A building facade scanning method and system

Abstract

This invention relates to the field of 3D modeling technology, specifically to a method and system for scanning building facades, comprising the following steps: acquiring facade point clouds and performing normal clustering, dividing geometric regions and calculating normals, curvature, and distances, analyzing topological propagation chains to constrain curvature and normals, adjusting control point displacements and smoothing coefficients to update coordinates, calculating boundary overlap ratios and curvature consistency to generate morphological fusion instructions, adjusting control point coordinates and normal directions, and outputting facade scanning results. In this invention, precise layering is achieved through point cloud spatial normal clustering, geometric constraints are established by combining normal, curvature, and distance calculations, topological propagation controls normal offsets and curvature gradients, enhancing morphological correction continuity, smoothly updating control points, balancing accuracy and smoothness, optimizing fusion by determining boundary overlap ratios and normal / curvature consistency, reducing discontinuities and misalignments, and comprehensively improving model accuracy and stability.

Description

Technical Field

[0001] This invention relates to the field of 3D modeling technology, and in particular to a method and system for scanning building facades. Background Technology

[0002] The field of 3D modeling technology encompasses the acquisition of data, geometric reconstruction, and spatial representation of objects or scenes to obtain 3D digital models that can be used for display, analysis, and design. Core components include the identification and reconstruction of target surface morphology, spatial structure, and geometric features, often achieved through multi-source data acquisition, spatial coordinate calculation, and model mesh generation. 3D modeling is applied in fields such as architecture, surveying, cultural relic preservation, virtual reality, and intelligent manufacturing. Its technological system covers multiple stages, including image acquisition, point cloud data processing, geometric registration, and model reconstruction, forming a complete technological chain from data acquisition to structural modeling.

[0003] Among them, the building facade scanning method and system refers to the process of acquiring facade surface data through laser ranging or structured light scanning, and converting the obtained spatial coordinate points into a three-dimensional dataset for shape reconstruction. This mainly involves setting the facade scanning path, extracting depth information from the scanning data, and reconstructing the shape geometry. Typically, based on the spatial positioning results of the scanning equipment, a depth reconstruction algorithm is used to model the facade shape, completing the data construction process of the three-dimensional model of the building facade.

[0004] Existing point cloud processing technologies often rely on aggregation methods based on single geometric parameters, which can easily lead to boundary confusion between regions with different curvatures, resulting in geometric feature distortion during model reconstruction. Current methods are insufficient in coordinating the control of normals and curvature, lacking in-depth analysis of spatial topological relationships, and prone to normal discontinuities and curvature abrupt changes, affecting surface smoothness and visual consistency. Point cloud correction often employs fixed smoothing filters, which have weak ability to preserve subtle structural features, easily leading to over-smoothing and distortion of facade features. The region fusion stage typically relies solely on coordinate proximity, lacking comprehensive consideration of normal and curvature consistency, resulting in cracks or misalignments at fusion boundaries. Because the data propagation path lacks global constraints, the overall model shape is difficult to coordinate and unify, and the reconstruction results exhibit local deviations, uneven boundaries, and geometric distortions on complex building surfaces, affecting the usability and engineering accuracy of the digital model of the building facade. Summary of the Invention

[0005] To address the technical problems existing in the prior art, this invention provides a method for scanning the exterior facade of a building, comprising the following steps: S1: Acquire point cloud data of building facade and perform spatial normal distribution clustering. Divide the point cloud into geometric regions and calculate the point normal vector, radius of curvature and spatial distance between points in multiple regions to generate facade geometric data. S2: Based on the spatial topological order analysis propagation chain of the facade geometry data, calculate the normal offset difference between adjacent points. If the offset difference exceeds the curvature gradient threshold, analyze the curvature change trend and constrain the curvature and normal of the control points to generate constrained propagation data. S3: Call the constraint propagation data to adjust the displacement of the control point normal direction, calculate the displacement difference between adjacent control points and analyze the smoothing coefficient, update the control point coordinates with weighted smoothing, and generate the exterior facade correction surface set. S4: Based on the modified surface set of the facade, calculate the boundary overlap ratio, normal angle offset and curvature consistency parameter of adjacent geometric regions. When any two parameters meet the consistency threshold, generate a morphological fusion instruction set. S5: Call the aforementioned morphological fusion instruction set, calculate the spatial distance difference of control points, the deviation of the normal direction and the tangential continuity offset sequence and weight them, perform proportional adjustment on the coordinates of the control points and the normal direction, and generate the facade scanning result.

[0006] As a further embodiment of the present invention, the facade geometric data includes geometric region division data, point normal vector set, curvature radius parameter set, and spatial distance between points; the constraint propagation data includes normal offset difference sequence and curvature change constraint parameters; the facade correction surface set includes smoothing coefficient, adjacent control point displacement difference parameter, and corrected normal vector set; the morphology fusion instruction set includes boundary overlap ratio parameter, normal angle offset parameter, and curvature consistency parameter; and the facade scanning result includes control point spatial distance difference set, normal direction deviation sequence, and tangential continuity offset sequence.

[0007] As a further aspect of the present invention, the specific steps of S1 are as follows: S101: Obtain the point cloud dataset of the building facade, organize the three-dimensional coordinate parameters of multiple points in the point cloud in a structured manner, calculate the Euclidean distance between multiple points based on the coordinate neighborhood relationship, and then perform spatial clustering on the distance value sequence to generate the facade spatial clustering partition set. S102: Based on the aforementioned facade space clustering partition set, extract the local neighborhood coordinate set of point clouds within multiple partitions and calculate the normal vector distribution value. Compare the angle difference of the normal vector with the direction angle of the cluster center. Based on the normal deviation angle threshold, redivide the boundaries of multiple partitions and generate partition normal feature sets. S103: Based on the partitioned normal feature set, calculate the local curvature radius of the points in the multiple partitions, perform a weighted average operation on the spatial distance between the points, and couple and map the curvature radius with the spatial distance result to generate the facade geometric data. The normal deviation angle threshold is determined by statistically analyzing the angular difference characteristics of the distribution of normal vectors in multiple regions of the facade point cloud.

[0008] As a further aspect of the present invention, the specific steps of S2 are as follows: S201: Based on the spatial topology analysis propagation chain of the facade geometry data, call the normal vectors and spatial coordinates of multiple points and group and pair the normal vectors of adjacent points, calculate the offset difference of each group of normal components on the three coordinate axes, iteratively calculate the cosine angle difference, and generate a normal offset angle sequence set. S202: Calculate the normal offset difference based on the normal offset angle sequence set. If the normal offset difference exceeds the curvature gradient threshold, analyze the rate of change of curvature radius of the point and adjacent points. Analyze the positive and negative trends of the rate of change of curvature through differential operation to generate a curvature change trend table. S203: Call the curvature change trend table, extract the curvature radius and normal vector of the corresponding control points for positive and negative values ​​of the rate of change, perform weighted constraint adjustment on the two, update the constraint control state of all points, and generate constraint propagation data.

[0009] As a further aspect of the present invention, the curvature gradient threshold is obtained by analyzing the variation distribution characteristics of the curvature radius of multiple regions in the point cloud of the facade, that is, after extracting the curvature radius of multiple region points, calculating the difference value sequence of the curvature radius of adjacent points, and performing frequency statistics on the difference sequence.

[0010] As a further aspect of the present invention, the specific steps of S3 are as follows: S301: Obtain the set of normal direction vectors of control points in the constraint propagation data, perform differential calculation on the normal direction vectors of adjacent control points to calculate the direction offset angle value, compare the direction offset angle value with the normal deviation angle threshold, remove abnormal control point vectors, and generate a normal displacement differential dataset. S302: Based on the normal displacement difference dataset, call the displacement vector difference between adjacent control points and perform amplitude analysis, calculate the displacement change rate sequence to extract the continuous interval of change rate, calculate the mean change rate within the interval as a smoothing coefficient, and generate a set of control point smoothing coefficients. S303: Based on the set of control point smoothing coefficients, perform weighted smoothing on the control point coordinates in the constraint propagation data, use the smoothing coefficients as weights in the control point coordinate update calculation, calculate the difference in the updated control point coordinates and perform coordinate aggregation to generate a set of exterior facade correction surfaces.

[0011] As a further aspect of the present invention, the normal deviation angle threshold is obtained by analyzing the angular distribution characteristics of the normal vectors in multiple regions of the facade point cloud, extracting the normal vectors of multiple regions, then calculating the angle between the normal vectors of adjacent points, and analyzing the angle distribution curve of the angle value to determine the stable range of the normal direction.

