area above +17 meters in height), the blue solid line indicates that the cross-sectional area has decreased sharply relative to the standard section (exceeding the preset reduction ratio). The system classifies this top area as the "tower crown section (or roof section)".

[0108] pass Figure 3 The vertical structure analysis shown in this application demonstrates that this application achieves automated and precise perception of the three-segment vertical geometric attributes of a complex target building: the podium, standard sections, and the crown. This step completely breaks through the blind spots caused by fixed-height slicing in existing technologies, laying a crucial data partitioning foundation for subsequent adaptive volume projection strategies that match different segments. This ensures that the final generated building outline will not be broken due to bottom occlusion or distorted due to top sparseness.

[0109] To further illustrate the specific processing steps and denoising effects of initial contour extraction, penetrating morphological filtering, and orthogonal shaping, please refer to [link to relevant documentation]. Figure 4 , Figure 4 This is a schematic diagram showing the comparison of the effects of contour filtering and shaping on the standard floor and roof floor of the building in the embodiments of this application.

[0110] exist Figure 4The figure visually demonstrates the physical effectiveness of the combined effects of a series of steps employed in this method, including high-precision contour extraction, penetrating morphological filtering, and orthogonal shaping. The upper part of the figure shows the processing result of the standard layer (standard segment), and the lower part shows the processing result of the roof layer (tower crown segment). The gray point set at the bottom of the figure represents the two-dimensional projected point cloud generated after adaptive volume projection.

[0111] Figure 4 The light green dashed line represents the "original grab" outline (i.e., the initial building outline) extracted using high-precision grab parameters. In order to fully capture the subtle undulations and grooves of the wall surface, this initial boundary is drawn in extremely small gaps, but this also extracts a large number of protruding parts of external accessories such as air conditioning units and drainage pipes (i.e., the burrs and noise protruding outwards from the green line).

[0112] The core parameters marked above the diagram, "Equipment Filtration = 0.25m, Compensation = +0.06m," clearly correspond to the penetration filtering and position compensation. The system first performs a "powerful corrosion" operation using a contraction distance (0.25m) greater than the equipment thickness, allowing thinner external structures to be directly "melted" and eliminated. Subsequently, the system performs a "rebound expansion" operation using an expansion distance of the same magnitude (0.25m), driving the main wall surface to spring back to its original position. Finally, a small compensation offset (+0.06m) is applied to slightly expand outwards, compensating for the wall thickness and corner thinning errors caused by the corrosion operation.

[0113] The red solid line with blue feature vertices in the figure is the "corrected contour" (i.e. the final regularized building vector contour) after the above filtering process, and after the direction north positioning based on the yellow dotted line (minimum bounding rectangle), forced projection straightening of line segments, and smoothing of minor jitter features.

[0114] Comparing the solid and dashed outlines, it's clear that the solid red line directly traverses and perfectly separates any obstructions from the external wall structure (i.e., the equipment layer protruding from the green line is discarded). The resulting red line is not only smooth and expansive but also exhibits a standard orthogonal right angle, perfectly fitting and retrieving the true geometric topological position of the target building's main wall. Figure 4 This method effectively solves the technical pain point of previous slice contours containing a large number of external equipment protrusions, which caused flight path jitter or mapping deviations.

[0115] To visually compare the drawbacks of traditional fixed-height slicing with the advantages of the core strategy of this application, please refer to [link / reference]. Figure 5 , Figure 5 This is a schematic diagram comparing the effects of extracting contours using traditional single-layer slicing and adaptive volume projection strategies in the embodiments of this application.

[0116] exist Figure 5 The paper details the differences in local and overall effects before and after optimization. The left column shows the original slice point cloud obtained using the existing traditional single-layer slicing method (i.e., before optimization); the right column shows the two-dimensional projection and extracted envelope results generated by applying the adaptive volume projection strategy of this application in the same corresponding building segment (i.e., after optimization).

