JP2018173705A - Three dimensional map separation device and three dimensional map separation program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a three dimensional map separation device and a three dimensional map separation program capable of precisely separating portions expressed by predetermined attributes in three dimensional map information for each attribute.SOLUTION: A three dimensional map separation device separates a plurality of planes constituting three dimensional map information expressed by polygon data for each attribute of each plane. The three dimensional map separation device comprises: a graph structure generating unit which generates graph structure representing adjacent relations between the plurality of planes by using the three dimensional map information; a first weight setting unit which finds a first weight indicative of possibility that the attribute of one plane is different from the attribute of the other plane, by using angles in which adjacent two planes face; a second weight setting unit which finds a second weight indicative of how much the plane has a specific attribute, from a height of a certain plane and the height of the ground in a spot of the plane; and a segmentation unit which performs graph cut processing to the graph structure, and separates each plane for each attribute, by using the first weigh and the second weigh.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a 3D map separation apparatus and a 3D map separation program for separating a part (for example, a building and a non-building) represented by a predetermined attribute in 3D map information for each attribute.

  In recent years, in various applications such as disaster response and security planning, a map that also includes height information created by performing stereo shooting and laser ranging from an aircraft, and a three-dimensional representation of the three-dimensional shape represented by polygon data Map information is used. However, since the three-dimensional map information is not defined as to what attributes each part represented by the polygon has, the application range is limited, and it cannot be said that it is always easy to use. Separating a set of polygons composed of parts represented by predetermined attributes such as a part corresponding to a building, a part corresponding to the ground, a part corresponding to a planting / road structure, etc. in the 3D map information If possible, an attribute label according to the usage can be given to each. Thereby, for example, it is considered that a portion where a person can move becomes clear and can be used for security planning such as an arrangement plan of guards, and the usability is improved.

  In Patent Document 1, by applying a filtering technique to a high-accuracy altitude model (ground position information) obtained by laser scanning from the sky, features are removed from the altitude model and measured. A technique for creating a ground surface elevation model representing a smooth ground surface as a whole in conformity with the altitude of the ground surface is disclosed.

Japanese Patent Laid-Open No. 2002-236019

  However, the above prior art does not take into account the separation of the portion represented by the predetermined attribute in the three-dimensional map information even when creating a ground surface elevation model. Therefore, it cannot be used for assigning an attribute label to a portion (a set of polygons) represented by a predetermined attribute in the three-dimensional map information.

  The present invention has been made in view of such problems, and a three-dimensional map separation device and a three-dimensional map separation program capable of accurately separating a portion represented by a predetermined attribute in three-dimensional map information for each attribute. The purpose is to provide.

  According to one aspect of the present invention, there is provided a three-dimensional map separation device for separating a plurality of surfaces constituting three-dimensional map information represented by polygon data for each surface attribute, The attribute of one surface and the attribute of the other surface are generated using a graph structure generation unit that generates a graph structure indicating the adjacent relationship between the plurality of surfaces using information, and an angle at which the two adjacent surfaces face each other. A first weight setting unit for obtaining a first weight indicating a different possibility, a second weight indicating a certain attribute of the surface from the height of a certain surface and the height of the ground at the point of the surface. A second weight setting unit that obtains a weight; and a segmentation unit that performs graph cut processing on the graph structure using the first weight and the second weight, and separates each surface for each attribute. A three-dimensional map separating apparatus is provided.

  The graph structure generation unit is adjacent to a surface node representing each surface and a first virtual node and a second virtual node that are virtual nodes corresponding to different attributes using the 3D map information. A first edge that connects the surface nodes of the surface; a first edge that connects the first virtual node and the surface node; and a second edge that connects each of the second virtual node and the surface node. The first weight setting unit sets the first weight for each of the first edges, and the second weight setting unit sets the graph structure for each of the second edges. The second weight may be set, and the segmentation unit may be separated into a plane node connected to the first virtual node and a plane node connected to the second virtual node.

  The second weight setting unit obtains a difference between the height of the surface and the height of the ground at the point of the surface, and sets the second weight that the attribute of the surface is more likely to be the ground as the difference is smaller. The second edge connected to the second virtual node is set to the second edge connected to the second virtual node, and the second weight connected to the second virtual node is set to the second weight that the attribute of the surface is likely to be non-ground as the difference is larger. The segmentation unit performs a graph cut process on the graph structure, sets the attribute of the surface represented by the surface node connected to the first virtual node as the ground, and is connected to the second virtual node. The attribute of the surface represented by the surface node may be separated as non-ground.