[0012] As a further aspect of the present invention, the specific steps of S4 are as follows: S401: Obtain the boundary coordinate data of adjacent geometric regions in the set of modified surfaces of the facade, calculate the overlapping area of ​​adjacent boundaries, and obtain the ratio of the overlapping area to the area enclosed by the two regions, record the overlap ratio sequence and filter out abnormal ratio items to generate a boundary overlap ratio sequence set. S402: Call the adjacent regions in the boundary overlap ratio sequence set, calculate the angle between the normal vectors of the adjacent regions based on the multi-region surface normal vector data, compare the angle value with the normal consistency threshold, filter the region pairs whose angle offset meets the normal consistency threshold condition, and generate a normal angle offset sequence set; S403: Based on the region pairs in the normal angle offset sequence set, extract the curvature parameter distribution of the corresponding region, calculate the curvature change rate of adjacent regions, and mark the fusion region when any two parameters among the boundary overlap ratio, normal angle offset and curvature consistency meet the consistency threshold, and generate a morphological fusion instruction set. The consistency threshold is set by analyzing the distribution pattern of characteristic parameters in multiple regions of the facade correction surface. The normal consistency threshold is set by analyzing the difference characteristics of the normal vectors of adjacent geometric regions in the facade correction surface set.

[0013] As a further aspect of the present invention, the specific steps of S5 are as follows: S501: Based on the control point coordinate data of adjacent regions in the morphological fusion instruction set, the relative positions of multiple control points in three-dimensional space are detected, the magnitude of the spatial coordinate difference vector is calculated, and abnormal deviations in the magnitude sequence are removed and smoothed to generate a spatial distance difference sequence. S502: Call the spatial distance difference sequence, retrieve the normal vector parameters of the corresponding control points, calculate based on the difference in normal angle between adjacent control points, compare with the direction deviation reference value, filter the normal changes that meet the direction deviation reference value, and generate a normal direction deviation sequence set; S503: Based on the spatial distance difference sequence set and the normal direction deviation sequence set, extract the tangential direction vector of adjacent control points, calculate the continuity offset between adjacent tangential directions, perform weighted superposition of the three types of sequences and perform proportional calculation, adjust the coordinates of control points and normal directions, and generate the facade scanning result. The directional deviation reference value is set by statistically analyzing the directional difference characteristics of the normal vectors of adjacent control points after morphological fusion.

[0014] A building facade scanning system includes: The data clustering module acquires point cloud data of the building facade and performs spatial normal distribution clustering. It divides the point cloud into geometric regions and calculates the point normal vectors, radii of curvature, and spatial distances between points in multiple regions, generating facade geometric data and transferring it to the topology analysis module. The topology analysis module analyzes the propagation chain based on the spatial topology order of the facade geometric data, calculates the normal offset difference between adjacent points, and if the offset difference exceeds the curvature gradient threshold, analyzes the curvature change trend and constrains the curvature and normal of the control points, generates constrained propagation data and transmits it to the normal adjustment module. The normal adjustment module calls the constraint propagation data to adjust the displacement of the control point normal direction, calculates the displacement difference between adjacent control points and analyzes the smoothing coefficient, updates the control point coordinates with weighted smoothing, generates the exterior facade correction surface set and passes it to the fusion analysis module. The fusion analysis module calculates the boundary overlap ratio, normal angle offset and curvature consistency parameters of adjacent geometric regions based on the exterior modified surface set. When any two parameters meet the consistency threshold, a morphological fusion instruction set is generated and transmitted to the morphological fusion module. The morphology fusion module calls the morphology fusion instruction set to calculate the spatial distance difference of control points, the deviation of the normal direction and the tangential continuity offset sequence and weights them together. It then performs proportional adjustments on the coordinates of the control points and the normal direction to generate the facade scanning results.

[0015] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, by clustering the spatial normal distribution of point clouds, scattered point clouds are divided into regions with consistent geometric features, enabling precise layered identification of building surface structures and avoiding the accumulation of surface errors caused by blurred point cloud boundaries in traditional modeling. Through comprehensive calculation of normal vectors, radii of curvature, and distances between points, spatial geometric constraints between points are established, providing a continuous foundation for subsequent morphological correction. In spatial topological sequence propagation, by detecting the difference in normal offset and curvature gradient threshold between adjacent points, a constraint propagation chain of curvature change trends is formed, strengthening the geometric continuity control of complex facade morphology. The weighted smoothing update mechanism of control points maintains surface smoothness while avoiding excessive smoothing of geometric features, achieving a balance between accuracy and smoothness. The joint determination of boundary overlap ratio, normal angle, and curvature consistency between adjacent regions allows the morphological fusion process to achieve a natural transition based on multi-dimensional geometric consistency conditions, reducing boundary discontinuities and geometric misalignments. By weighted superposition of the spatial distance and normal direction of control points, facade reconstruction that balances morphological continuity and structural coordination is achieved. The overall logic forms a coherent data flow structure in the stages of point cloud partitioning, normal constraint propagation, curvature smoothing adjustment and morphological fusion, which significantly improves the geometric accuracy, surface continuity and morphological consistency of the facade model, and can effectively enhance the realism and structural stability of the reconstructed building surface. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of the steps of the present invention; Figure 2 This is a detailed schematic diagram of S1 of the present invention; Figure 3 This is a detailed schematic diagram of S2 of the present invention; Figure 4 This is a detailed schematic diagram of S3 of the present invention; Figure 5 This is a detailed schematic diagram of S4 of the present invention; Figure 6 This is a detailed schematic diagram of S5 of the present invention; Figure 7 This is a system module diagram of the present invention. Detailed Implementation

[0018] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0019] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0020] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0021] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0022] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0023] Please see Figure 1 This invention provides a method for scanning the exterior facade of a building, comprising the following steps: S1: Acquire point cloud data of building facade and perform spatial normal distribution clustering. Divide the point cloud into geometric regions and calculate the point normal vector, radius of curvature and spatial distance between points in multiple regions to generate facade geometric data. S2: Based on the spatial topological order analysis propagation chain of the facade geometry data, calculate the normal offset difference between adjacent points. If the offset difference exceeds the curvature gradient threshold, analyze the curvature change trend and constrain the curvature and normal of the control points to generate constrained propagation data. S3: Call the constraint propagation data to adjust the displacement of the control points in the normal direction, calculate the displacement difference between adjacent control points and analyze the smoothing coefficient, update the control point coordinates with weighted smoothing, and generate the set of exterior facade correction surfaces. S4: Based on the exterior facade modified surface set, calculate the boundary overlap ratio, normal angle offset and curvature consistency parameter of adjacent geometric regions. When any two parameters meet the consistency threshold, generate the morphological fusion instruction set. S5: Call the morphological fusion instruction set, calculate the spatial distance difference of control points, the deviation of the normal direction and the tangential continuity offset sequence and weight them, perform proportional adjustment on the coordinates of the control points and the normal direction, and generate the facade scanning result.

[0024] The facade geometric data includes geometric region division data, point normal vector set, curvature radius parameter set, and spatial distance between points. The constraint propagation data includes normal offset difference sequence and curvature change constraint parameters. The facade correction surface set includes smoothing coefficient, adjacent control point displacement difference parameter, and corrected normal vector set. The morphology fusion instruction set includes boundary overlap ratio parameter, normal angle offset parameter, and curvature consistency parameter. The facade scanning results include control point spatial distance difference set, normal direction deviation sequence, and tangential continuity offset sequence.

[0025] Please see Figure 2 The specific steps of S1 are as follows: S101: Obtain the point cloud dataset of the building facade, organize the three-dimensional coordinate parameters of multiple points in the point cloud in a structured manner, calculate the Euclidean distance between multiple points based on the coordinate neighborhood relationship, and then perform spatial clustering on the distance value sequence to generate the facade spatial clustering partition set. The 3D coordinate parameters of multiple points in the acquired building facade point cloud dataset are structured and organized to form an initial dataset containing the 3D coordinates (X, Y, Z) of each point. For example, a building including walls and windows is scanned using a terrestrial 3D laser scanner to obtain its facade point cloud. The average density of the point cloud is 400 points per square meter, i.e., the point spacing is approximately 0.05 meters. A portion of the points are selected for illustration, and their coordinate parameters are shown in Table 1.