[0117] First, regarding the issue of broken contours (as shown in the first row of the comparison panel in the figure): In the original slice on the left, due to the severe occlusion of buildings (such as the bottom podium section) by surrounding roadside trees and other vegetation, a large area of ​​point cloud data is missing at a specific height level. Direct horizontal slicing only yields fragmented and incomplete line segment combinations. In the corresponding optimized result on the right, this system precisely applies an adaptive volume projection strategy of "borrowing points upwards," extending the projection sampling range upwards to the unobstructed wall area of ​​the next higher level. This utilizes the clear wall data from higher levels to compensate for the missing data at the lower level. It can be seen that the red line on the right successfully crosses the data void area, achieving perfect repair of the broken contour.

[0118] Secondly, regarding the issues of point cloud noise and heavy boundaries (as shown in the comparison panels in the second and third rows of the figure): In the original slice on the left, due to geometric topological errors often occurring at building corners or floor slab junctions in the mesh generated by UAV oblique photography, the original slice boundary is covered with a large number of heavy noise points, messy burrs, and redundant point sets adhered between layers. In the corresponding optimized result on the right, this system strictly implemented a "bidirectional shrinkage" strategy for the standard segment (removing redundant point clouds disturbed at the upper and lower edges, and purifying and retaining the point cloud in the middle segment), and implemented a "downward fusion" strategy for the tower crown segment. The resulting red line outline completely eliminated the original rough and blurry boundaries.

[0119] In summary, compared to the original slice on the left, which contained breaks, voids, and messy noise, the red line result generated on the right using the adaptive "volume-for-surface" projection strategy exhibits a high degree of continuity, regularity, and geometric integrity. This step fully utilizes the complementary advantages of inter-layer data along the vertical direction of the building's Z-axis, removing the most challenging issues of basic voids and adhesions for subsequent precise morphological filtering.

[0120] The point cloud building outline regularization generation system in this application embodiment is described below from the perspective of hardware processing. Please refer to [link to relevant documentation]. Figure 6 This is a schematic diagram of a physical device structure for a point cloud building outline regularization generation system in this application embodiment.

[0121] It should be noted that, Figure 6 The structure of the point cloud building outline regularization generation system shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0122] like Figure 6 As shown, the point cloud building outline regularization generation system includes a central processing unit (CPU) 601, which can perform various appropriate actions and processes based on programs stored in read-only memory (ROM) 602 or programs loaded from storage section 608 into random access memory (RAM) 603, such as performing the methods described in the above embodiments. The RAM 603 also stores various programs and data required for system operation. The CPU 601, ROM 602, and RAM 603 are interconnected via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0123] The following components are connected to I / O interface 605: input section 606 including audio input devices, push-button switches, etc.; output section 607 including a liquid crystal display (LCD) and audio output devices, indicator lights, etc.; storage section 608 including a hard disk, etc.; and communication section 609 including a network interface card such as a LAN (Local Area Network) card, modem, etc. Communication section 609 performs communication processing via a network such as the Internet. Drive 610 is also connected to I / O interface 605 as needed. Removable media 611, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 610 as needed so that computer programs read from them can be installed into storage section 608 as needed.

[0124] Specifically, according to embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program including a computer program for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 609, and / or installed from removable medium 611. When the computer program is executed by central processing unit (CPU) 601, it performs the various functions defined in this application.

[0125] It should be noted that specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0126] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. Each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those shown in the drawings.

[0127] Specifically, the point cloud building outline regularization generation system of this embodiment includes a processor and a memory. The memory stores a computer program. When the computer program is executed by the processor, it implements the point cloud building outline regularization generation method based on adaptive volume projection provided in the above embodiment.

[0128] In another aspect, this application also provides a computer-readable storage medium, which may be included in the point cloud building contour regularization generation system described in the above embodiments; or it may exist independently and not assembled into the point cloud building contour regularization generation system. The storage medium carries one or more computer programs, which, when executed by a processor of the point cloud building contour regularization generation system, cause the point cloud building contour regularization generation system to implement the point cloud building contour regularization generation method based on adaptive volume projection provided in the above embodiments.

