3D model generation device and 3D model generation program

Abstract

To accurately reproduce a complicated roof structure without previously setting a roof type.SOLUTION: A three-dimensional model generation device acquires three-dimensional point group data, and two-dimensional building contour line data. The three-dimensional model generation device includes: three-dimensional point group data acquisition means 101; roof surface estimation means 106 for overlapping the three-dimensional point group data and the two-dimensional building contour line data on the basis of positional information, dividing inside a building contour line indicated by the two-dimensional building contour line data into grid meshes having predetermined sizes, grouping the divided grid meshes on the basis of the distribution tendency of the three-dimensional point group data, and thereby estimating a roof surface; correction means 107 for correcting the grid meshes on the basis of information on other grid mesh adjacent to the grid mesh, and thereby correcting the roof surface; and three-dimensional model generation means 120 for generating a three-dimensional model on the basis of the roof surface corrected by the correction means 107.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to a three-dimensional model generation apparatus and a three-dimensional model generation program that generate a three-dimensional model based on point cloud data.

Background Art

[0002] In order to generate a three-dimensional model, a method using laser scanner data is generally often used.

[0003] Laser scanner data is obtained by irradiating the ground surface with laser light from a laser scanner mounted on an aircraft such as an airplane or a helicopter and analyzing these received light data. Then, a three-dimensional model such as a three-dimensional building shape is generated using this three-dimensional point cloud data.

[0004] In the reconstruction of a three-dimensional model such as this three-dimensional building shape, in recent years, a method of automatically generating at the model detail level LOD (Level of Detail) 2 level has been eagerly desired. At the LOD2 level, mainly information on the building outline, the surfaces, lines, and vertices that make up the roof is required.

[0005] Conventionally, many methods have been proposed for finding a two-dimensional building outline from aerial photographs or DSM (Digital Surface Model) data. Among them, as a method for extracting a roof surface and a roof contour line, a method limited to a predetermined roof shape type or a combined roof shape thereof (for example, see Patent Document 1), or a method targeting basic roof shapes with a small number of roof surfaces such as gable roofs and hip roofs (for example, see Patent Document 2) is well known.

Prior Art Documents

Patent Documents

[0006]

Patent Document 1

Patent Document 2

[0007] However, the technologies described in Patent Documents 1 and 2 made it difficult to accurately reproduce buildings with a large number of roof surfaces or buildings with complex shapes that have drop boundary lines.

[0008] For example, in the technology described in Patent Document 1, it is necessary to pre-set the roof shape type, making it difficult to accurately reproduce buildings with a large number of roof surfaces for which no specific roof shape type exists.

[0009] Furthermore, since the technology described in Patent Document 2 targets basic roof shapes with a small number of roof surfaces, it is difficult to accurately reproduce complex roof structures, such as roofs with a multi-layered structure.

[0010] This invention has been made in view of the above-mentioned problems, and aims to provide a 3D model generation device and a 3D model generation program that can accurately reproduce buildings with complex roof shapes. [Means for solving the problem]

[0011] To solve the above objective, the first feature of the 3D model generation apparatus according to the present invention is: A 3D model generation device that generates a 3D model based on 3D point cloud data and 2D building outline data, An acquisition means for acquiring the aforementioned 3D point cloud data and the aforementioned 2D building outline data, A roof surface estimation means that estimates the roof surface by superimposing the three-dimensional point cloud data and the two-dimensional building outline data based on positional information, dividing the area within the building outline indicated by the two-dimensional building outline data into a grid mesh of a predetermined size, and grouping the divided grid meshes based on the distribution trend of the three-dimensional point cloud data, Correction means for correcting the roof surface by correcting the grid mesh based on information from other grid meshes adjacent to the aforementioned grid mesh, A 3D model generation means that generates a 3D model based on the roof surface corrected by the correction means, It is equipped with this.

[0012] The second feature of the 3D model generation device according to the present invention is, The roof surface estimation means is RAN S The method involves using AC to estimate the roof surface by assigning a roof surface number to the 3D point cloud data, and then determining the roof surface number with the highest number of points in the divided grid mesh as the roof surface number of the grid mesh.

[0013] The third feature of the 3D model generation device according to the present invention is, The correction means is the grid mesh of The objective is to correct the roof surface by selecting an arbitrary grid mesh that does not contain the aforementioned 3D point cloud data as the grid mesh of interest, and then using the roof surface number with the highest number among the eight grid meshes adjacent to the grid mesh of interest as the roof surface number of the grid mesh of interest.

[0014] The fourth feature of the 3D model generation device according to the present invention is, The correction means is the grid mesh of When an arbitrary grid mesh containing the aforementioned 3D point cloud data is designated as the grid mesh of interest, if the grid mesh of interest does not belong to any of the four grid meshes adjacent to it on all sides (up, down, left, and right) that form a group of roof surfaces, the roof surface is corrected by setting it to "unassigned to a group".

[0015] The fifth feature of the 3D model generation apparatus according to the present invention is: The three-dimensional model generation means uses the position where the contour line of the roof surface corrected by the correction means and the building outer shape line indicated by the two-dimensional building outer shape line data overlap on a plane as the building outer wall surface, and among the positions where the contour lines of the roof surface corrected by the correction means overlap each other, the position with a drop of a predetermined drop threshold or more is used as the drop wall surface, and wall surface generation means for generating the building outer wall surface and the drop wall surface as a wall surface model; Roof surface generation means for generating a roof surface model in three dimensions based on the roof surface corrected by the correction means; Integration means for integrating the wall surface model generated by the wall surface generation means and the roof surface model generated by the roof surface generation means as a building model; It is further provided with the above.

[0016] The sixth feature of the three-dimensional model generation device according to the present invention is that A first group is generated by grouping the vertices of the model surface of the building model integrated by the integration means based on the position coordinates on the horizontal plane, and a second group is generated by further grouping the first group with the position coordinates in the vertical direction. When there are a plurality of vertices in the generated second group, vertex correction means for integrating them into any one vertex; It is provided with the above.