[0026] Table 1. Example of point cloud coordinate parameters for building facades.

[0027] As shown in Table 1, points P1, P2, and P3 are located on the building wall, points P4 and P5 are located in another area of ​​the same wall, and points P6, P7, and P8 are located on the recessed structure of the window. The Euclidean distance between multiple points is calculated based on their coordinate neighborhood relationships. This process does not use a specific algorithm but directly calculates the linear distance between any two points in the point set. Taking points P1 (5.21, 10.33, 2.50) and P2 (5.26, 10.34, 2.51) as an example, the Euclidean distance calculation process is as follows: First, the squares of the differences between the X, Y, and Z coordinates are calculated, which are (5.26 - 5.21). 2 =0.0025, (10.34-10.33) 2 =0.0001, and (2.51-2.50) 2 =0.0001; then add the three squared differences together to get 0.0027; take the square root (0.0027). 2 =0.052 meters. Similarly, calculate the distance between P1 and P4 (6.50, 10.35, 2.52), the sum of the squares of their coordinate differences is (6.50 - 5.21).2 +(10.35-10.33) 2 +(2.52-2.50) 2 =1.6641+0.0004+0.0004=1.6649, the square root of which gives a distance of approximately 1.29 meters. Spatial clustering is performed on the distance value sequence. Specifically, a neighborhood distance threshold is set, and each point is iterated. If the number of points within a point's neighborhood (i.e., the range of points less than the threshold) reaches a minimum threshold, then that point and all its neighbors are grouped into one cluster. The neighborhood distance threshold is set based on the average point spacing of the point cloud, and is set to 3 times the average point spacing, i.e., 0.05*3=0.15 meters. The minimum threshold is set to 10. This setting is verified through the following experiment: Five sets of sample point clouds with different point densities are selected, and the neighborhood distance thresholds of 2, 3, 4, and 5 times the average point spacing are tested respectively, along with minimum thresholds of 5, 10, and 15, to observe the clustering effect. Experimental data shows that when the distance threshold is 3 times the point spacing and the minimum number threshold is 10, the main wall surface can be effectively separated from structures of different depths such as windows and balconies, with a misclassification rate of less than 2%. For example, when judging P1, if there are points P2, P3, etc. within a 0.15-meter range, and the total number exceeds 10, then P1, P2, P3, etc., are grouped into one cluster (cluster C1); while P4, because its distance from P1 is 1.29 meters, exceeds 0.15 meters, it does not belong to the neighborhood of P1. Finally, a spatial clustering partition set for the facade is generated.

[0028] S102: Based on the spatial clustering partition set of the facade, extract the local neighborhood coordinate set of the point cloud in multiple partitions and calculate the normal vector distribution value. Compare the angle difference of the normal vector with the direction angle of the cluster center. According to the normal deviation angle threshold, redivide the boundary of multiple partitions and generate partition normal feature set. Based on the generated facade spatial clustering partition sets, such as wall partition C1 and window partition C2, the local neighborhood coordinate sets of point clouds within multiple partitions are extracted, and the normal vector distribution values ​​are calculated. Specifically, for a given target point, a certain number of neighboring points are selected to form a local neighborhood. The number of neighboring points, k, is set to 20. This value is verified experimentally: on five sets of standard point cloud models containing typical features such as planes, corners, and curved surfaces, the normal vector estimation effects with k values ​​of 10, 20, 30, and 40 are tested and compared with the model's true normal vectors. Experimental data shows that when k is between 15 and 25, the average angular error between the calculated normal vector and the true normal vector is the smallest, all less than 3 degrees; when k is below 15, it is sensitive to noise; when k is above 25, it leads to smoothing of details at edges and corners. Therefore, k=20 is chosen as a value that balances computational accuracy and detail preservation. For example, extract the coordinates of point P5 (6.55, 10.36, 2.53) and its 20 nearest neighbors in partition C1 to form a local neighborhood coordinate set. Perform principal component analysis on this coordinate set: first, calculate the covariance matrix of the coordinates of these 21 points, and then calculate the eigenvalues ​​and eigenvectors of this matrix. The eigenvector corresponding to the smallest eigenvalue is the normal vector of point P5, denoted as N-P5. The calculated N-P5 is (0.98, 0.18, 0.05). Compare the angle difference of the normal vectors with the cluster center direction angle, where the cluster center direction angle is defined as the angle between the vector pointing from the centroid of the point cloud to the centroid of the partition and the coordinate axis. The centroid of partition C1 is (5.80, 10.34, 2.51), and the centroid of the point cloud is (6.15, 10.50, 3.10), so the cluster center direction vector is (-0.35, -0.16, -0.59). The angle difference between the normal vector N-P5 of point P5 and the direction vector is the normal deviation angle. The boundaries of the multi-partition are redefined based on the normal deviation angle threshold, which is set to 20 degrees. This threshold is set using the following experiment: Prepare a test point cloud with known geometric relationships (e.g., plane, 90-degree exterior angle, 90-degree interior angle), and calculate the normal deviation angle of each point. Data shows that the deviation angle of points in the planar region is distributed between 0-10 degrees, while the deviation angle of points in the 90-degree interior and exterior angle regions is significantly greater than 30 degrees. Therefore, setting the threshold to 20 degrees effectively distinguishes between planar points and boundary points. If the calculated normal deviation angle of point P5 is 8 degrees, less than the 20-degree threshold, it is retained in partition C1. Conversely, if the normal deviation angle of a boundary point is 35 degrees, greater than 20 degrees, it is removed from the original partition and marked as a point to be determined or assigned to an adjacent partition. By performing this judgment on all partition boundary points, a partition normal feature set is generated.

[0029] S103: Based on the partitioned normal feature set, calculate the local radius of curvature of points in multiple partitions, perform a weighted average operation on the spatial distance between points, and couple and map the radius of curvature with the spatial distance results to generate the facade geometric data. The normal deviation angle threshold is determined by statistically analyzing the angular difference characteristics of the distribution of normal vectors in multiple regions of the exterior facade point cloud. Based on the generated partitioned normal feature set, the local radius of curvature of points in multiple partitions is calculated. For point P5 (6.55, 10.36, 2.53) in partition C1, the curvature is estimated using the normal vector information of P5 and its 20 neighboring points. Specifically, the process involves analyzing the degree of change in the normal vectors of each point in P5's neighborhood, quantified by the average angle difference between the neighboring point normal vectors and the P5 normal vector. The radius of curvature is inversely proportional to this average angle difference. For example, if the area where P5 is located is a flat wall, the directions of the normal vectors of its neighboring points are basically consistent, and the average angle difference is close to 0 degrees, resulting in a very large radius of curvature, such as 250.0 meters. If another point P9 is located at the intersection of a wall and a windowsill, the directions of the normal vectors of its neighboring points change significantly, and the average angle difference is large, such as 15 degrees, resulting in a smaller radius of curvature, such as 0.2 meters. A weighted average calculation is performed on the spatial distances between points, and the radius of curvature is coupled and mapped to the spatial distance results. Here, "spatial distance" refers to the Euclidean distance between the point and the centroid of its corresponding partition. The spatial distance between point P5 (6.55, 10.36, 2.53) and the centroid of its partition C1 (5.80, 10.34, 2.51) is 0.75 meters. The specific operation of the coupling mapping involves a weighted sum of the radius of curvature and the spatial distance to generate a comprehensive geometric data value. The calculation method is: Geometric data value = Weight coefficient 1 × Radius of curvature + Weight coefficient 2 × Spatial distance. The weight coefficients are set to balance the contribution of different geometric attributes to the final data value. Calibration was performed using an experimental dataset containing 50 different building facade samples. The radius of curvature and spatial distance of all points in the dataset were normalized, and their contribution to distinguishing the main building structures (walls, windows, eaves) was analyzed. Linear regression analysis determined that when weight coefficient 1 is set to 0.4 and weight coefficient 2 is set to 0.6, the generated geometric data value has the highest distinguishability among different building structures. Therefore, weight coefficient 1 is set to 0.4, and weight coefficient 2 is set to 0.6. Substituting the example of point P5, its normalized radius of curvature is 0.95 (normalized based on global maximum and minimum values), and its normalized spatial distance is 0.30. The geometric data value is 0.4 × 0.95 + 0.6 × 0.30 = 0.38 + 0.18 = 0.56. This result indicates that point P5 is geometrically a stable structural point with small curvature and a certain distance from the partition center. This calculation process is applied to all points to generate the facade geometric data.

