Urban building three-dimensional fine model extraction method based on tomographic SAR point cloud regularization

Abstract

The invention provides an urban building three-dimensional fine model extraction method based on tomographic SAR point cloud regularization. The method comprises the following steps: separating building point cloud and ground point cloud from tomographic SAR point cloud; segmenting the building point cloud to obtain a facade point cloud and a roof point cloud; in the presence of the roof point cloud, reconstructing a roof contour by using the roof point cloud; under the condition that the roof point cloud is missing, the building thickness is estimated by using a ground blank area; and reconstructing a building three-dimensional model by using the roof contour reconstructed under the existence or absence of the facade footprint and the roof point cloud. According to the method, the facade is gradually segmented from thick to thin along the wall surface direction to reconstruct the fine facade footprint, and the representative line segments of the roof contour are selected and connected by combining the facade direction, so that the regular roof contour can be simply and conveniently extracted; and in the absence of the roof point cloud, estimating the thickness of the building roof by using the shielded area of the ground, and reconstructing a three-dimensional model of the building.

Description

Technical Field

[0001] The present invention belongs to the technical field of SAR point cloud processing, and particularly relates to a method for extracting a three-dimensional fine model of urban buildings based on the regularization of tomographic SAR point clouds. Background Art

[0002] Tomographic synthetic aperture radar technology (TomoSAR) is an advanced interferometric technique aimed at achieving true three-dimensional synthetic aperture radar imaging. Through spectral analysis, a synthetic aperture is established in the elevation direction to three-dimensionally locate each scatterer, and it can resolve multiple / principal scatterers within a resolution cell. Converting the high-density scatterer distribution obtained from TomoSAR into the ground coordinate system can generate high-quality TomoSAR point clouds, which contain rich three-dimensional position information of scatterers and are thus very suitable for generating three-dimensional urban models. However, compared with light detection and ranging (LiDAR), airborne TomoSAR point clouds face various challenges, such as incomplete point clouds caused by shadows, excessive noise, and low three-dimensional accuracy, which pose certain challenges to modeling.

[0003] Most existing methods use linear fitting to extract the elevation positions of roofs, use region growing on both sides of the elevation to extract roof point clouds, and then perform contour reconstruction based on the roof point clouds. Most reconstruction methods adopt modeling methods similar to LiDAR. Mainly, the roof point clouds are projected onto the ground and then smoothed and filled using morphological filtering, and then edge extraction methods (such as α-shape) are used to extract the preliminary contour. The edges are regularized according to the contour orientation, and finally, a three-dimensional model is reconstructed by combining height information.

[0004] Existing methods have the following defects: 1. Using the linear fitting method depends on the integrity and accuracy of roof point clouds and ignores the curved detail information of elevations; 2. Traditional roof contour regularization mainly estimates the main direction based on roof edges and does not utilize the orientation information of elevations; 3. Existing reconstruction techniques mainly use roofs for contour extraction and cannot perform reconstruction in the case of missing roof point clouds. Summary of the Invention

[0005] In order to solve the above problems existing in the prior art, the present invention provides a method for extracting a three-dimensional fine model of urban buildings based on the regularization of tomographic SAR point clouds. The technical problems to be solved by the present invention are achieved through the following technical solutions:

[0006] A method for extracting a three-dimensional fine model of urban buildings based on the regularization of tomographic SAR point clouds includes:

[0007] S100, separating building point clouds and ground point clouds from tomographic SAR point clouds;

[0008] S200. Segment the building point cloud to obtain an elevation point cloud and a roof point cloud.

[0009] S300. In the case where the roof point cloud exists, reconstruct the roof contour using the roof point cloud and the elevation point cloud.

[0010] S400. In the case where the roof point cloud is missing, estimate the building thickness using the ground blank area, and reconstruct the roof contour in combination with the elevation footprint.

[0011] S500. Reconstruct the 3D building model using the elevation footprint and the roof contour reconstructed in the presence or absence of the roof point cloud.

[0012] Advantageous effects:

[0013] 1. The present invention proposes an effective solution for segmenting a single elevation point cloud by combining height information, which can effectively avoid the loss of elevation edge information in the method of directly using line fitting to extract the elevation, and at the same time can retain elevations with a small projected area that are not easily detected by the line detection algorithm.