[0017] The first feature of the three-dimensional model generation program according to the present invention is that A three-dimensional model generation program executed by a three-dimensional model generation device that generates a three-dimensional model based on three-dimensional point cloud data and two-dimensional building outer shape line data, An acquisition step of acquiring the three-dimensional point cloud data and the two-dimensional building outer shape line data; The three-dimensional point cloud data and the two-dimensional building outer shape line data are superimposed based on position information, the building outer shape line indicated by the two-dimensional building outer shape line data is divided into grid meshes of a predetermined size, and the divided grid meshes are grouped based on the distribution tendency of the three-dimensional point cloud data to estimate the roof surface. Roof surface estimation step; A correction step of correcting the roof surface by correcting the lattice mesh based on information of other lattice meshes adjacent to the lattice mesh; And a three-dimensional model generation step of generating a three-dimensional model based on the roof surface corrected by the correction step.

Effect of the Invention

[0018] According to the three-dimensional model generation device and the three-dimensional model generation program according to the present invention, a building having a complicated roof shape can be accurately reproduced as a three-dimensional model.

Brief Description of the Drawings

[0019] [Figure 1] It is a schematic configuration diagram showing the schematic configuration of a three-dimensional model generation device which is an embodiment of the present invention. [Figure 2] An example of color-separated three-dimensional point cloud data is shown by the roof surface estimation means provided in the three-dimensional model generation device which is an embodiment of the present invention. (a) is a plan view showing an example of a roof surface extracted by RANSAC, (b) is a side view showing an example of a roof surface extracted by RANSAC, (c) is a plan view showing an example of a roof surface obtained by merging surfaces considered to be the same plane, and (d) is a side view showing an example of a roof surface obtained by merging surfaces considered to be the same plane. [Figure 3] (a) is a view showing an example of a roof surface where an enclave has occurred, and (b) shows an example of dividing the enclave. [Figure 4] It is a view showing an example of excluding a wall surface. [Figure 5] It is a view showing the result of noise removal by the roof surface estimation means of the three-dimensional model generation device which is an embodiment of the present invention. (a) shows an example of three-dimensional point cloud data after noise removal, and (b) shows an example of three-dimensional point cloud data removed as noise. [Figure 6] It is an explanatory view explaining the lattice mesh division process by the roof surface estimation means of the three-dimensional model generation device which is an embodiment of the present invention. [Figure 7] This is an explanatory diagram illustrating a small-area mesh correction process by a correction means for a 3D model generation apparatus, which is one embodiment of the present invention. (a) shows the roof surface before the small-area mesh correction process, and (b) shows the roof surface after the small-area mesh correction process. [Figure 8] This is an explanatory diagram illustrating the missing mesh completion process by a correction means for a 3D model generation device, which is one embodiment of the present invention. [Figure 9] This is an explanatory diagram illustrating the isolated mesh recompletion process by a correction means for a 3D model generation device, which is one embodiment of the present invention. [Figure 10] This is an explanatory diagram illustrating the integration process by a correction means for a 3D model generation apparatus, which is one embodiment of the present invention. (a) is a diagram showing an example of a roof surface before integration processing, and (b) is a diagram showing an example of a roof surface after integration processing. [Figure 11] This is an explanatory diagram illustrating the straightening process by a correction means for a 3D model generation apparatus, which is one embodiment of the present invention. (a) is a diagram showing an example of a roof surface before the straightening process, and (b) is a diagram showing an example of a roof surface after the straightening process. [Figure 12] This is an explanatory diagram illustrating the ground search process performed by the ground search means of a 3D model generation device, which is an embodiment of the present invention. [Figure 13] This is an explanatory diagram illustrating the roof surface generation process by the roof surface generation means of a 3D model generation apparatus, which is one embodiment of the present invention. [Figure 14] This is an explanatory diagram illustrating the roof surface generation process by the roof surface generation means of a 3D model generation apparatus, which is one embodiment of the present invention. [Figure 15] This is an explanatory diagram illustrating the wall generation process by the wall generation means of a 3D model generation apparatus, which is one embodiment of the present invention. [Figure 16] This is an explanatory diagram illustrating the vertex correction process performed by the vertex correction means of a 3D model generation device, which is an embodiment of the present invention. [Figure 17] This is an explanatory diagram illustrating the vertex correction process performed by the vertex correction means of a 3D model generation device, which is an embodiment of the present invention. [Figure 18] This is a flowchart showing the processing procedure in a 3D model generation device, which is one embodiment of the present invention. [Modes for carrying out the invention]

[0020] Embodiments of the present invention will be described below with reference to the drawings. Throughout the drawings, identical or equivalent parts and components are denoted by the same or equivalent reference numerals. However, it should be noted that the drawings are schematic and may differ from reality. Furthermore, there are parts where the dimensional relationships and proportions differ between drawings.

[0021] Furthermore, the embodiments shown below are illustrative examples of devices and the like for realizing the technical concept of this invention, and the technical concept of this invention is not limited to the arrangement of each component as described below. The technical concept of this invention can be modified in various ways within the scope of the claims.

[0022] The following describes a 3D model generation device, which is one embodiment of the present invention.

[0023] Figure 1 is a schematic diagram showing the general configuration of a 3D model generation device, which is one embodiment of the present invention.

[0024] As shown in Figure 1, the 3D model generation device 1 includes a 3D point cloud data acquisition means 101, a cluster division means 102, a 3D point cloud data storage means 103, a 2D building outline data acquisition means 104, a 2D building outline data storage means 105, a roof surface estimation means 106, a correction means 107, a roof surface data storage means 108, a ground search means 111, a ground data storage means 112, a vertex correction means 115, a 3D model data storage means 116, and a 3D model generation means 120.

[0025] The 3D point cloud data acquisition means 101 acquires 3D point cloud data. 3D point cloud data is measurement data that combines coordinate data calculated based on reflected light from a laser scanner or the like mounted on an aircraft or drone, and the time information of the received light. While the 3D point cloud data here refers to measurement data from an aircraft or drone as described above, it is not limited to measurement data acquired by irradiation from an aircraft or drone; any data containing at least 3D coordinate data is acceptable.