[0030] Please see Figure 3 The specific steps of S2 are as follows: S201: Spatial topology analysis propagation chain based on facade geometry data, calls the normal vectors and spatial coordinates of multiple points and groups and pairs the normal vectors of adjacent points, calculates the offset difference of each group of normal components on the three coordinate axes, iteratively calculates the cosine angle difference, and generates a set of normal offset angle sequences. Based on the generated facade geometry data, a spatial topology analysis propagation chain is constructed. This propagation chain is an ordered sequence of points, formed by tracing the nearest neighbor in the point cloud data along a specific direction (e.g., the positive X-axis). Starting from point P5 (6.55, 10.36, 2.53), its nearest neighbor in the positive X-axis direction is P10 (6.60, 10.36, 2.53), and the nearest neighbor of P10 is P11 (6.65, 10.37, 2.54), thus forming the propagation chain P5-P10-P11. The normal vectors and spatial coordinates of adjacent points on the chain are used to group and pair the normal vectors of adjacent points. For example, the normal vectors of point P5 and its neighbor P10 are extracted and denoted as N-P5 and N-P10. According to the calculation in S102, N-P5 is (0.98, 0.18, 0.05). Point P10 is located on the same flat wall surface, and its normal vector N-P10 is calculated to be (0.97, 0.22, 0.06). Calculate the offset difference of this set of normal components on the three coordinate axes, specifically the difference in the X component: Y component difference: Z-component difference: Next, the cosine angle difference is calculated iteratively. For P5 and P10, the dot product of their normal vectors is... Since the normal vector is a unit vector with a magnitude of 1, the cosine of the angle between the two vectors is 0.9932. The angle difference, calculated using the inverse cosine function, is approximately 6.7 degrees. Continuing along the propagation chain, we process the next pair, P10 and P11. We set P11 to begin entering the transition area between the wall and the window frame, with its normal vector N-P11 being (0.92, 0.35, 0.15). The dot product of the normal vectors of P10 and P11 is... The corresponding angular difference is approximately 11.9 degrees. This calculation process is applied to each pair of adjacent points throughout the propagation chain, generating a set of normal offset angle sequences that record continuous angular changes. For example, for a propagation chain containing 10 points, a sequence of 9 angle values ​​is obtained, such as {6.7, 7.1, 6.9, 11.9, 25.5, 26.1, 12.2, 7.5, 7.3}.

[0031] S202: Calculate the normal offset difference based on the normal offset angle sequence set. If the normal offset difference exceeds the curvature gradient threshold, analyze the rate of change of curvature radius of the point and adjacent points. Analyze the positive and negative trends of the rate of change of curvature through difference operations to generate a curvature change trend table. Based on the generated normal offset angle sequence set {6.7, 7.1, 6.9, 11.9, 25.5, 26.1, 12.2, 7.5, 7.3}, the difference in normal offset within the sequence is calculated. Specifically, the difference between the second angle and the first angle is... Degree; the difference between the third and second is Degree; the difference between the fourth and third is Degree; the difference between the fifth and fourth is The angle difference sequence is applied to the entire sequence, resulting in the sequence of adjacent angle differences: {0.4, -0.2, 5.0, 13.6, 0.6, -13.9, -4.7, -0.2}. The absolute value of this normal offset difference is then determined to exceed the curvature gradient threshold, which is set to 10 degrees. This threshold is based on experiments with 20 standard point cloud models containing transition regions of planes, cylinders, spheres, and edges. The experiments calculated the adjacent normal offset angle differences for different geometric transition regions. The results show that the absolute values ​​of the angle differences within the planar region are all less than 3 degrees, the angle differences in the smooth curved surface transition region are between 3 and 8 degrees, while the angle differences at the intersection of the plane and the curved surface or edge are all greater than 12 degrees. Using 10 degrees as the threshold can identify regions where geometric features change drastically, with a false positive rate of less than 1.5% in the experiments. In this example, the absolute values ​​of the angle differences 13.6 and -13.9 both exceed the 10-degree threshold. For these points, analyze the rate of change of their radius of curvature and that of their adjacent points. Locate the pair of points that produce a 13.6-degree angular difference, corresponding to P11 and P12 in the propagation chain. Obtain or calculate the radius of curvature of these points from S103, setting the radius of curvature of P11 to 28.0 meters (smooth transition zone) and the radius of curvature of P12 to 0.8 meters (near the window corner). The rate of change of curvature is analyzed through differential operations, calculated as the difference between the radii of curvature of the two points divided by the Euclidean distance between the two points. The Euclidean distance between P11 and P12 is 0.05 meters. The rate of change is... This is a negative value, indicating a sharp decrease in the radius of curvature and an increase in curvature. Similarly, analyzing the point pair (P13 and P14) that produces a -13.9 degree angle difference, if the radius of curvature of P13 is 0.7 meters and the radius of curvature of P14 is 35.0 meters, then the rate of change is... This is a positive value. Generate a table showing the trend of curvature changes.

[0032] Table 2 Curvature Change Trend Table

[0033] Table 2 shows the curvature and trends of key points in the propagation chain. Points with an absolute rate of change less than 200 are classified as stable, while those greater than 200 are classified as significant. The results indicate that P12 and P14 are key control points for the geometric features.

[0034] S203: Call the curvature change trend table, extract the curvature radius and normal vector of the control points with positive and negative change rates respectively, perform weighted constraint adjustment on the two, update the constraint control state of all points, and generate constraint propagation data; The generated curvature change trend table is used to filter control points with "positively significant" and "negatively significant" trends. According to Table 2, the control points are P12 and P14. For P12 (negatively significant), its curvature radius is extracted to be 0.8 meters, and its normal vector N-P12 is (0.85, 0.40, 0.34). For P14 (positively significant), its curvature radius is extracted to be 35.0 meters, and its normal vector N-P14 is (0.96, 0.26, 0.10). Weighted constraint adjustment is performed on the curvature radius and normal vector. A "constraint strength" value is generated for each control point, which quantifies the strength of the point as a geometric feature point. The constraint strength is calculated as follows: Constraint strength = Curvature weight × (1 / Normalized curvature radius) + Change rate weight × Normalized curvature change rate. The normalization process scales the value to between 0 and 1. The curvature weight and rate of change weight were set to 0.7 and 0.3, respectively. This weight configuration was tested on point cloud data of 10 buildings with different styles. The performance of three weight sets (0.5, 0.5), (0.7, 0.3), and (0.3, 0.7) in feature point extraction accuracy was compared. The results showed that the combination of a curvature weight of 0.7 and a rate of change weight of 0.3 achieved the highest accuracy (96%) in identifying corners and edges that matched the actual building structure. Taking P12 as an example, the maximum curvature radius in the dataset was set to 250 meters, and the absolute value of the maximum rate of change was 700. The normalized curvature radius of P12 was... The normalized curvature change rate of P12 is... Therefore, the constraint strength of P12 is calculated as follows: The advantage of this approach is that by weighting the radius of curvature and the rate of change of curvature, abrupt changes in geometric features (such as corners) can obtain extremely high constraint strength values, thus being prioritized for identification and anchoring in subsequent processing. The constraint control state is updated for all points. The constraint state of a control point is its calculated constraint strength value. The constraint state of non-control points is updated through propagation. Specifically, for any non-control point, its constraint state value is equal to the sum of the influences of all upstream control points on it in the propagation chain. The influence of a single control point is defined as its constraint strength value multiplied by a distance-decaying coefficient, calculated as 1 divided by (1 plus the square of the distance between the point and the control point in the propagation chain). For example, if the distance between point P11 and control point P12 is 1, then the constraint state value obtained by P11 from P12 is 218.98 × [1 / (1+1...]. 2 )]=109.49. This method performs calculations on all points in the propagation chain to generate constraint propagation data.