[0129] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0130] As used in the above embodiments, depending on the context, the term "when..." can be interpreted as "if...", "after...", "in response to determining...", or "in response to detecting...". Similarly, depending on the context, the phrase "when determining..." or "if (the stated condition or event) is interpreted as meaning "if determining...", "in response to determining...", "when (the stated condition or event) is detected", or "in response to detecting (the stated condition or event)".

[0131] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A method for generating point cloud building contour regularization based on adaptive volume projection, characterized in that, The method, applied to a point cloud building contour regularization generation system, includes: Obtain the original 3D point cloud of the target building and preprocess the original 3D point cloud; Vertical structural feature analysis is performed on the preprocessed original 3D point cloud to obtain the point cloud density features along the Z-axis. Based on the point cloud density characteristics, the target building is divided into multiple structural segments in the vertical direction, and the segment type matching each structural segment is determined. Based on the segmentation type, a differentiated adaptive volume projection strategy is applied to determine the projection sampling range of each structural segment; The original three-dimensional point cloud in each structural segment is projected and mapped along the Z-axis to the corresponding two-dimensional projected point cloud, and the initial building outline is extracted from the two-dimensional projected point cloud using a contour grabbing algorithm. The initial building outline is subjected to a penetrating morphological filtering process consisting of a strong erosion operation and a springback expansion operation to obtain the filtered main wall outline. The main wall outline is orthogonally shaped to generate a regularized final building vector outline.

2. The method of claim 1, wherein, Based on the point cloud density features, the step of dividing the target building into multiple structural segments in the vertical direction and determining the matching segment type for each structural segment specifically includes: Based on the point cloud density characteristics, the number of point clouds in each height layer is counted along the Z-axis with a preset step size, or the projected area of ​​the point clouds in each height layer projected onto the horizontal plane, and a Z-axis density histogram is generated. Based on the density abrupt change points in the Z-axis density histogram, the target building is divided into structural segments of the segment type, which at least include a podium segment, a standard segment, and a tower crown segment.

3. The method according to claim 2, characterized in that, The step of dividing the target building into structural segments, including at least a podium segment, a standard segment, and a tower crown segment, based on density abrupt change points in the Z-axis density histogram specifically includes: Calculate the rate of change of point cloud between adjacent height layers in the Z-axis density histogram, where the rate of change of point cloud includes the rate of change of point cloud quantity or the rate of change of projected area. When the absolute value of the point cloud change rate exceeds a preset abrupt change threshold, the matching position is determined as a density abrupt change point; Identify a first mutation point and a second mutation point, wherein the first Z-axis height of the first mutation point is lower than the second Z-axis height of the second mutation point; The height range from the bottom of the building to the first abrupt change point is divided into a podium segment, characterized by the corresponding projected area or point cloud number being the largest among all segments; The height range between the first mutation point and the second mutation point is divided into a standard segment, and the standard segment is characterized by the point cloud number change rate or the projected area change rate being less than a preset stable threshold. The height range from the second mutation point to the top of the target building is divided into a tower crown segment. The tower crown segment is characterized by a reduction in the corresponding projected area or number of point clouds compared to the standard segment by more than a preset reduction ratio. Within each height range, calculate the contour similarity of multiple height layer slices; When the contour similarity between consecutive preset number height layers is higher than a preset similarity threshold, it is determined that the target interval corresponding to the consecutive preset number height layers has geometric stability, and the target interval is aggregated into a structural segment.