[0014] 2. The present invention adopts a progressive fitting method to gradually refine and segment the elevation footprint from coarse to fine. Compared with the existing method of using line fitting, it can obtain a more complex and practical-shaped elevation footprint.

[0015] 3. In the case where the roof point cloud is not missing, after obtaining the preliminary roof contour by α-shape, the present invention normalizes the preliminary roof contour in combination with the elevation orientation information. By extracting parallel or perpendicular line segments of the elevation orientation, a practical roof contour can be simply and effectively extracted.

[0016] 4. In the case where the roof point cloud is missing, the present invention proposes a method for estimating the roof thickness by combining ground point cloud information. By estimating the ground blank area and combining the elevation height information, the approximate width of the roof can be estimated, thereby generating a 3D model of the building in the case where the roof point cloud is missing.

[0017] The following will further elaborate on the present invention in conjunction with the drawings and embodiments. Description of the Drawings

[0018] Figure 1 is a flowchart of a method for extracting a 3D fine model of urban buildings based on tomographic SAR point cloud regularization provided by the present invention;

[0019] Figure 2 is a process diagram of a method for extracting a 3D fine model of urban buildings based on tomographic SAR point cloud regularization provided by the present invention;

[0020] Figure 3 is a flowchart of building and ground point cloud segmentation provided by the present invention;

[0021] Figure 4 It is a schematic diagram of the point cloud data used in the experiment;

[0022] Figure 5 It is a schematic diagram of the result after SMRF filtering;

[0023] Figure 6 It is a schematic diagram of the result after density clustering;

[0024] Figure 7 It is a schematic diagram of the result of the segmentation of the facade and roof point clouds;

[0025] Figure 8 It is a median distribution diagram of a single building;

[0026] Figure 9 It is a schematic diagram of the result of the preliminary segmentation after threshold segmentation and removal of short segments;

[0027] Figure 10 It is a distribution diagram of the front and back horizontal lines before and after further refined segmentation;

[0028] Figure 11 It is a schematic diagram of the result of the final connection of the short lines in the previous step;

[0029] Figure 12 It is a schematic diagram of the edge contour of the α - shape of the extracted roof point cloud;

[0030] Figure 13 It is a schematic diagram of the final roof contour;

[0031] Figure 14 It is a schematic diagram of the fusion of the reconstructed facade footprint combination and the final roof contour in the previous step;

[0032] Figure 15 It is a schematic diagram of the reconstruction result of a single building;

[0033] Figure 16 It is a contour reconstruction result diagram after the roof thickness estimation in the case of missing roof point cloud;

[0034] Figure 17 It is a schematic diagram of the reconstruction result of the entire scene;

[0035] Figure 18 It is a schematic diagram of the comparison between the model reconstruction and the oblique photography stereo model. Specific implementation manners

[0036] The present invention will be further described in detail below in conjunction with specific embodiments, but the implementation manners of the present invention are not limited thereto.

[0037] Extracting the outlines and heights of buildings from TomoSAR point clouds can generate high-quality 3D models. However, due to the limitations of TomoSAR observation methods and algorithms, there are problems such as incomplete point clouds, multipath interference, and low position accuracy. Traditional facade reconstruction methods based on tomographic SAR point clouds often rely on plane approximation, resulting in the inability to reflect the detailed shape of buildings in the modeling.

[0038] To address this problem, the present invention provides a method for extracting a three-dimensional fine model of urban buildings based on the regularization of tomographic SAR point clouds. By gradually segmenting the facade from coarse to fine along the wall orientation, a fine facade footprint can be reconstructed. Existing methods only utilize roof point clouds during roof reconstruction and do not use the orientation information provided by the facade, resulting in insufficient consistency between the roof and facade shapes. To address this problem, the present invention proposes a method for selecting and connecting representative line segments of the roof contour by combining the facade orientation, which can simply and conveniently extract a regularized roof edge. Existing technologies perform model reconstruction based on the extracted roof contour and will be unable to perform model reconstruction in the case of missing roof point clouds. To address this problem, the present invention provides a method for estimating the thickness of a building roof using the occluded area on the ground, which can achieve model reconstruction in the case of missing roof point clouds.