[0035] Please see Figure 4 The specific steps of S3 are as follows: S301: Obtain the set of normal direction vectors of control points in the constraint propagation data, perform differential calculation on the normal direction vectors of adjacent control points to calculate the direction offset angle value, compare the direction offset angle value with the normal deviation angle threshold, remove abnormal control point vectors, and generate a normal displacement differential dataset. The generated constraint propagation data is acquired, and control points with constraint strength values ​​higher than a preset strength threshold and their normal direction vectors are extracted to form an initial set of control point normal direction vectors. The strength threshold is set to 100.0, based on the following experiment: the constraint strength of all points is calculated on 30 sets of building point cloud samples containing clear edges, corners, and smooth transition areas. Experimental data shows that the constraint strength values ​​of points representing the main structural lines of the building (such as wall corners and window frame edges) are all above 100.0, while the constraint strength values ​​of flat areas or points with small noise are all below 50.0. Therefore, choosing 100.0 as the threshold can effectively filter out key geometric feature points. Based on this standard, P12 (constraint strength 218.98) and P14 (constraint strength can be estimated as a high value based on its positive significant rate of change, set to 235.50 here) are selected from the results of S202 and S203 as adjacent control points, and control point P18 (constraint strength 150.70) is further filtered out after setting the propagation chain. The corresponding normal vectors are N-P12(0.85, 0.40, 0.34), N-P14(0.96, 0.26, 0.10), and N-P18(0.45, 0.88, 0.17). The directional offset angle is obtained by performing a difference calculation on the normal vectors of adjacent control points. This calculation process is as follows: first, calculate the dot product of adjacent normal vectors, and then obtain the included angle using the inverse cosine function. Taking N-P12 and N-P14 as examples, their dot product is (0.85×0.96)+(0.40×0.26)+(0.34×0.10)=0.816+0.104+0.034=0.954, corresponding to a directional offset angle of 17.4 degrees. Next, the dot product of N-P14 and N-P18 is calculated: (0.96×0.45)+(0.26×0.88)+(0.10×0.17)=0.432+0.2288+0.017=0.6778, corresponding to a directional offset angle of 47.3 degrees. These angle values ​​are compared with a normal deviation angle threshold, which is set at 40 degrees. The experimental verification process for this threshold is as follows: point cloud models of 10 buildings with different structural styles (Gothic, Modernist, etc.) are analyzed, and control point sequences on their clear continuous feature lines (such as window sill outlines and roof lines) are extracted. The angle difference of the normal vectors of adjacent control points is calculated. Statistical analysis of experimental data shows that more than 98% of adjacent control points belonging to the same continuous structural line have a normal vector angle difference of less than 40 degrees. Therefore, 40 degrees is set as the threshold to determine whether control points belong to the same geometric feature. The angle between P12 and P14 is 17.4 degrees, which is less than 40 degrees, so they are considered continuous features. The angle between P14 and P18 is 47.3 degrees, which is greater than 40 degrees, so P18 is considered an outlier control point, presumably belonging to another irrelevant feature line or a spurious feature point caused by noise. Therefore, the vector of P18 is removed from the dataset.This process iterates through all control point pairs to generate a normal displacement difference dataset.

[0036] S302: Based on the normal displacement difference dataset, call the displacement vector difference between adjacent control points and perform amplitude analysis, calculate the displacement change rate sequence to extract the continuous interval of change rate, calculate the mean of the change rate within the interval as the smoothing coefficient, and generate a set of control point smoothing coefficients. Based on the generated normal displacement difference dataset, which contains a filtered sequence of continuous control points, such as {P12, P14}, the displacement vector difference between adjacent control points in this sequence is used for magnitude analysis. The displacement vector difference is the vector difference between the spatial coordinates of two points. The coordinates of point P12 are obtained or set as (6.70, 10.38, 2.55), and the coordinates of point P14 are set as (6.72, 10.38, 3.75). The coordinates of the next continuous control point P16 in the sequence are set as (6.75, 10.39, 4.94). First, the displacement vector difference between P12 and P14 is calculated: the difference in coordinates X is... Meters, Y difference is meters, Z difference is Meters. The displacement vector is (0.02, 0.00, 1.20). Its magnitude (i.e., the Euclidean distance between the two points) is the square root of the sum of the squares of the three components, i.e. Meters. Next, calculate the displacement vector difference between P14 and P16: (6.75-6.72, 10.39-10.38, 4.94-3.75) = (0.03, 0.01, 1.19), with an amplitude of Meters. Apply this process to the entire control point sequence to obtain a displacement amplitude sequence, such as {1.2002, 1.1908, 1.2105}. Then calculate the displacement rate of change sequence, which is the ratio of each subsequent amplitude to the previous amplitude. The first rate of change is... The second rate of change The resulting rate of change sequence is {0.992, 1.017}. A continuous interval of rate of change is extracted, i.e., a continuous subsequence of rate of change values ​​falling within a specific range. This range is set to [0.90, 1.10]. This interval is based on the analysis of point clouds of regular geometries (such as standard doors and windows), where equidistant feature points calculate displacement rate of change fluctuations within the range of 0.90 to 1.10. If the rate of change is within this interval, it indicates that the density between control points is relatively uniform. In this example, both 0.992 and 1.017 are within [0.90, 1.10], therefore they constitute a continuous interval of rate of change. The mean of the rate of change within this interval is calculated as a smoothing coefficient, with a mean of [value missing]. This smoothing coefficient will be applied to the control points (P14 and P16) corresponding to this interval. A set of control point smoothing coefficients is generated as shown in Table 3.

[0037] Table 3. Smoothing Coefficients for Control Points

[0038] As shown in Table 3, this table lists the displacement change rates of some control points and the smoothing coefficient values ​​of their respective continuous intervals. The change rate of point P19, 1.350, exceeds the range of [0.90, 1.10], therefore it forms its own interval, and its smoothing coefficient is its own change rate value.

[0039] S303: Based on the set of control point smoothing coefficients, perform weighted smoothing on the coordinates of control points in the constraint propagation data, use the smoothing coefficients as weights in the control point coordinate update calculation, calculate the difference in control point coordinates after the update and perform coordinate aggregation to generate a set of exterior facade correction surfaces. Based on the generated set of control point smoothing coefficients, the coordinates of the control points in the constraint propagation data are subjected to weighted smoothing. This process updates the position of each control point using the coordinates of its immediate and adjacent control points, as well as its own smoothing coefficient. The smoothing coefficient is used as a weight in the control point coordinate update calculation. Specifically, the new coordinates of a control point are equal to the weighted average of its old coordinates and the midpoints of its two immediate and adjacent coordinates. The weights are assigned as follows: the weight of the old coordinates is (1 - smoothing coefficient), and the weight of the midpoints of the adjacent coordinates is the smoothing coefficient. When the smoothing coefficient is close to 1.0, the new coordinates will be more biased towards the midpoints of the two adjacent points. For example, control point P14 has coordinates of (6.72, 10.38, 3.75), the previous point P12 has coordinates of (6.70, 10.38, 2.55), and the next point P16 has coordinates of (6.75, 10.39, 4.94). The smoothing coefficient for P14, obtained from Table 3, is 1.0045. Since this coefficient is greater than 1, it indicates that the point needs to be expanded outwards towards the midpoint of its neighbor. Therefore, the weight calculation is adjusted as follows: New coordinates = Old coordinates + Smoothing coefficient × (Neighbor midpoint - Old coordinates). The midpoint coordinates of P12 and P16 are (6.70 + 6.75) / 2, (10.38 + 10.39) / 2, and (2.55 + 4.94) / 2, respectively. The vector difference between the midpoint and the old coordinates on P14 is... The new coordinates for P14 are calculated as follows: The advantage of this approach is that by introducing a smoothing coefficient as a dynamic weight, areas with uneven point distribution (rate of change deviating from 1) can be adjusted more significantly, while areas with uniform distribution remain stable, thus preserving the macroscopic shape of the feature lines while smoothing noise. After performing this update on all control points, the coordinate difference between the updated control points is calculated. For example, P12 remains unchanged, and the new coordinates of P14 are (6.725, 10.385, 3.745), calculating a new distance of approximately 1.196 meters between the two points. Finally, coordinate aggregation is performed, which uses the updated control point sequence as interpolation control points to generate a smooth B-spline curve. This curve represents a key geometric feature line on the building facade, such as the outline of a window. All calculated feature lines (such as corner lines, window sill lines, eaves lines, etc.) are combined and used as a skeleton to refit the original point cloud data or partitions, generating a set of modified facade surfaces.