[0026] In this scenario, the aircraft may fly over the same location multiple times to acquire multiple 3D point cloud data points for the same location. In this case, differences in factors such as the laser irradiation angle may result in some data being lost or misaligned.

[0027] Therefore, in order to use more reliable 3D point cloud data, it is necessary to divide the 3D point cloud data by time and make it easier to handle.

[0028] Therefore, the cluster division means 102 divides the 3D point cloud data acquired by the 3D point cloud data acquisition means 101 into clusters for each time period. The cluster division means 102 then gives higher priority to the clusters with a larger number of data points among the divided clusters. The cluster division means 102 then selects the clusters with the highest priority rankings for the building outline, which will be described later, in order until the point cloud coverage rate exceeds 80%.

[0029] The 3D point cloud data storage means 103 is composed of a storage medium such as an HDD (Hard Disk Drive) or an SSD (Solid State Drive), and stores the 3D point cloud data included in the clusters adopted by the cluster partitioning means 102.

[0030] The 2D building outline data acquisition means 104 acquires 2D building outline data TaTo obtain the data, the 2D building outline data is vector data in which the outline of a 2D building is drawn, and the outline has 2D coordinate data.

[0031] The 2D building outline data storage means 105 is composed of a storage medium such as an HDD (Hard Disk Drive) or an SSD (Solid State Drive), and the 2D building outline data acquired by the 2D building outline data acquisition means 104 Ta Remember.

[0032] The roof surface estimation means 106 overlays 3D point cloud data and 2D building outline data based on positional information, divides the area within the building outline indicated by the 2D building outline data into a grid mesh of a predetermined size, and estimates the roof surface by grouping the divided grid meshes based on the distribution trend of the 3D point cloud data.

[0033] For example, the roof surface estimation means 106 uses the 3D point cloud data stored in the 3D point cloud data storage means 103 to perform RAN S Using AC (RANDOM SAmple Consensus) for plane estimation, roof surface numbers are assigned to 3D point cloud data to identify roof surfaces. The roof surface number with the highest number of points in each divided grid mesh is then used as the grid mesh's roof surface number to estimate the roof surfaces. Here, RANSAC is a method for learning the parameters of a mathematical model by excluding the influence of outliers from data containing outliers.

[0034] More specifically, the roof surface estimation means 106 performs a planar fitting using RANSAC on the point cloud of the target building (referred to as point cloud A0) to estimate the largest plane (roof surface) and obtain a 3D planar equation. Next, it assigns a roof surface number to the point cloud on the estimated plane and removes the point cloud to which the roof surface number has been assigned from point cloud A0 (this point cloud is referred to as point cloud A1). The roof surface estimation means 106 performs a planar fitting using RANSAC on point cloud A1 to obtain a 3D planar equation of the plane, assign a roof surface number to the point cloud on the plane, and remove the point cloud. By repeating this process, the 3D planar equations of multiple planes and the assignment of roof surface numbers to the point clouds are obtained. The roof surface extraction process is terminated when the remaining point cloud falls below a certain number of points. Through this process, the roof surface estimation means 106 obtains a 3D planar equation for each roof surface and assigns a roof surface number to it. The roof surface estimation means 106 may also simultaneously assign a different color to each roof surface number to make it easier to visualize which roof surface each point belongs to. 3D point cloud data. If there are multiple roof surfaces extracted by RANSAC, any roof surface that is similar to an already acquired roof surface, that is, if the normal vectors of the roof surfaces are similar (within a predetermined angular range) and the distance between surfaces is short, will be merged as if it were identical to an existing surface.

[0035] Note that while RANSAC was used for plane estimation here, any method capable of plane estimation is acceptable, not necessarily limited to RANSAC.

[0036] Figure 2 shows an example of 3D point cloud data that has been assigned roof surface numbers and color-coded by a roof surface estimation means 106 provided in a 3D model generation apparatus 1, which is one embodiment of the present invention. (a) is a plan view showing an example of a roof surface extracted by RANSAC, (b) is a side view showing an example of a roof surface extracted by RANSAC, (c) is a plan view showing an example of a merged roof surface, and (d) is a side view showing an example of a merged roof surface.

[0037] As shown in Figures 2(a) and (b), roof surfaces D101, D102, D103, and D104 were obtained as roof surfaces extracted by RANSAC.

[0038] Since roof surfaces D101 and D102 have similar normal vectors and are close in distance from each other, the roof surface estimation means 106 merges roof surface D101 with roof surface D102, which has more 3D point cloud data, to generate roof surface D105 shown in Figures 2(c) and (d). Similarly, since roof surfaces D103 and D104 have similar normal vectors and are close in distance from each other, the roof surface estimation means 106 merges roof surface D104 with roof surface D103, which has more 3D point cloud data, to generate roof surface D106 shown in Figures 2(c) and (d).

[0039] The roof surface estimation means 106 further performs a detached area division process. Specifically, after acquiring the roof surface using RANSAC, the roof surface estimation means 106 performs clustering based on the distance between points in order to divide the detached roof surface. By grouping neighboring points within a set distance from the point of interest into the same cluster, regions with continuous point clouds become the same roof surface, while discontinuous, detached 3D point cloud data become separate roof surfaces.

[0040] Figure 3(a) shows an example of a roof surface with an isolated section, and Figure 3(b) shows an example of an isolated section being divided.

[0041] As shown in Figure 3(a), roof surfaces D201a and D201b are recognized as the same roof surface D201 because their normal vectors are similar.

[0042] Therefore, the roof surface estimation means 106 divides roof surface D201a and roof surface D201b as different roof surfaces by performing a detached area division process. For example, as shown in Figure 3(b), the roof surface estimation means 106 leaves roof surface D201a as roof surface D201 and separates roof surface D201b from roof surface D202.

[0043] Furthermore, the roof surface estimation means 106 excludes 3D point cloud data belonging to roof surfaces where the angle between the normal vector of the roof surface and the z-axis is close to 90°, for example, within the range of 90±2°, as wall surfaces.

[0044] Figure 4 shows an example where the wall surface is excluded.