  A correct label for which the attribute of the surface corresponding to the area inside the reduced outer peripheral line reduced by a predetermined amount of the area formed by the outer peripheral line of the building represented by the two-dimensional map data is a building, and the area formed by the outer peripheral line is determined. A correct label setting unit for obtaining a correct answer label in which the attribute of the surface corresponding to a region outside the enlarged outer peripheral line enlarged by a predetermined amount is a non-building is provided, and the second weight setting unit uses the correct label as the building Then, the second weight that the attribute of the surface is likely to be a building is set to the second edge connected to the first virtual node, and the attribute of the surface is determined to be a non-building by using the correct label as the non-building. The second weight is set to be the second edge connected to the second virtual node, and the segmentation unit performs a graph cut process on the graph structure and connects to the first virtual node. Is an attribute of the surface represented by the surface node and buildings is, the attributes of the surface represented by the surface node connected to the second virtual node may be separated as non-building.

  When the region formed by the enlarged outer peripheral line of a certain outer peripheral line overlaps with the region formed by the enlarged outer peripheral line of another outer peripheral line, the correct label setting unit is configured such that the certain outer peripheral line is included in the overlapping region. The area formed by the second predetermined outer circumference expanded by the second predetermined amount smaller than the predetermined amount and the area formed by the other outer peripheral line is formed by the region outside the second enlarged outer peripheral line expanded by the second predetermined amount. You may set the correct answer label of a non-building.

  The second weight setting unit may obtain the altitude using altitude data indicating the height of the ground stored in advance.

  The second weight setting unit may obtain the height of the ground using two-dimensional map data representing an area corresponding to the ground and the three-dimensional map information.

  According to another aspect of the present invention, there is provided a three-dimensional map separation program for separating a plurality of faces constituting three-dimensional map information represented by polygon data for each face attribute, Using the three-dimensional map information, a graph structure generation unit that generates a graph structure indicating an adjacent relationship between the plurality of surfaces, an angle between two adjacent surfaces, an attribute of one surface, and the other surface The first weight setting unit for obtaining a first weight indicating the possibility that the attribute of the surface is different from the height of a certain surface and the height of the ground at the point of the surface indicates how much the surface seems to be a specific attribute. A second weight setting unit that obtains a second weight to be shown, and a segmentation that performs graph cut processing on the graph structure using the first weight and the second weight, and separates each surface for each attribute Function as part 3D map separation program is provided.

  A region having a predetermined attribute expressed by three-dimensional map information can be accurately separated for each attribute.

The block diagram which shows schematic structure of the three-dimensional map separation apparatus 100 which concerns on one Embodiment. The flowchart which shows the processing operation of the three-dimensional map separation apparatus 100. The figure which shows typically the graph structure produced | generated by the graph structure production | generation part 41. FIG. The figure explaining the smooth term S. The figure explaining the method of setting the correct label area | region of a building. The figure explaining the method of setting the correct label area | region of a non-building. The figure explaining the method of setting the correct label area | region of a non-building. The figure which put together the data term set to each 2nd edge. The figure which illustrates a graph cut typically. The figure which shows the polygon data in the three-dimensional map information 31. FIG. The figure obtained by performing segmentation with respect to the three-dimensional map information 31. FIG.

  Embodiments according to the present invention will be specifically described below with reference to the drawings.

(First embodiment)
FIG. 1 is a block diagram illustrating a schematic configuration of a three-dimensional map separating apparatus 100 according to an embodiment. The three-dimensional map separating apparatus 100 includes an input device 1, an output device 2, a storage unit 3, and a control unit 4. Note that the three-dimensional map separation device 100 may be configured by a single device, or may be configured by a plurality of devices.

  The three-dimensional map separating apparatus 100 separates a plurality of surfaces (polygons) constituting three-dimensional map information expressed by polygon data, which will be described later, for each surface attribute. The attribute is, for example, a building, the ground, planting, a road structure, or the like. In the present embodiment, an example is shown in which each surface is separated into a building and the other (hereinafter referred to as a non-building).