[0040] Please see Figure 5 The specific steps of S4 are as follows: S401: Obtain the boundary coordinate data of adjacent geometric regions in the set of modified surfaces of the facade, calculate the overlapping area of ​​adjacent boundaries, and calculate the ratio of the overlapping area to the area enclosed by the two regions' boundaries. Record the overlap ratio sequence and filter out abnormal ratio items to generate a boundary overlap ratio sequence set. Obtain the generated set of modified facade surfaces. This set contains several geometrically optimized regions, such as the large wall area labeled R1 and its adjacent window sill area R2. Extract the boundary coordinate data of these two adjacent geometric regions, which are ordered sequences of 3D points defined by the B-spline curve control points in S303. Perform an overlap area calculation on the adjacent boundaries of R1 and R2. Specifically, first, project the boundary coordinate points of the two regions onto a common reference plane, which is determined by the average normal vector of region R1. Then, on the 2D projection plane, the boundaries of R1 and R2 form two polygons. Calculate the area of ​​the intersection of these two polygons, which is the overlap area. For example, the projected area of ​​R1 (i.e., the area enclosed by its boundaries) is calculated to be 12.5 square meters, and the projected area of ​​R2 is 1.8 square meters. Through geometric calculations, the overlap area on the projection plane is found to be 0.9 square meters. Next, calculate the ratio of this overlap area to the area enclosed by the boundaries of the two regions. The ratio here is calculated as follows: Overlap ratio = Overlap area / (Area R1 + Area R2). Substituting the example data, the overlap ratio is... This calculation is applied to all adjacent region pairs, recording them to form an overlap ratio sequence. This sequence is then filtered to remove outlier ratios. An outlier ratio is defined as an item with a value below a specific lower threshold, set at 0.01. This threshold was verified through the following experiment: Point cloud models of 20 buildings were selected, containing 500 pairs of clearly separated regions (e.g., non-touching windows) and 500 pairs of clearly contacting regions (e.g., walls and window frames). The overlap ratio of all region pairs was calculated. Experimental data showed that for clearly separated region pairs, the overlap ratio caused by point cloud noise or calculation errors was below 0.005; while for clearly contacting region pairs, the overlap ratio was above 0.015. Therefore, setting the threshold to 0.01 can effectively filter out weak overlaps caused by calculation errors. If the calculated overlap ratio of another pair of regions R3 and R4 is 0.004, since this value is below 0.01, this item will be considered an outlier ratio and removed from the sequence. After filtering, a boundary overlap ratio sequence set is generated.

[0041] S402: Call the adjacent regions in the boundary overlap ratio sequence set, calculate the angle between the normal vectors of the adjacent regions based on the surface normal vector data of the multi-region region, compare the angle value with the normal consistency threshold, filter the region pairs whose angle offset meets the normal consistency threshold condition, and generate the normal angle offset sequence set; The generated boundary overlap ratio sequence set is invoked, which contains region pairs such as (R1, R2) that have passed the initial overlap ratio screening. The angle between the normal vectors of adjacent regions is calculated based on the multi-region surface normal vector data. The surface normal vector data for each region is calculated in S102 and may be updated in subsequent steps; here, the average of the normal vectors of all points within the region is taken as the representative normal vector for that region. For example, the average normal vector N-R1 of region R1 is (0.996, -0.087, 0.012), and the average normal vector N-R2 of region R2 is (0.990, -0.130, 0.025). The angle between these two vectors is calculated by first calculating the dot product of the two unit vectors, i.e. The included angle was then calculated using the inverse cosine function, and its value was approximately 4.0 degrees. This included angle value was compared with a normal consistency threshold, which was set to 15 degrees. This threshold was based on point cloud data analysis of 30 standard geometric models containing intersecting planes at different angles. In the experiment, the included angle of the average normal vector of the point set representing each plane was calculated. The results showed that when the included angle between two planes is less than 15 degrees, the surfaces appear as continuous or smoothly transitioning visually and geometrically. When the included angle is greater than 15 degrees, a clear transition is observed. Therefore, 15 degrees was chosen as the standard for judging whether two regions are consistent in the normal direction. In this example, the included angle of normals between R1 and R2, 4.0 degrees, is less than 15 degrees, therefore, this pair of regions is determined to satisfy the normal consistency condition. Taking another pair of regions (R5, R6) as an example, its calculated normal angle is 88.2 degrees, which is much greater than 15 degrees. This indicates that the two regions (such as the wall and the vertical window side) are inconsistent in the normal direction, so this pair of regions will not be selected to proceed to the next step. By performing this angle calculation and comparison on all pairs of regions in the boundary overlap ratio sequence set, regions that meet the normal consistency threshold condition are selected, and a normal angle offset sequence set is generated.

[0042] S403: Based on the region pairs in the normal angle offset sequence set, extract the curvature parameter distribution of the corresponding region, calculate the curvature change rate of adjacent regions, and mark the fusion region when any two parameters among the boundary overlap ratio, normal angle offset and curvature consistency meet the consistency threshold, and generate a morphological fusion instruction set. The consistency threshold is set by analyzing the distribution pattern of characteristic parameters in multiple regions of the facade correction surface. The normal consistency threshold is set by analyzing the difference characteristics of the normal vectors of adjacent geometric regions in the modified surface set of the facade; Based on the generated set of normal angle offset sequences, which includes region pairs such as (R1, R2), the curvature parameter distribution of the corresponding regions in these region pairs is extracted. The curvature parameter distribution is part of the generated geometric data; here, the average curvature value of all points within each region is extracted as the representative curvature of that region. For example, the average curvature of region R1 (large wall surface) is 0.002, and the average curvature of region R2 (windowsill) is 0.003. The rate of change of curvature between these two adjacent regions is calculated by dividing the absolute value of the difference between the average curvatures of the two regions by the smaller of the two values. Next, we determine whether the three parameters—boundary overlap ratio, normal angle offset, and curvature consistency—satisfy their respective consistency thresholds. The setting of these three thresholds and their experimental verification process are shown in Table 4.

[0043] Table 4. Threshold Parameters for Morphological Fusion Consistency

[0044] As shown in Table 4, three parameters are used to judge the region pair (R1, R2): its boundary overlap ratio is 0.0629 (greater than 0.03, satisfied), its normal angle offset is 4.0 degrees (less than 15 degrees, satisfied), and its curvature change rate is 0.5 (less than 0.8, satisfied). When any two of these three parameters meet the corresponding consistency threshold, the pair of regions is marked as a fused region. In this example, all three parameters of (R1, R2) are satisfied, so it is marked as a fused region. The advantage of this method is that by setting the judgment condition that any two of the boundary overlap ratio, normal angle offset, and curvature consistency meet the threshold, it is possible to correctly fuse regions that should belong to the same whole based on the other two macroscopically consistent geometric features even if one parameter fails to meet the standard due to local noise or minor structural changes in the point cloud data. For example, even if the curvature of a pair of regions changes significantly due to local contamination (curvature change rate greater than 0.8), as long as their overlap is high and their normals are basically consistent, they will still be judged as fused. This judgment logic is applied to all region pairs that have passed the S402 filter, generating an instruction list that contains information on all region pairs that should be merged. This list is the morphological fusion instruction set.

[0045] Please see Figure 6 The specific steps of S5 are as follows: S501: Based on the control point coordinate data of adjacent regions in the morphological fusion instruction set, the relative positions of multiple control points in three-dimensional space are detected, the magnitude of the spatial coordinate difference vector is calculated, and abnormal deviations in the magnitude sequence are removed and smoothed to generate a spatial distance difference sequence. Based on the generated morphological fusion instruction set, adjacent region pairs marked as fusion regions are extracted, such as region R1 and region R2. The coordinate data of control points on the shared boundary of these two regions are obtained, forming an ordered sequence of control points, denoted as {CP1, CP2, CP3, CP4, CP5}. The relative positions of adjacent control points in this sequence in 3D space are detected. This detection is accomplished by calculating the magnitude of the spatial coordinate difference vector. Taking points CP2 and CP3 as an example, the coordinates of CP2 obtained from the coordinate update result of S303 are set to (6.725, 10.385, 3.745) meters, and the coordinates of CP3 are set to (6.751, 10.394, 4.935) meters. 6.751 - 6.725 = 0.026 meters; Y-axis difference: 10.394 - 10.385 = 0.009 meters; Z-axis difference: 4.935 - 3.745 = 1.190 meters. The coordinate difference vector is (0.026, 0.009, 1.190). Next, the magnitude of this vector, i.e., the Euclidean distance between the two points, is calculated: Meters. This calculation process is applied to all adjacent point pairs {CP1-CP2, CP2-CP3, CP3-CP4, CP4-CP5} in the sequence, generating a modulus sequence, for example, {1.2002, 1.1903, 1.5520, 1.1895} meters. Then, outliers in this modulus sequence are removed. The removal process is as follows: first, the mean and standard deviation of the sequence are calculated, with the mean being... The standard deviation is approximately 0.17 meters. Items exceeding the mean ± 2 standard deviations are defined as outliers. This 2 standard deviation criterion is based on the analysis of 20 sets of building feature line segment samples containing 5000 control points. Statistical results show that over 95% of the continuous point spacing falls within the range of the mean ± 2 standard deviations. In this example, the judgment interval is... The range is [0.943, 1.623] meters. The modulus 1.5520 falls within this range, but due to a data acquisition error, the distance between CP4 and CP5 was recorded as 0.5 meters, so this value will be removed. Taking the original sequence as an example, the value 1.5520 is removed because it has a significant jump compared to other values. The sequence after removal is {1.2002, 1.1903, 1.1895}. The sequence after removing outliers is smoothed using a moving average with a window size of 3. For a value in the sequence, its smoothed new value is replaced by the average of itself and its immediate and next-to-immediate (if any) values. For example, the new value of 1.1903 is... / 3=1.1933. Apply this smoothing operation to the entire sequence to generate a set of spatial distance difference sequences.