[0045] As shown in Figure 4, the 3D point cloud data D302 is a 3D point cloud data in which the angle between the normal vector and the z axis is close to 90°, so the 3D point cloud data D302 is excluded as a wall.

[0046] The roof surface estimation means 106 also performs noise reduction. Specifically, the roof surface estimation means 106 collects the 3D point cloud data of the roof surface after the acquisition of the roof surface by RANSAC and performs PCA (principal component analysis) analysis. PCA analysis is a method of multivariate analysis in which a small number of uncorrelated variables called principal components that best represent the overall variability are synthesized from a large number of correlated variables. Based on the results of the PCA analysis, the roof surface estimation means 106 determines whether the 3D point cloud data is classified as a distribution of PCA analysis result points (linear, surface, or random), removes the 3D point cloud data classified as "random" as noise, and stores it in the 3D point cloud data storage means 103.

[0047] Figure 5 shows the results of noise reduction by the roof surface estimation means 106 of a 3D model generation apparatus 1, which is one embodiment of the present invention. Figure 5(a) shows an example of 3D point cloud data after noise reduction, and Figure 5(b) shows an example of 3D point cloud data that has been removed as noise.

[0048] As shown in Figure 5(b), the 3D point cloud data classified as "random" in the PCA analysis results is removed as noise, resulting in the 3D point cloud data with noise removed, as shown in Figure 5(a).

[0049] The roof surface estimation means 106 overlays the 3D point cloud data stored in the 3D point cloud data storage means 103 and the 2D building outline data stored in the 2D building outline data storage means 105 based on positional information, and divides the area within the building outline indicated by the 2D building outline data into a grid mesh of a predetermined size.

[0050] Figure 6 is an explanatory diagram illustrating the grid mesh division process by the roof surface estimation means 106 of a 3D model generation apparatus 1, which is one embodiment of the present invention.

[0051] As shown in Figure 6(a), the 3D point cloud data D401 and the 2D building outline data D402 are superimposed based on positional information.

[0052] The roof surface estimation means 106 uses the longest side of the building outline shown by the 2D building outline data D402 as the reference side D402a to generate a bounding box D403 that encloses the building outline in a rectangle.

[0053] As shown in Figure 6(b), the roof surface estimation means 106 divides the area within the building outline indicated by the 2D building outline data D402 into a predetermined size rectangular grid mesh based on the generated bounding box D403.

[0054] The roof surface estimation means 106 estimates the roof surface by grouping the divided grid mesh based on the distribution trend of the 3D point cloud data D401. Specifically, the roof surface estimation means 106 uses RAN S Using AC-based plane estimation, roof surface numbers are assigned to the 3D point cloud data D401 to identify roof surfaces. The roof surface is then estimated by selecting the roof surface number with the highest number of points in each divided grid mesh as the roof surface number for that mesh. As a result, only one roof surface is determined for each grid mesh, and identical roof surfaces are grouped together and estimated as a single roof surface.

[0055] As shown in Figure 6(b), for example, the grid mesh is grouped to estimate roof surface D404a, roof surface D404b, etc.

[0056] As shown in Figure 6(b), in the area D404 within the building outline indicated by the 2D building outline data D402 where the 3D point cloud data D401 does not exist, an unset grid mesh will be used.

[0057] The correction means 107 corrects the roof surface by correcting the grid mesh based on information from other grid meshes adjacent to the grid mesh. Specifically, the correction means 107 performs small-area mesh correction processing, missing mesh completion processing, and isolated mesh recompletion processing.

[0058] Figure 7 is an explanatory diagram illustrating the small-area mesh correction processing by the correction means 107 of a 3D model generation apparatus 1, which is one embodiment of the present invention. Figure 7(a) shows the roof surface before the small-area mesh correction processing, and Figure 7(b) shows the roof surface after the small-area mesh correction processing.

[0059] As shown in Figure 7(a), the correction means 107 counts the number of consecutive grid meshes within the same roof surface. In the example shown in Figure 7(a), roof surfaces D501 to D505 are estimated, and each of these roof surfaces D501 to D505 contains consecutive grid meshes. For example, the number of grid meshes on roof surface D503 is 24, the number of grid meshes on roof surface D504 is 17, and the number of grid meshes on roof surface D505 is 9.

[0060] If the number of continuous grid meshes within the roof surface is less than a preset threshold Th1, the correction means 107 sets the grid meshes included in that roof surface as an unset mesh as a small region.

[0061] For example, if the threshold Th1 is set to 30, then as shown in Figure 7(b), the number of grid meshes included in roof surfaces D503 to D505 is less than 30 each. Therefore, roof surfaces D503 to D505 are set as sub-regions D513 to D515, and the grid meshes included in these sub-regions D513 to D515 are set as unset meshes.

[0062] On the other hand, since the number of grid meshes included in roof surfaces D501 to D502 is 30 or more, the grid meshes included in roof surfaces D501 to D502 are considered to belong to roof surfaces D501 to D502.

[0063] Furthermore, the correction means 107 performs missing mesh completion processing. Specifically, the correction means 107 performs grid mesh completion processing. of When an arbitrary grid mesh that does not contain the 3D point cloud data D401 is selected as the grid mesh of interest, the roof surface is corrected by setting the roof surface number of the eight adjacent grid meshes to the grid mesh of interest as the roof surface number of the grid mesh of interest, which has the highest number. The roof surface number is a number used to uniquely identify a roof surface.

[0064] Figure 8 is an explanatory diagram illustrating the missing mesh completion process by the correction means 107 of the 3D model generation apparatus 1, which is one embodiment of the present invention.

[0065] As shown in Figure 8, for example, a grid mesh of When an arbitrary grid mesh D600 that does not contain the 3D point cloud data D401 is chosen as the grid mesh of interest, the grid meshes adjacent to this grid mesh D600 are the eight neighboring grid meshes D601 to D608.

[0066] Of these grid meshes D601 to D608, the roof surface that contains the 3D point cloud data D401 and has the most roof surfaces will be designated as the roof surface corresponding to the grid mesh D600 of interest.