[0039] Based on computer vision processing methods, tomographic radar imaging theory, etc., the present invention adopts a research idea combining theoretical methods with verification of measured data and proposes a method for extracting a three-dimensional fine model of urban buildings based on the regularization of tomographic SAR point clouds. The present invention mainly consists of two steps: the refined fitting of the facade and the standardized extraction of the roof contour. First, simple morphological filtering combined with a density map method is used to separate building point clouds and ground point clouds from the scene point clouds, and density clustering is used to complete the classification of individual buildings. Secondly, after estimating the main orientation of the building using the facade point clouds, the facade is rotated to be parallel to the x-axis, and a coarse-to-fine segmentation strategy is adopted to refine the segmentation of the facade. The turning points of the facade footprint are found based on the change in the median at different positions of the point clouds. Then, different segments are connected in a horizontal and vertical connection manner to obtain the facade footprint; thirdly, the initial contour of the rotated roof point clouds is obtained using the alpha shape, and only the representative feature line segments parallel or perpendicular to the main direction are retained. The line segments are also connected in a horizontal and vertical connection manner to obtain a regularized contour. Finally, in the case of missing roof point clouds, combined with the height of the facade and the radar incident angle, the ground blank area is used to estimate the building thickness, which is combined with the facade footprint to generate the roof contour. Combining the roof contour and the facade footprint can reconstruct the three-dimensional model of the building.

[0040] Combined Figure 1 and Figure 2 ,the present invention provides a method for extracting a three-dimensional fine model of urban buildings based on the regularization of tomographic SAR point clouds, including:

[0041] S100, Separate building point clouds and ground point clouds from tomographic SAR point clouds;

[0042] S200, Segment the building point clouds to obtain facade point clouds and roof point clouds;

[0043] S300, In the case where roof point clouds exist, reconstruct the roof contour using the roof point clouds and facade point clouds;

[0044] S400, In the case where roof point clouds are missing, estimate the building thickness using the ground blank area, and reconstruct the roof contour in combination with the facade footprints;

[0045] In this step, in the case where roof point clouds are missing, combine the height of the facade and the radar incident angle, and estimate the building thickness through the ground blank area; generate the roof contour in the case of missing roof point clouds in combination with the building thickness and facade footprints.

[0046] S500, Reconstruct the 3D building model using the facade footprints and the roof contour reconstructed in the presence or absence of roof point clouds.

[0047] In a specific embodiment of the present invention, S100 includes:

[0048] S110, Obtain tomographic SAR point clouds, preliminarily filter out the ground point clouds using the SMRF filtering method, and generate ground plane parameters using the ground point clouds;

[0049] S120, Project the remaining non-ground point clouds after S110 onto the ground and divide them into 2m grids, and use the number of point clouds in the grid as the grid density; use half of the maximum grid density as the threshold to segment high-density and low-density grids;

[0050] S130, Combine the point clouds in the low-density grids with the ground point clouds after SMRF filtering, and calculate the distance D from each combined point cloud to the ground plane; classify the point clouds with D < 10 as the final non-building point clouds, and classify the point clouds with D > 10 in combination with the high-density grid point clouds as the final building point clouds;

[0051] S140, Use the density clustering algorithm DBSCAN to cluster and segment the final building point clouds into individual building point clouds.

[0052] In a specific embodiment of the present invention, refer to Figure 3 , S200 includes:

[0053] S210, Project individual building point clouds onto the ground and divide them into 2m grids, calculate the number of point clouds in the grid as the grid density, use half of the maximum grid density as the threshold to separate the high-density grid point clouds, and statistically average the height of the highest point in each grid to obtain H1 as the estimated facade height;

[0054] S220. Obtain the building base height H2 according to the ground parameters, and the building height H = H1 - H2;

[0055] S230. Take threshold = H - 3 as the height segmentation boundary to separate the roof point cloud. Re-divide the remaining point cloud into 1m grids, and calculate the maximum height H max and the minimum height H min of each grid point cloud. Points in the grid cloud that satisfy (H max - H min ) > H / 2 && H max > H - 10 are used as the facade point cloud.

[0056] In a specific embodiment of the present invention, S300 includes:

[0057] S310. In the presence of the roof point cloud, estimate the main orientation of the building using the facade point cloud, and divide it into refined segments through a segmentation strategy, and connect different segments to reconstruct the facade footprint;

[0058] S320. Use alpha shape to obtain the preliminary contour of the rotated roof point cloud; retain the representative feature line segments in the preliminary contour, and connect the representative feature line segments to obtain the roof contour in the presence of the roof point cloud.