[0046] S502: Call the spatial distance difference sequence, retrieve the normal vector parameters of the corresponding control points, calculate based on the difference in the normal angle between adjacent control points, compare with the direction deviation reference value, filter the normal changes that meet the direction deviation reference value, and generate a normal direction deviation sequence set; The generated spatial distance difference sequence corresponds to the control point sequence {CP1, CP2, CP3}, and the normal vector parameters of these control points are retrieved. These normal vectors are calculated in S102 and may be adjusted in S303. Let the normal vector N-CP2 of control point CP2 be (0.84, 0.41, 0.35), and the normal vector N-CP3 of CP3 be (0.82, 0.45, 0.36). The angle difference between these two adjacent control points is calculated based on their normal vectors. The calculation process involves first finding the dot product of the two unit normal vectors: The dot product value is the cosine of the angle between the two vectors. The angle difference, obtained using the inverse cosine function, is approximately 2.1 degrees. This calculation is applied to all adjacent control point pairs in the sequence, resulting in a normal angle difference sequence, for example, {1.9, 2.1, 15.3, 2.5} degrees. Each angle difference in the sequence is compared to a direction deviation benchmark value, set at 10 degrees. This benchmark value was determined through analysis of continuous structural lines (such as eaves and waistlines) totaling over 5000 meters in length from 50 different architectural models (covering modern, classical, and industrial styles). Control points were extracted from these known, smooth, continuous structural lines, and the normal angle difference between adjacent points was calculated. Statistical data showed that 99% of the angle differences were below 10 degrees. Therefore, 10 degrees was used as the benchmark for judging whether the normal change is smooth and continuous. For the angle difference sequence {1.9, 2.1, 15.3, 2.5}, each item is compared: 1.9 degrees is less than 10 degrees, which meets the requirement; 2.1 degrees is less than 10 degrees, which meets the requirement; 15.3 degrees is greater than 10 degrees, which does not meet the requirement; 2.5 degrees is less than 10 degrees, which meets the requirement. Through this comparison process, all items with angle differences less than 10 degrees are selected. These items constitute the set of normal changes that meet the direction deviation reference value. Finally, these qualified angle difference values ​​are combined to generate the normal direction deviation sequence set, such as {1.9, 2.1, 2.5}.

[0047] S503: Based on the spatial distance difference sequence set and the normal direction deviation sequence set, extract the tangential direction vector of adjacent control points, calculate the continuity offset between adjacent tangential directions, perform weighted superposition of the three types of sequences and perform proportional calculation, adjust the coordinates of control points and normal directions, and generate the facade scanning results. The directional deviation baseline value is set by statistically analyzing the directional difference characteristics of the normal vectors of adjacent control points after morphological fusion. Based on the generated spatial distance difference sequence set and the generated normal direction deviation sequence set, the tangential direction vectors of adjacent control points are extracted. For the control point sequence {CP1, CP2, CP3, CP4}, the tangential direction vector T-CP2 of control point CP2 is obtained by normalizing the coordinate difference vector between it and the next point CP3. Using the coordinates of CP2 and CP3 in S501, their coordinate difference vector is (0.026, 0.009, 1.190), with a magnitude of 1.1903. After normalization, T-CP2 is... Similarly, calculate the tangential direction vector T-CP3 of CP3, set to (0.025, 0.007, 0.999). Calculate the continuity offset between these two adjacent tangential vectors, i.e., calculate the angle between them. The dot product is (0.022*0.025)+(0.008*0.007)+(0.999*0.999)= The included angle is approximately 3.0 degrees. This yields the third sequence, the tangential continuity offset sequence. These three sequences (spatial distance difference, normal direction deviation, and tangential continuity offset) are then weighted and superimposed. First, the values ​​of the three sequences are normalized and mapped to a unified interval. Then, the normalized values ​​are weighted and summed to obtain a comprehensive adjustment factor. The weight coefficients are set as follows: spatial distance weight... Normal direction weight Tangential direction weight The experimental verification process for these weights is as follows: Ten sets of standard architectural model point clouds with different levels of noise and deformation were constructed. Different weight combinations (e.g., {0.3, 0.4, 0.3}, {0.2, 0.5, 0.3}, etc.) were applied for final coordinate adjustment, and the results were compared with the original noise-free model to calculate the root mean square error. The results show that the combination with the highest weight (0.5) for normal deviation has the best restoration effect and the lowest root mean square error, which is 15% lower than other combinations on average. Taking CP3 as an example, its normalized three deviation values ​​are set to 0.1 (distance), 0.2 (normal), and 0.15 (tangential), then its comprehensive adjustment factor is... The coordinates and normal direction of control point CP3 are adjusted. The coordinate adjustment is the factor multiplied by a vector pointing towards the midpoint of the neighboring points, and the normal adjustment is the factor multiplied by a vector pointing towards the mean normal of the neighboring points. After adjusting all control points, the facade scan result is generated through surface reconstruction calculation.

[0048] Please see Figure 7 A building facade scanning system includes: The data clustering module acquires point cloud data of the building facade and performs spatial normal distribution clustering. It divides the point cloud into geometric regions and calculates the point normal vectors, radii of curvature, and spatial distances between points in multiple regions, generating facade geometric data and transferring it to the topology analysis module. The topology analysis module analyzes the propagation chain of spatial topology order based on the facade geometric data, calculates the normal offset difference between adjacent points, and if the offset difference exceeds the curvature gradient threshold, analyzes the curvature change trend and constrains the curvature and normal of the control points, generates constrained propagation data and transmits it to the normal adjustment module. The normal adjustment module calls the constraint propagation data to adjust the displacement of the control points in the normal direction, calculates the displacement difference between adjacent control points and analyzes the smoothing coefficient, updates the control point coordinates with weighted smoothing, generates the set of exterior facade correction surfaces and passes them to the fusion analysis module. The fusion analysis module, based on the exterior facade modified surface set, calculates the boundary overlap ratio, normal angle offset and curvature consistency parameters of adjacent geometric regions. When any two parameters meet the consistency threshold, it generates a morphology fusion instruction set and passes it to the morphology fusion module. The morphology fusion module calls the morphology fusion instruction set, calculates the spatial distance difference of control points, the deviation of the normal direction and the tangential continuity offset sequence and weights them together, performs proportional adjustment on the coordinates of the control points and the normal direction, and generates the facade scanning result.

[0049] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for scanning building facades, characterized in that, Includes the following steps: S1: Acquire point cloud data of building facade and perform spatial normal distribution clustering. Divide the point cloud into geometric regions and calculate the point normal vector, radius of curvature and spatial distance between points in multiple regions to generate facade geometric data. S2: Based on the spatial topological order analysis propagation chain of the facade geometry data, calculate the normal offset difference between adjacent points. If the offset difference exceeds the curvature gradient threshold, analyze the curvature change trend and constrain the curvature and normal of the control points to generate constrained propagation data. S3: Call the constraint propagation data to adjust the displacement of the control point normal direction, calculate the displacement difference between adjacent control points and analyze the smoothing coefficient, update the control point coordinates with weighted smoothing, and generate the exterior facade correction surface set. S4: Based on the modified surface set of the facade, calculate the boundary overlap ratio, normal angle offset and curvature consistency parameter of adjacent geometric regions. When any two parameters meet the consistency threshold, generate a morphological fusion instruction set. S5: Call the aforementioned morphological fusion instruction set, calculate the spatial distance difference of control points, the deviation of the normal direction and the tangential continuity offset sequence and weight them, perform proportional adjustment on the coordinates of the control points and the normal direction, and generate the facade scanning result.