[0067] Here, among the grid meshes D601 to D608, only grid meshes D605 to D608 contain the 3D point cloud data D401. Grid meshes D601 to D604 are unset grid meshes that do not contain the 3D point cloud data D401. Since grid meshes D605 to D608 all represent the same roof surface, the correction means 107 integrates the grid mesh of interest D600, which has the most roof surfaces, into the roof surface to which grid meshes D605 to D608 belong.

[0068] Then, the correction means 107 selects one of the grid meshes D601 to D604 that does not contain the 3D point cloud data D401 adjacent to grid mesh D600 as the grid mesh of interest and similarly performs missing mesh completion processing. In this way, the correction means 107 performs missing mesh completion processing until all unset meshes within the building outline are completed.

[0069] Furthermore, the correction means 107 performs isolated mesh re-interpolation processing. Specifically, the correction means 107 performs grid mesh re-interpolation processing. of When an arbitrary grid mesh containing the 3D point cloud data D401 is selected as the grid mesh of interest, if the grid mesh of interest does not belong to any of the roof surfaces formed by grouping the four grid meshes adjacent to the grid mesh above, below, left, and right, the roof surface is corrected by setting it to "unassigned to a group".

[0070] Figure 9 is an explanatory diagram illustrating the isolated mesh recompletion process by the correction means 107 of the 3D model generation apparatus 1, which is one embodiment of the present invention.

[0071] In the example shown in Figure 9(a), when an arbitrary grid mesh D700 containing the 3D point cloud data D401 is designated as the grid mesh of interest, the grid mesh of interest D700 belongs to the same roof surface as the grid mesh D701, which is the rightmost of the four grid meshes adjacent to the grid mesh of interest D700 in all directions. In this case, the correction means 107 does not correct the grid mesh of interest D700.

[0072] On the other hand, as shown in Figure 8(b), when an arbitrary grid mesh D710 containing the 3D point cloud data D401 is designated as the grid mesh of interest, the grid mesh of interest D710 belongs to a different roof surface than the one to which the grid mesh D711, which is adjacent to the grid mesh D711 on the right, belongs, among the four grid meshes D711 to D714 adjacent to the grid mesh of interest D710 on the top, bottom, left, and right. Furthermore, since the grid meshes D712 to D714 are unassigned meshes, the grid mesh of interest D710 does not belong to any of the roof surfaces formed by grouping the four grid meshes D711 to D714 adjacent to the grid mesh of interest D710 on the top, bottom, left, and right.

[0073] Therefore, as shown in Figure 9(c), the correction means 107 corrects the roof surface by identifying the target grid mesh D710 as an isolated mesh and setting it to an unassigned mesh D720 that does not belong to a group.

[0074] In this way, the correction means 107 changes all isolated meshes to unassigned meshes and then performs the missing mesh completion process again. This allows them to be assigned to the appropriate roof surface.

[0075] Furthermore, the correction means 107 integrates the grid mesh for each roof surface.

[0076] Figure 10 is an explanatory diagram illustrating the integration process by the correction means 107 of a 3D model generation apparatus 1, which is one embodiment of the present invention. (a) is a diagram showing an example of the roof surface before integration processing, and (b) is a diagram showing an example of the roof surface after integration processing.

[0077] As shown in Figure 10(a), the correction means 107 integrates the grid mesh for each roof surface, and cuts out the integrated polygon along the building outline so that it does not extend beyond the building outline, thereby obtaining the roof contour line. In other words, the correction means 107 generates the roof contour line by deleting the portion that extends beyond the building outline line.

[0078] Furthermore, the correction means 107 straightens the roof contour line.

[0079] Figure 11 is an explanatory diagram illustrating the straightening process performed by the correction means 107 of a 3D model generation apparatus 1, which is one embodiment of the present invention. (a) is a diagram showing an example of a roof surface before the straightening process, and (b) is a diagram showing an example of a roof surface after the straightening process.

[0080] As shown in Figure 11(a), the correction means 107 acquires the common line segment D800 of the roof surface boundary line generated by the integrated processing by the correction means 107.

[0081] The correction means 107 then extracts the portion of the common line segment D800 that is longer than or equal to the threshold Th2 and sets it as the non-straightening line segment D802. On the other hand, the correction means 107 also extracts the portion of the common line segment D800 that is shorter than the threshold Th2 and sets it as the line segments to be straightened D801 and D803. In this way, the common line segment D800 is divided into the line segments to be straightened D801 and D802 and the non-straightening line segment D802.

[0082] Then, as shown in Figure 11(b), the correction means 107 straightens the straightening targets D801 and D802 using, for example, Douglas-Peucker, to generate the straightened roof surface boundary line D805.

[0083] The roof surface data storage means 108 stores the roof surface data (2D roof surface contour data and roof surface equations) corrected by the correction means 107.

[0084] The ground search means 111 searches for the height (z coordinate) of the ground on which the building is installed.

[0085] Figure 12 is an explanatory diagram illustrating the ground search process performed by the ground search means 111 of the 3D model generation device 1, which is one embodiment of the present invention.

[0086] As shown in Figure 12, the ground search means 111 generates a bounding box D902 for the building outline D901.

[0087] The ground search means 111 then sets an extended range D903 by extending the bounding box D902 up, down, left, and right by +α (m) (here, the initial value of α is assumed to be 1m), and acquires 3D point cloud data D401 within the set extended range D903. There are no height restrictions when acquiring the 3D point cloud data D401.

[0088] If the ground search means 111 can acquire 3D point cloud data D401 from within the extended range D903, it sets the minimum height point of the acquired 3D point cloud data D401 as the ground height and terminates the process.

[0089] On the other hand, if the 3D point cloud data D401 cannot be obtained from within the extended range D903, the ground search means 111 further expands the search range. Here, the ground search means 111 sets an extended range D904 by expanding the bounding box D902 up, down, left, and right by +α (m) (here, α: 5m), and obtains the 3D point cloud data D401 within the set extended range D904.

[0090] Similarly, if the ground search means 111 is able to obtain 3D point cloud data D401 from within the extended range D904, it sets the minimum height point of the obtained 3D point cloud data D401 as the ground height and terminates the process.