[0059] In a specific embodiment of the present invention, S310 includes:

[0060] S311. Project the facade point cloud onto the ground, and use RANSAC to fit a straight line as the main facade direction. Subsequently, rotate the facade parallel to the x-axis according to the main direction;

[0061] S312. Adopt a segmentation strategy from coarse to fine to refine the rotated facade, and find the turning points of the facade footprint according to the change in the number of digits of the point cloud at different positions in the segment; connect different segments through horizontal and vertical connection methods to obtain the reconstructed facade footprint.

[0062] In a specific embodiment of the present invention, S312 includes:

[0063] S3121. Divide each 1m interval along the x-axis, calculate the median of the y values of the points in each interval, draw a median distribution graph. If the median jump between adjacent intervals exceeds 1m, record the x coordinate of the midpoint of the interval as the preliminary segmentation point, and remove the preliminary segments with a length less than 2m;

[0064] S3122. Divide each preliminary segment into 0.5m intervals along the x-axis, calculate the median of the y values of the points in each interval. If the median jump between adjacent intervals exceeds 0.5m, record the x coordinate of the midpoint of the interval as the refined segmentation point, and simultaneously constrain the coordinate difference of the midpoint x of the interval not to exceed 3m;

[0065] In this step, each preliminary segmentation is refined to obtain a refined segmented point cloud.

[0066] S3123. Calculate the median of the y-values of each refined segmented point cloud, and generate a short horizontal line using this median; compare the differences in the y-values of adjacent horizontal lines. If the difference is less than 0.5, take the average of the two horizontal lines as the y-values of the two horizontal lines for constraint.

[0067] S3124. Horizontally extend the previous horizontal line to the middle of the x-axis of the adjacent horizontal line, then vertically extend it to the y-value of the next horizontal line, and then horizontally extend it to the start of the next horizontal line to obtain a connected elevation footprint.

[0068] S3125. Rotate the connected elevation footprint counterclockwise with the point cloud to the original coordinates to obtain the reconstructed elevation footprint.

[0069] In a specific embodiment of the present invention, S320 includes:

[0070] S321. Rotate the roof point cloud to be parallel to the x-axis, then project it onto the ground and divide it into grids to form a binary image.

[0071] S322. Use morphological operations to smooth and fill the binary image, take the largest connected region for α-shape fitting to obtain a preliminary roof contour.

[0072] S323. Take the main direction of the building and its perpendicular direction as regularization constraints, take the short line segments parallel to the two directions in the α-shape, and obtain the roof contour in the presence of the roof point cloud by horizontally or vertically connecting the short point segments.

[0073] It should be noted that the initially obtained polygon is irregular and needs to be refined / regularized. Since the main direction of the building is determined, in this step, the main direction and its perpendicular direction are used as regularization constraints. Only take the short line segments parallel to the two directions in the α-shape, and then, similar to the connection method in the elevation reconstruction, perform horizontal or vertical connection to obtain the final roof contour.

[0074] In a specific embodiment of the present invention, S400 includes:

[0075] S410. Calculate the remaining area between the projected coverage area of the roof point cloud and the projected coverage area of the elevation point cloud. If the distribution of the point cloud in the remaining area in the main direction is less than half of the elevation length, and the distribution length perpendicular to the main direction is less than 5 m, it is considered that the roof is missing and the roof thickness parameter needs to be estimated.

[0076] S420. Obtain the target point cloud less than 5 m away from the ground plane according to the ground plane parameters, and project the target point cloud onto the ground plane.

[0077] Assume that the radar incident angle θ and the squint angle β are known, and calculate the vertical projection length L1 by combining the elevation height and the incident direction.

[0078] S430, project the elevation point cloud onto the ground plane, and use RANSAC for line fitting. Take the length of 5m on both sides of the midpoint of the fitted line and the width of 0.5m to obtain a rectangular strip, and extend the rectangular strip in the direction perpendicular to the line to obtain the actual occlusion length on the ground; among them, if the number of points in the rectangular strip exceeds the threshold, stop the extension, and record the extension length L2 as the actual occlusion width on the ground;

[0079] S440, take L = L2 - L1 as the finally estimated roof width;

[0080] S450, move both ends of the reconstructed elevation footprint along the vertical direction of the elevation by L to generate two new boundary points and connect them to obtain the roof contour without the roof point cloud.