2. The method for scanning building facades according to claim 1, characterized in that, The facade geometric data includes geometric region division data, point normal vector set, curvature radius parameter set, and spatial distance between points. The constraint propagation data includes normal offset difference sequence and curvature change constraint parameters. The facade correction surface set includes smoothing coefficient, adjacent control point displacement difference parameter, and corrected normal vector set. The morphology fusion instruction set includes boundary overlap ratio parameter, normal angle offset parameter, and curvature consistency parameter. The facade scanning results include control point spatial distance difference set, normal direction deviation sequence, and tangential continuity offset sequence.

3. The method for scanning building facades according to claim 1, characterized in that, The specific steps of S1 are as follows: S101: Obtain the point cloud dataset of the building facade, organize the three-dimensional coordinate parameters of multiple points in the point cloud in a structured manner, calculate the Euclidean distance between multiple points based on the coordinate neighborhood relationship, and then perform spatial clustering on the distance value sequence to generate the facade spatial clustering partition set. S102: Based on the aforementioned facade space clustering partition set, extract the local neighborhood coordinate set of point clouds within multiple partitions and calculate the normal vector distribution value. Compare the angle difference of the normal vector with the direction angle of the cluster center. Based on the normal deviation angle threshold, redivide the boundaries of multiple partitions and generate partition normal feature sets. S103: Based on the partitioned normal feature set, calculate the local curvature radius of the points in the multiple partitions, perform a weighted average operation on the spatial distance between the points, and couple and map the curvature radius with the spatial distance result to generate the facade geometric data. The normal deviation angle threshold is determined by statistically analyzing the angular difference characteristics of the distribution of normal vectors in multiple regions of the facade point cloud.

4. The method for scanning building facades according to claim 1, characterized in that, The specific steps of S2 are as follows: S201: Based on the spatial topology analysis propagation chain of the facade geometry data, call the normal vectors and spatial coordinates of multiple points and group and pair the normal vectors of adjacent points, calculate the offset difference of each group of normal components on the three coordinate axes, iteratively calculate the cosine angle difference, and generate a normal offset angle sequence set. S202: Calculate the normal offset difference based on the normal offset angle sequence set. If the normal offset difference exceeds the curvature gradient threshold, analyze the rate of change of curvature radius of the point and adjacent points. Analyze the positive and negative trends of the rate of change of curvature through differential operation to generate a curvature change trend table. S203: Call the curvature change trend table, extract the curvature radius and normal vector of the corresponding control points for positive and negative values ​​of the rate of change, perform weighted constraint adjustment on the two, update the constraint control state of all points, and generate constraint propagation data.

5. The method for scanning building facades according to claim 4, characterized in that, The curvature gradient threshold is obtained by analyzing the variation distribution characteristics of the curvature radius in multiple regions of the facade point cloud. Specifically, after extracting the curvature radius of multiple region points, the difference value sequence of the curvature radius of adjacent points is calculated, and frequency statistics are performed on the difference sequence.

6. The method for scanning building facades according to claim 1, characterized in that, The specific steps for S3 are as follows: S301: Obtain the set of normal direction vectors of control points in the constraint propagation data, perform differential calculation on the normal direction vectors of adjacent control points to calculate the direction offset angle value, compare the direction offset angle value with the normal deviation angle threshold, remove abnormal control point vectors, and generate a normal displacement differential dataset. S302: Based on the normal displacement difference dataset, call the displacement vector difference between adjacent control points and perform amplitude analysis, calculate the displacement change rate sequence to extract the continuous interval of change rate, calculate the mean change rate within the interval as a smoothing coefficient, and generate a set of control point smoothing coefficients. S303: Based on the set of control point smoothing coefficients, perform weighted smoothing on the control point coordinates in the constraint propagation data, use the smoothing coefficients as weights in the control point coordinate update calculation, calculate the difference in the updated control point coordinates and perform coordinate aggregation to generate a set of exterior facade correction surfaces.

7. The method for scanning building facades according to claim 6, characterized in that, The normal deviation angle threshold is determined by analyzing the angular distribution characteristics of the normal vectors in multiple regions of the facade point cloud, extracting the normal vectors of multiple regions, calculating the angle between the normal vectors of adjacent points, and analyzing the angle distribution curve of the angle value to determine the stable range of the normal direction.

8. The method for scanning building facades according to claim 1, characterized in that, The specific steps of S4 are as follows: S401: Obtain the boundary coordinate data of adjacent geometric regions in the set of modified surfaces of the facade, calculate the overlapping area of ​​adjacent boundaries, and obtain the ratio of the overlapping area to the area enclosed by the two regions, record the overlap ratio sequence and filter out abnormal ratio items to generate a boundary overlap ratio sequence set. S402: Call the adjacent regions in the boundary overlap ratio sequence set, calculate the angle between the normal vectors of the adjacent regions based on the multi-region surface normal vector data, compare the angle value with the normal consistency threshold, filter the region pairs whose angle offset meets the normal consistency threshold condition, and generate a normal angle offset sequence set; S403: Based on the region pairs in the normal angle offset sequence set, extract the curvature parameter distribution of the corresponding region, calculate the curvature change rate of adjacent regions, and mark the fusion region when any two parameters among the boundary overlap ratio, normal angle offset and curvature consistency meet the consistency threshold, and generate a morphological fusion instruction set. The consistency threshold is set by analyzing the distribution pattern of characteristic parameters in multiple regions of the facade correction surface. The normal consistency threshold is set by analyzing the difference characteristics of the normal vectors of adjacent geometric regions in the facade correction surface set.

9. The method for scanning building facades according to claim 1, characterized in that, The specific steps of S5 are as follows: S501: Based on the control point coordinate data of adjacent regions in the morphological fusion instruction set, the relative positions of multiple control points in three-dimensional space are detected, the magnitude of the spatial coordinate difference vector is calculated, and abnormal deviations in the magnitude sequence are removed and smoothed to generate a spatial distance difference sequence. S502: Call the spatial distance difference sequence, retrieve the normal vector parameters of the corresponding control points, calculate based on the difference in normal angle between adjacent control points, compare with the direction deviation reference value, filter the normal changes that meet the direction deviation reference value, and generate a normal direction deviation sequence set; S503: Based on the spatial distance difference sequence set and the normal direction deviation sequence set, extract the tangential direction vector of adjacent control points, calculate the continuity offset between adjacent tangential directions, perform weighted superposition of the three types of sequences and perform proportional calculation, adjust the coordinates of control points and normal directions, and generate the facade scanning result. The directional deviation reference value is set by statistically analyzing the directional difference characteristics of the normal vectors of adjacent control points after morphological fusion.

10. A building facade scanning system, characterized in that, The system is used to implement the building facade scanning method according to any one of claims 1-9, the system comprising: The data clustering module acquires point cloud data of the building facade and performs spatial normal distribution clustering. It divides the point cloud into geometric regions and calculates the point normal vectors, radii of curvature, and spatial distances between points in multiple regions, generating facade geometric data and transferring it to the topology analysis module. The topology analysis module analyzes the propagation chain based on the spatial topology order of the facade geometric data, calculates the normal offset difference between adjacent points, and if the offset difference exceeds the curvature gradient threshold, analyzes the curvature change trend and constrains the curvature and normal of the control points, generates constrained propagation data and transmits it to the normal adjustment module. The normal adjustment module calls the constraint propagation data to adjust the displacement of the control point normal direction, calculates the displacement difference between adjacent control points and analyzes the smoothing coefficient, updates the control point coordinates with weighted smoothing, generates the exterior facade correction surface set and passes it to the fusion analysis module. The fusion analysis module calculates the boundary overlap ratio, normal angle offset and curvature consistency parameters of adjacent geometric regions based on the exterior modified surface set. When any two parameters meet the consistency threshold, a morphological fusion instruction set is generated and transmitted to the morphological fusion module. The morphology fusion module calls the morphology fusion instruction set to calculate the spatial distance difference of control points, the deviation of the normal direction and the tangential continuity offset sequence and weights them together. It then performs proportional adjustments on the coordinates of the control points and the normal direction to generate the facade scanning results.

Images