[0091] On the other hand, if the 3D point cloud data D401 cannot be obtained from within the extended range D904, the ground search means 111 further expands the search range. Here, the ground search means 111 sets an extended range D905 by expanding the bounding box D902 above, below, left, and right by +α (m) (here, α: 20m), and obtains the 3D point cloud data D401 within the set extended range D905.

[0092] Similarly, if the ground search means 111 is able to obtain 3D point cloud data D401 from within the extended range D905, it sets the minimum height point of the obtained 3D point cloud data D401 as the ground height and terminates the process.

[0093] On the other hand, if the 3D point cloud data D401 cannot be obtained from within the extended range D905, the ground search means 111 terminates processing by setting the height (z coordinate) to "0".

[0094] In this case, if the 3D point cloud data D401 could not be obtained from within the extended range D905, the ground search means 111 terminated the process by setting the height (z coordinate) to "0". However, the extended range is not limited to three extensions; it can be extended any number of times.

[0095] The ground data storage means 112 stores the acquired ground data (ground height data).

[0096] The 3D model generation means 120 generates a 3D model based on roof surface data and ground data. The 3D model generation means 120 includes a roof surface generation means 121, a wall surface generation means 122, a bottom surface generation means 123, and an integration means 124.

[0097] The roof surface generation means 121 generates a three-dimensional roof surface model based on the roof surface data (two-dimensional roof surface contour data and roof surface equations) stored in the roof surface data storage means 108.

[0098] Figures 13 and 14 are explanatory diagrams illustrating the roof surface generation process by the roof surface generation means 121 of a 3D model generation apparatus 1, which is one embodiment of the present invention. Figure 13(a) is a plan view, and Figure 13(b) is a side view.

[0099] As shown in Figures 13(a) and (b), the roof surface generating means 121 determines whether or not it has a hierarchical structure.

[0100] Specifically, the roof surface generation means 121 performs an inclusion check based on the roof surface data (2D roof surface contour data and roof surface equations) to determine whether a roof surface contour is contained within another roof surface contour. Then, based on the inclusion check result, the roof surface generation means 121 recognizes the hierarchical structure (parent-child relationship) of the roof surfaces.

[0101] In the example shown in Figures 13(a) and (b), the roof surface generation means 121 recognizes that the roof surface contours of roof surface E102 and roof surface E103 are contained within the roof surface contour of roof surface E101. The roof surface generation means 121 also recognizes that the roof surface contour of roof surface E104 is contained within the roof surface contour of roof surface E102. Roof surface E105 does not have a hierarchical structure.

[0102] Furthermore, if there is no hierarchical structure, a two-dimensional roof contour is obtained by projecting a two-dimensional roof contour.

[0103] As shown in Figure 14, if there is no hierarchical structure, the roof surface generation means 121 obtains the contour lines of a three-dimensional roof surface (contour lines connecting vertices E211 to E214) by projecting the two-dimensional roof contour lines (contour lines connecting vertices E201 to E204) stored in the roof surface data storage means 108 onto the plane E205 estimated by RANSAC (the plane indicated by the roof surface equation: ax+by+cz+d=0).

[0104] As a result, the roof surface generation means 121 can generate a roof surface model by generating a three-dimensional roof surface contour line from a two-dimensional roof contour line based on the roof surface data.

[0105] The wall surface generation means 122 generates a three-dimensional wall surface model based on the two-dimensional roof surface contour data stored in the roof surface data storage means 108 and the ground data stored in the ground data storage means 112.

[0106] Figure 15 is an explanatory diagram illustrating the wall surface generation process by the wall surface generation means 122 of a three-dimensional model generation apparatus 1, which is one embodiment of the present invention. Figure 15(a) is a plan view, and Figure 15(b) is a perspective view.

[0107] The wall surface generation means 122 systematically checks the roof contour lines and building outline lines to determine the outermost wall line (location of the building's exterior wall surface) or the drop boundary wall line (location of the drop wall surface). Specifically, as shown in Figures 15(a) and (b), the wall surface generation means 122 defines the outermost wall line E310 (location of the building's exterior wall surface) as the position where the two-dimensional roof contour line E301, indicated by the two-dimensional roof contour line data stored in the roof surface data storage means 108, and the building outline line E302, indicated by the two-dimensional building outline line data D402, overlap on a plane.

[0108] Similarly, the wall surface generation means 122 defines the position where the two-dimensional roof surface contour line E303, indicated by the two-dimensional roof surface contour line data stored in the roof surface data storage means 108, and the building outline line E302, indicated by the two-dimensional building outline line data D402, overlap on a plane as the outermost wall line E311 (position of the building's outer wall surface).

[0109] Furthermore, the wall surface generation means 122 defines the position where two-dimensional roof surface contour lines indicated by the two-dimensional roof surface contour line data stored in the roof surface data storage means 108 overlap, that is, the position where the two-dimensional roof surface contour line E301 and the two-dimensional roof surface contour line E303 overlap and there is a drop greater than a predetermined drop threshold, as the drop boundary wall line (position of the drop wall surface) E312.

[0110] The wall surface generation means 122 then generates a wall surface model based on the height (z-coordinate) calculated by the ground search means 111, the contour line of the three-dimensional roof surface generated by the roof surface generation means 121, the position of the building's exterior wall surface (outermost wall line), and the position of the drop wall surface (drop boundary wall line).

[0111] As a result, as shown in Figure 15(b), the wall surface generation means 122 calculates contour line E310a based on the outermost wall line E310 and the ground height (z coordinate) calculated by the ground search means 111, and calculates contour line E310b based on the outermost wall line E310 and the contour line of the roof surface in three dimensions.

[0112] Furthermore, as shown in Figure 15(b), the wall surface generation means 122 calculates contour line E312a based on the outermost wall line E310 and the contour line of the roof surface in three dimensions, and calculates contour line E312b based on the outermost wall line E311 and the contour line of the roof surface in three dimensions.

[0113] As a result, the wall generation means 122 can generate a wall surface in three dimensions.

[0114] The base generation means 123 generates a three-dimensional base based on the ground data stored in the ground data storage means 112. Specifically, the base generation means 123 generates a three-dimensional base model based on the ground height (z coordinate) calculated by the ground search means 111 and the two-dimensional building outline data D402.