[0081] In a specific embodiment of the present invention, S500 includes:

[0082] S510, fuse the roof contour and its ground projection, and find the point closest to the end point of the ground projection on the roof contour boundary point as the segmentation point;

[0083] S520, use the segmentation point to segment the roof contour, and compare the length differences between the two segments of the contour and the elevation footprint after segmentation;

[0084] S530, replace the contour with the smaller length difference with the elevation footprint to obtain the final building contour, and build a building 3D model in combination with the building height.

[0085] The effectiveness of the present invention can be further illustrated by the following measured data processing results.

[0086] Experimental process and result analysis of measured data:

[0087] Verification of measured data:

[0088] The experimental data comes from the SARMV3D dataset published in the Journal of Radar. The airborne TomoSAR point cloud is generated by tomographic synthetic aperture radar imaging of some urban areas in Yuncheng City, Shanxi Province, China, with a size of approximately 183m × 227m. The point cloud contains 1,740,399 points, and the pixel size is 0.1499m (range) × 0.07m (azimuth). Figure 4 The point cloud data used in the experiment is shown, and the measured data is processed using the method of the present invention. Figure 5 It is the result after SMRF filtering. Figure 5On the left of the middle figure is the preliminary ground point cloud after filtering, and on the right is the preliminary ground point cloud after filtering. It can be seen that SMRF can better separate multipath false points and ground points, but it will also classify the roofs of low-rise buildings as ground points, so further processing is needed to extract the complete building point cloud. Figure 6 This is the result after density clustering, and it can be verified that different building point clouds can be separated through density clustering. Figure 7 This is the segmentation result of the facade and roof point clouds. It can be verified that by combining density and height information, the roof point clouds and facade point clouds can be segmented more completely, while retaining the edge information of the facade. Figure 8 This is the median distribution graph of a single building. From Figure 8 it can be seen that the median fluctuates accordingly with the concave and convex shape of the facade, and the fluctuation is obvious. Figure 9 This is the preliminary segmentation result after threshold segmentation and removal of short segments. It can be seen that different concave and convex surfaces can be roughly segmented. Figure 10 This is the distribution graph of the front and back horizontal lines before and after further refined segmentation. It can be seen from the figure that the concave and convex details of the facade can be better depicted with short lines. Figure 11 This is the final connection result of the short lines in the previous step. By horizontally and vertically connecting the facade, a complete facade footprint contour can be obtained, and it can be seen from the projection of the facade that it fits well with the ups and downs and turns of the facade. Figure 12 This is the α-shape edge contour of the extracted roof point cloud. The alpha shape can only determine the rough contour of a single building. Therefore, the obtained polygon is irregular and needs to be refined / regularized. Figure 13 This is the result of roof contour regularization. Figure 13 In the left figure in the middle, only short line segments parallel or perpendicular to the main direction are extracted, and in the right figure, the regularized contour formed by connecting the short line segments in sequence. Figure 14 This is the result of fusing the reconstruction of the facade footprint and the roof contour in the previous step. By finding the nodes in the roof contour that are closest to the endpoints of the facade footprint, the roof contour is segmented and the part to be replaced is selected to obtain the final building contour. Figure 15 This is the reconstruction result of a single building. Combining the estimated height of the facade and the obtained building contour. Figure 16 This is the contour reconstruction result after roof thickness estimation in the case of missing roof point cloud. The estimated value of the roof thickness of this building is 16.5, and the actual thickness measured using the oblique photography stereo model is about 16. This shows the feasibility of the estimation method. Figure 17 This is the reconstruction result of the entire scene. Figure 18 This is a comparison diagram between the model reconstruction and the oblique photography stereo model. It can be seen that the facade part of the reconstructed model is basically close to the actual building, and the building with missing roof point cloud in the original point cloud can also be effectively reconstructed.

[0089] From the experimental verification with the measured data provided by the present invention, the correctness, effectiveness and reliability of the present invention are verified.

[0090] It should be noted that the terms "first" and "second" in the present invention are only used for descriptive purposes and cannot be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the present invention, "a plurality of" means two or more, unless otherwise specifically defined.