4. The method according to claim 1, characterized in that, The step of determining the projection sampling range of each structural segment by applying a differentiated adaptive volume projection strategy according to the segmentation type specifically includes: If the segment type is a standard segment, a bidirectional indentation strategy is applied to the standard segment, the bidirectional indentation strategy including: Remove the point cloud data corresponding to the first preset ratio at the top and bottom of the standard segment along the Z-axis direction, and extract and retain the point cloud in the middle area as the first projection sampling range; If the segment type is a podium segment, an upward borrowing strategy is applied to the podium segment. The upward borrowing strategy includes: The top sampling boundary of the podium section in the Z-axis direction is extended upward by a second preset proportion into the adjacent upper structure segment, and the point cloud contained between the bottom boundary of the podium section and the extended top boundary is defined as the second projection sampling range. If the segment type is a tower crown segment, a downward fusion strategy is executed on the tower crown segment, and the downward fusion strategy includes: The bottom sampling boundary of the tower crown segment in the Z-axis direction is extended downward by a third preset proportion into the adjacent lower structure segment, and the point cloud contained between the extended bottom boundary and the top boundary of the tower crown segment is defined as the third projection sampling range.

5. The method according to claim 1, characterized in that, The steps of performing penetrating morphological filtering on the initial building outline by sequentially performing strong erosion and springback expansion operations to obtain the filtered main wall outline specifically include: A first contraction distance is determined, wherein the first contraction distance is greater than a preset threshold for the thickness of building ancillary facilities; The initial building outline is subjected to a strong inward erosion operation using the first shrinkage distance to shrink inward and eliminate protruding ancillary facility outlines with a thickness less than the building ancillary facility thickness threshold, and generate a core wall outline that is offset inward as a whole. Determine a first expansion distance that matches the first contraction distance; Using the first expansion distance, an outward rebound expansion operation is performed on the core wall contour, driving the core wall contour to rebound outward to the main building entity scale position before the removal of auxiliary facilities, thus obtaining the transition main contour; The transition main contour is buffered outward by using a preset micro-compensation offset to compensate for the reduction error in wall thickness and corners caused by the strong corrosion operation, thereby generating the main wall contour that fits the boundary of the real wall point cloud.

6. The method according to claim 1, characterized in that, The step of orthogonally shaping the main wall outline to generate a regularized final building vector outline specifically includes: Calculate the minimum bounding rectangle of the main wall outline, and determine the main axis direction of the building outline based on the orientation of the minimum bounding rectangle; Rotate and align the entire outline of the main wall to the principal axis coordinate system defined by the main axis direction of the building; In the principal axis coordinate system, line segments whose direction and the coordinate axis are within the preset tolerance angle range are identified, and the identified line segments are forcibly projected and straightened into horizontal or vertical orthogonal line segments. Traverse the geometric features of the straightened main wall outline. When the size of a local geometric feature is less than a preset first length threshold, perform noise reduction and smoothing operations on the local micro-jitter features. When the size of the structural groove formed by local geometric features is greater than a preset second length threshold, the structural groove is retained and orthogonal normalization is performed. The contour, after feature filtering and orthogonal normalization, is subjected to an inverse rotation transformation to restore it to the original coordinate system of the initial input point cloud, thereby generating the regularized final building vector contour.

7. The method according to claim 1, characterized in that, After the step of orthogonalizing and shaping the main wall outline to generate a regularized final building vector outline, the method further includes: Calculate the fitting deviation between the final building vector profile and the original 3D point cloud; When the fitting deviation is less than the preset deviation threshold, the final building vector profile is output as a vector format file; When the fitting deviation is greater than or equal to the preset deviation threshold, adjust the preset tolerance angle range or slightly compensate the offset, and re-execute the orthogonalization shaping process.

8. A point cloud building outline regularization generation system, characterized in that, The point cloud building outline regularization generation system includes: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code includes computer instructions, and the one or more processors call the computer instructions to cause the point cloud building outline regularization generation system to perform the method as described in any one of claims 1-7.

9. A computer-readable storage medium comprising instructions, characterized in that, When the instruction is run on the point cloud building outline regularization generation system, the point cloud building outline regularization generation system performs the method as described in any one of claims 1-7.

10. A computer program product, comprising a computer program, characterized in that, When the computer program runs on the point cloud building outline regularization generation system, the point cloud building outline regularization generation system performs the method as described in any one of claims 1-7.

Images