[0115] The integration means 124 integrates the three-dimensional roof surface model generated by the roof surface generation means 121, the three-dimensional wall surface model generated by the wall surface generation means 122, and the three-dimensional base surface model generated by the base surface generation means 123 to generate a three-dimensional building model.

[0116] The vertex correction means 115 generates a first group by grouping the vertices of the model faces of the building model integrated by the integration means 124 based on their position coordinates on the horizontal plane, generates a second group by further grouping the first group by their vertical position coordinates, and if there are multiple vertices in the generated second group, it integrates them into any one vertex.

[0117] Figures 16 and 17 are explanatory diagrams illustrating the vertex correction process performed by the vertex correction means 115 of a 3D model generation apparatus 1, which is one embodiment of the present invention.

[0118] As shown in Figure 16(a), the vertex correction means 115 generates a first group E410 to E460 by grouping the vertices of the model faces of the building model E400, which have been integrated by the integration means 124, on the horizontal plane (xy plane).

[0119] As shown in Figure 16(b), the vertex correction means 115 generates second groups E411 and E412 by further grouping the first group E410 by vertical position coordinates (z coordinates). Similarly, the vertex correction means 115 generates second groups E421 and E422 by further grouping the first group E420 by vertical position coordinates (z coordinates), generates second groups E431 and E432 by further grouping the first group E430 by vertical position coordinates (z coordinates), generates second groups E441 and E442 by further grouping the first group E440 by vertical position coordinates (z coordinates), generates second groups E451 and E452 by further grouping the first group E450 by vertical position coordinates (z coordinates), and generates second groups E461 and E462 by further grouping the first group E460 by vertical position coordinates (z coordinates).

[0120] Then, if there are multiple vertices in the generated second group, the vertex correction means 115 merges them into any single vertex.

[0121] Figure 17(a) shows an example of the second group before correction by integration, and Figure 17(b) shows an example of the second group after correction by integration.

[0122] As shown in Figure 17(a), the second group E421 includes the vertices of P1(x1,y1,z1) and P2(x2,y2,z2). The second group E422 also includes the vertices of P3(x3,y3,z3) and P4(x4,y4,z4).

[0123] Thus, there may be multiple vertices within the second group, which can be considered an error. Also, we assume that z1 is greater than or equal to z2.

[0124] Therefore, as shown in Figure 17(b), the vertex correction means 115 merges P2(x2,y2,z2) with P1(x1,y1,z1) by making the coordinates of P2(x2,y2,z2) the same as the coordinates of P1(x1,y1,z1). This allows for merging with a higher vertex (a vertex with a larger z-coordinate value).

[0125] Furthermore, the vertex correction means 115 determines that P3(x3,y3,z3) and P4(x4,y4,z4) belong to the same first group. Therefore, it integrates the x and y coordinates of P3 and P4 into the x and y coordinates of P1. The vertex correction means 115 also integrates the z coordinate of P3 into the z coordinate of P4.

[0126] This allows us to integrate P4 into P3, obtaining P3(x1,y1,z3) and P4(x1,y1,z3).

[0127] Note that while P2 was merged into P1 here, P1 could also be merged into P2. Similarly, while P4 was merged into P3 here, P3 could also be merged into P4.

[0128] The 3D model data storage means 116 stores the 3D model data whose vertices have been corrected by the vertex correction means 115.

[0129] Figure 18 is a flowchart showing the processing procedure in a 3D model generation apparatus 1, which is one embodiment of the present invention.

[0130] As shown in Figure 18, in step S101, the 3D point cloud data acquisition means 101 acquires 3D point cloud data.

[0131] In step S103, the 2D building outline data acquisition means 104 acquires 2D building outline data.

[0132] In step S105, the cluster division means 102 performs a cluster division process. Specifically, the cluster division means 102 divides the 3D point cloud data acquired by the 3D point cloud data acquisition means 101 into clusters for each time period, and among the divided clusters, it gives higher priority to clusters with a large number of data points, and adopts the clusters with the highest priority rankings for the building outline, up to a point cloud coverage rate of 80%.

[0133] In step S107, the roof surface estimation means 106 estimates the roof surface using RANSAC (RANDOM SAmple Consensus) based on the 3D point cloud data stored in the 3D point cloud data storage means 103.

[0134] In step S109, the correction means 107 performs a small-area mesh correction process based on information from other grid meshes adjacent to the grid mesh. Specifically, the correction means 107 counts the number of consecutive grid meshes within the same roof surface, and if the number of consecutive grid meshes within the roof surface is less than a preset threshold Th1, it sets the grid meshes included in that roof surface as an unset mesh small area.

[0135] In step S111, the correction means 107 performs missing mesh completion processing. Specifically, the correction means 107 corrects the roof surface by integrating the target grid mesh, which is the roof surface that contains the most of the eight grid meshes adjacent to the target grid mesh, when an arbitrary grid mesh among the grid meshes that does not contain the 3D point cloud data D401 is designated as the target grid mesh.

[0136] In step S113, the correction means 107 performs isolated mesh interpolation. When an arbitrary grid mesh containing the 3D point cloud data D401 is selected as the grid mesh of interest, if the grid mesh of interest does not belong to any of the roof surfaces formed by grouping the four grid meshes adjacent to the grid mesh above, below, left, and right of the grid mesh of interest, the roof surface is corrected by setting it to "unassigned to a group".

[0137] In step S115, the ground search means 111 searches the ground on which the building is installed.

[0138] In step S121, the roof surface generation means 121 generates a three-dimensional roof surface model based on the roof surface data (two-dimensional roof surface contour data and roof surface equations) stored in the roof surface data storage means 108.

[0139] In step S123, the wall surface generation means 122 generates a three-dimensional wall surface model based on the two-dimensional roof surface contour data stored in the roof surface data storage means 108 and the ground data stored in the ground data storage means 112.

[0140] In step S125, the base generation means 123 generates a three-dimensional base based on the ground data stored in the ground data storage means 112.