[0091] Although the present application has been described in conjunction with various embodiments herein, however, in the process of implementing the claimed present application, those skilled in the art can understand and achieve other variations of the disclosed embodiments by viewing the accompanying drawings, the disclosure, and the appended claims. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "one" does not exclude a plurality of cases.

[0092] The above content is a further detailed description of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is only limited to these descriptions. For those of ordinary skill in the technical field to which the present invention pertains, without departing from the concept of the present invention, several simple deductions or substitutions can be made, and all should be regarded as belonging to the protection scope of the present invention.

Claims

1. A method for extracting a three-dimensional fine model of urban buildings based on the regularization of tomographic SAR point clouds, characterized in that, Including: S100, separating building point clouds and ground point clouds from tomographic SAR point clouds; S200, segmenting the building point clouds to obtain facade point clouds and roof point clouds; S300, when roof point clouds exist, reconstructing the roof contour using the roof point clouds and facade point clouds; S400, when roof point clouds are missing, estimating the building thickness using the ground blank area and reconstructing the roof contour in combination with the facade footprint; S500, reconstructing the 3D building model using the facade footprint and the roof contour reconstructed with or without roof point clouds.

2. The method for extracting a three-dimensional fine model of urban buildings based on the regularization of tomographic SAR point clouds according to claim 1, wherein S100 includes: S110, obtaining tomographic SAR point clouds, initially filtering out the ground point clouds using the SMRF filtering method, and generating ground plane parameters using the ground point clouds; S120, projecting the remaining non-ground point clouds after S110 onto the ground and dividing them into 2m grids, and taking the number of point clouds in the grid as the grid density; using half of the maximum grid density as the threshold to segment high-density and low-density grids; S130, merging the point clouds in the low-density grids with the ground point clouds after SMRF filtering, and calculating the distance D from each merged point cloud to the ground plane; classifying the point clouds with D < 10 as the final non-building point clouds, and classifying the point clouds with D > 10 combined with the high-density grid point clouds as the final building point clouds; S140, using the density clustering algorithm DBSCAN to cluster and segment the final building point clouds into individual building point clouds.

3. The method for extracting a three-dimensional fine model of urban buildings based on the regularization of tomographic SAR point clouds according to claim 1, characterized in that S200 includes: S210, projecting individual building point clouds onto the ground and dividing them into 2m grids, calculating the number of point clouds in the grid as the grid density, using half of the maximum grid density as the threshold to separate the high-density grid point clouds, and statistically averaging the height of the highest point in each grid to obtain H1 as the facade estimated height; S220, obtaining the building base height H2 according to the ground parameters, and the building height H = H1 - H2; S230. Take threshold = H - 3 as the height segmentation boundary to separate the roof point cloud. For the remaining point cloud, re-divide it into 1m grids and calculate the maximum height H of each grid point cloud max and the minimum height H min , and the grid point cloud that satisfies (H max - H min ) > H / 2 && H max > H - 10 is used as the facade point cloud 4. The method for extracting a three-dimensional fine model of urban buildings based on the regularization of tomographic SAR point clouds according to claim 1, characterized in that, S300 includes: S310, when roof point clouds exist, estimating the main orientation of the building using the facade point clouds, and dividing into refined segments through a segmentation strategy, and connecting different segments to reconstruct the facade footprint; S320, obtaining the preliminary contour of the rotated roof point clouds using alpha shape; retaining the representative feature line segments in the preliminary contour, and connecting the representative feature line segments to obtain the roof contour when roof point clouds exist.

5. The method for extracting a three-dimensional fine model of urban buildings based on the regularization of tomographic SAR point clouds according to claim 4, wherein S310 includes: S311, projecting the facade point clouds onto the ground, and using RANSAC to fit a straight line as the main facade direction, and then rotating the facade to be parallel to the x-axis according to the main direction; S312, adopting a segmentation strategy from coarse to fine to refine the rotated facade, and finding the facade footprint turning points according to the digit change of the point clouds at different positions in the segments; connecting different segments through horizontal and vertical connection methods to obtain the reconstructed facade footprint.

6. The method for extracting a three-dimensional fine model of urban buildings based on the regularization of tomographic SAR point clouds according to claim 5, characterized in that S312 includes: S3121, dividing each 1m interval along the x-axis, calculating the median of the y values of the points in each interval, drawing the median distribution graph, if the median jump between adjacent intervals exceeds 1m, recording the x coordinate of the midpoint of the interval as the preliminary segmentation point, and removing the preliminary segments with a length less than 2m; S3122. Divide each preliminary segment into intervals along the x-axis at intervals of 0.5 m, calculate the median of the y-values of the points in each interval. If the jump of the medians between adjacent intervals exceeds 0.5 m, record the x-coordinate of the midpoint of the interval as the refinement segmentation point, and simultaneously constrain that the coordinate difference of the midpoint x of the interval cannot exceed 3 m. S3123. Calculate the median of the y-values of the point cloud of each refinement segmentation point, and generate a short horizontal line using this median. Compare the differences in the y-values of adjacent horizontal lines. If the difference is less than 0.5, take the average of the two horizontal lines as the y-value of the two horizontal lines for constraint. S3124. Horizontally extend the previous horizontal line to the middle of the x-axis of the adjacent horizontal line, then vertically extend it to the y-value of the next horizontal line, and then horizontally extend it to the starting point of the next horizontal line to obtain the connected elevation footprint. S3125. Rotate the connected elevation footprint and the point cloud counterclockwise to the original coordinates to obtain the reconstructed elevation footprint.

7. The method for extracting the three-dimensional fine model of urban buildings based on the regularization of tomographic SAR point clouds according to claim 4, wherein S320 includes: S321. Rotate the roof point cloud to be parallel to the x-axis, project it onto the ground, and divide it into grids to form a binary image. S322. Use morphological operations to smooth and fill the binary image, take the largest connected region for α-shape fitting to obtain the preliminary roof contour. S323. Take the main direction of the building and its perpendicular direction as regularization constraints, take the short line segments parallel to the two directions in the α-shape, and obtain the roof contour in the presence of the roof point cloud by horizontally or vertically connecting the short point segments.

8. The method for extracting a three-dimensional fine model of urban buildings based on the regularization of tomographic SAR point clouds according to claim 1, characterized in that S400 includes: In the case of missing roof point cloud, combine the height of the elevation and the radar incident angle, and estimate the building thickness through the ground blank area; combine the building thickness and the elevation footprint to generate the roof contour in the case of missing roof point cloud.

9. The method for extracting a three-dimensional fine model of urban buildings based on the regularization of tomographic SAR point clouds according to claim 8, wherein In the case of missing roof point cloud, combine the height of the elevation and the radar incident angle, and estimate the building thickness through the ground blank area; Combining the building thickness and the elevation footprint to generate the roof contour in the case of missing roof point cloud includes: S410. Calculate the remaining area between the projected coverage area of the roof point cloud and the projected coverage area of the elevation point cloud. If the distribution of the point cloud in the remaining area in the main direction is less than half of the elevation length, and the distribution length perpendicular to the main direction is less than 5 m, it is considered that the roof is missing and the roof thickness parameter needs to be estimated. S420. Obtain the target point cloud less than 5 m from the ground plane according to the ground plane parameters, and project the target point cloud onto the ground plane. S430. Project the elevation point cloud onto the ground plane, use RANSAC for line fitting, take a rectangle strip with a length of 5 m on both sides of the midpoint of the fitted line and a width of 0.5 m, and extend the rectangle strip in the direction perpendicular to the line to obtain the actual ground occlusion length; where, if the number of points in the rectangle strip exceeds the threshold, stop the extension, and record the extension length L2 as the actual ground occlusion width. S440. Take L = L2 - L1 as the finally estimated roof width. S450. Move the two ends of the reconstructed elevation footprint along the vertical direction of the elevation by L to generate two new boundary points and connect them to obtain the roof contour in the case of missing roof point cloud.

10. The method for extracting a three-dimensional fine model of urban buildings based on the regularization of tomographic SAR point clouds according to claim 1, characterized in that, S500 includes: S510, fuse the roof contour and its ground projection, and find the point on the boundary point of the roof contour that is closest to the end point of the ground projection as the segmentation point; S520, use the segmentation point to segment the roof contour, and compare the length differences between the two segments of the contour and the elevation footprint after segmentation; S530, replace the contour with the smaller length difference with the elevation footprint to obtain the final building contour, and build a 3D building model in combination with the building height.

Images