[0141] In step S127, the integration means 124 combines the three-dimensional roof surface model generated by the roof surface generation means 121 with the three-dimensional wall surface model generated by the wall surface generation means 122. Face Mo Dell and the three-dimensional base model generated by the base generation means 123 are integrated to generate a three-dimensional building model.

[0142] In step S129, the vertex correction means 115 corrects the vertices of the three-dimensional building model. Specifically, the vertex correction means 115 generates a first group by grouping the vertices of the model faces of the building model integrated by the integration means 124 based on their position coordinates on the horizontal plane, generates a second group by further grouping the first group by their vertical position coordinates, and if there are multiple vertices in the generated second group, integrates them into any one vertex.

[0143] As described above, the 3D model generation apparatus 1, which is one embodiment of the present invention, acquires 3D point cloud data and 2D building outline data. It comprises a 3D point cloud data acquisition means 101, a roof surface estimation means 106 that superimposes the 3D point cloud data and the 2D building outline data based on positional information, divides the area within the building outline indicated by the 2D building outline data into a grid mesh of a predetermined size, and estimates the roof surface by grouping the divided grid meshes based on the distribution trend of the 3D point cloud data, a correction means 107 that corrects the roof surface by correcting the grid mesh based on information of other grid meshes adjacent to the grid mesh, and a 3D model generation means 120 that generates a 3D model based on the roof surface corrected by the correction means 107.

[0144] This allows for the accurate reproduction of complex roof structures without having to pre-set roof shape types.

[0145] Furthermore, the above-described embodiment can also be realized by running a program installed on a computer. [Explanation of Symbols]

[0146] 1. 3D Model Generation Device 101 3D point cloud data acquisition method 102 Cluster partitioning means 103 3D point cloud data storage means 104 2D Building outline data acquisition method 105 2D Building outline data storage means 106 Roof surface estimation means 107 Correction means 108 Roof surface data storage means 111 Ground search means 112 Ground data storage means 115 Vertex Correction Method 116 3D model data storage means 120 3D Model Generation Method 121 Roof surface generation means 122 Wall Generation Methods 123 Bottom Surface Generation Method 124 Integration methods

Claims

1. A 3D model generation device that generates a 3D model based on 3D point cloud data and 2D building outline data, An acquisition means for acquiring the aforementioned three-dimensional point cloud data and the aforementioned two-dimensional building outline data, A roof surface estimation means that estimates the roof surface by superimposing the three-dimensional point cloud data and the two-dimensional building outline data based on positional information, dividing the area within the building outline indicated by the two-dimensional building outline data into a grid mesh of a predetermined size, and grouping the divided grid meshes based on the distribution trend of the three-dimensional point cloud data, Correction means for correcting the roof surface by correcting the grid mesh based on information from other grid meshes adjacent to the aforementioned grid mesh, The system includes a three-dimensional model generation means that generates a three-dimensional model based on the roof surface corrected by the correction means, The roof surface estimation means is By using RANSAC for plane estimation, a roof surface number is assigned to the 3D point cloud data to identify the roof surface, and the roof surface is estimated by selecting the roof surface number with the highest number of points in the divided grid mesh as the roof surface number of the grid mesh. A three-dimensional model generation apparatus characterized by the following features.

2. The correction means is When an arbitrary grid mesh from the aforementioned grid meshes that does not contain the three-dimensional point cloud data is designated as the grid mesh of interest, the roof surface is corrected by setting the roof surface number of the eight adjacent grid meshes to the grid mesh of interest as the roof surface number of the grid mesh of interest, which has the highest number of roof surface numbers. The three-dimensional model generation apparatus according to claim 1.

3. The correction means is When an arbitrary grid mesh containing the three-dimensional point cloud data is selected as the grid mesh of interest, and the grid mesh of interest does not belong to any of the roof surfaces formed by grouping the four grid meshes adjacent to the grid mesh above, below, left, and right of the grid mesh of interest, the roof surface is corrected by setting it to "unassigned to a group". The three-dimensional model generation apparatus according to claim 1.

4. The three-dimensional model generation means is A wall surface generation means generates a wall surface model by defining the building exterior wall surface at the position where the contour line of the roof surface corrected by the correction means and the building outline shown by the two-dimensional building outline data overlap on a plane, defining the drop wall surface at the position where the drop difference exceeds a predetermined drop threshold among the positions where the contour lines of the roof surfaces corrected by the correction means overlap, and generating the building exterior wall surface and the drop wall surface as a wall surface model. A roof surface generation means that generates a roof surface model in three dimensions based on the roof surface corrected by the correction means, An integration means for integrating the wall model generated by the wall generation means and the roof model generated by the roof generation means into a building model, The 3D model generation apparatus according to claim 1, further comprising the features described above.

5. The vertex correction means generates a first group by grouping the vertices of the model surface of the building model integrated by the integration means based on their position coordinates on the horizontal plane, generates a second group by further grouping each of the first groups based on their vertical position coordinates, and if there are multiple vertices in the generated second group, integrates them into any one vertex. The three-dimensional model generation apparatus according to claim 4, characterized by comprising the following features.

6. A 3D model generation program executed by a 3D model generation device that generates a 3D model based on 3D point cloud data and 2D building outline data, An acquisition step to acquire the aforementioned 3D point cloud data and the aforementioned 2D building outline data, A roof surface estimation step in which the three-dimensional point cloud data and the two-dimensional building outline data are superimposed based on positional information, the area within the building outline indicated by the two-dimensional building outline data is divided into a grid mesh of a predetermined size, and the divided grid mesh is grouped based on the distribution trend of the three-dimensional point cloud data to estimate the roof surface, A correction step of correcting the roof surface by correcting the grid mesh based on information from other grid meshes adjacent to the aforementioned grid mesh, The system includes a 3D model generation step that generates a 3D model based on the roof surface corrected by the correction step, The aforementioned roof surface estimation step is, By using RANSAC for plane estimation, a roof surface number is assigned to the 3D point cloud data to identify the roof surface, and the roof surface is estimated by selecting the roof surface number with the highest number of points in the divided grid mesh as the roof surface number of the grid mesh. A 3D model generation program characterized by having the following features.

Images