  The input device 1 is a user interface for performing input to the storage unit 3 and the control unit 4 and includes a keyboard, a mouse, a touch panel, and the like. The output device 2 is a user interface for outputting the processing result in the control unit 4, and is a display, a printer, an information storage device to a storage medium, a communication device capable of transmitting information via a network, or the like.

  The storage unit 3 is a hard disk, for example, and stores various information and programs. Specifically, the storage unit 3 stores 3D map information 31, 2D map data 32, altitude data 33, and the like input in advance via the input device 1. In addition, the storage unit 3 temporarily stores graph data 34 generated by the control unit 4 described later and used for subsequent processing.

  Here, the three-dimensional map information 31 is polygon data representing the shape of the ground surface, and may include buildings, planting, road structures (signs, footbridges, etc.) (see FIG. 9A). The three-dimensional map information 31 is created by the following process, for example. First, the ground is photographed from above with a plurality of cameras having different photographing angles. Subsequently, the position in the height direction is calculated by the principle of stereo vision, and a point group having three-dimensional coordinates is obtained. After that, polygon processing and simplification processing are performed on the point group. Then, the texture created from the shooting data is mapped. Thus, 3D map information can be created.

  Note that the 3D map information 31 is simply polygon data with a large texture, and does not include information on what a specific place is. For example, it is not known from the 3D map information 31 which place is a building and which place is a road.

  The two-dimensional map data 32 is data composed of a polygon group representing a building outer peripheral line (in other words, an area corresponding to the ground) representing an existing area of the building. For example, the two-dimensional map data 32 is included in the base map information of the Geographical Survey Institute. The “periphery of the building” can be used. Note that the map information that is not polygonized may be used, and the control unit 4 may generate the two-dimensional map data 32 by polygonizing.

  The altitude data 33 is data indicating the altitude for each point. For example, “numerical altitude data” included in the base map information of the Geographical Survey Institute can be used. This numerical elevation data is data created by aviation laser surveying, and measures altitude by measuring the time from when laser light is emitted to the ground surface until it is reflected and returned. In the case of this digital elevation data, the data is prepared at intervals of about 5 m, and the error is about 0.3 m.

  The control unit 4 includes a graph creation unit 40 and a segmentation unit 45, and the graph creation unit 40 includes a graph structure generation unit 41, a smooth term calculation unit 42, a correct answer label setting unit 43, and a data term calculation unit 44. . Each of these units may be realized, for example, by a processor executing a predetermined program stored in the storage unit 3.

  The graph creation unit 40 creates a graph structure using the polygon data of the 3D map information 31 in order to segment the building portion and other portions in the 3D map information 31. Specifically, it is as follows.

  A graph structure generation unit 41 in the graph creation unit 40 generates a graph structure from the three-dimensional map information 31 expressed by polygon data. The graph structure indicates the adjacency relationship between a plurality of surfaces constituting the 3D map information 31. More specifically, the graph structure includes an edge that connects a node (hereinafter referred to as a surface node) representing each surface (polygon) defined in each polygon data and a surface node representing another surface adjacent to the surface. (Hereinafter referred to as the first edge). Further, the graph structure includes a building node B (first virtual node) that is a virtual node representing a building (Building) and a non-building node O (second virtual node) that is a virtual node representing the other (Other). Node). Further, the graph structure includes an edge connecting the building node B and each of the area nodes and an edge connecting the non-building node O and each of the area nodes (hereinafter referred to as the building node B, the non-building node O and each of the area nodes). Edge to be connected is referred to as a second edge). The generated graph structure is stored in the storage unit 3 as graph data 34. Note that the two virtual nodes correspond to different attributes.

  The smooth term calculation unit 42 (first weight setting unit) refers to the three-dimensional map information 31 and the graph data 34, and uses one of the two facing surfaces as an angle, and one surface is a building and the other surface. Is a weight indicating the possibility of being a non-building (hereinafter referred to as “smooth term S” or “first weight”). A smooth term S can be set for each of the first edges. The smooth term S can be said to indicate how much the first edge seems to be a boundary between a building and a non-building. The smooth term S has a smaller value as the possibility of being a boundary is higher.

  The correct answer label setting unit 43 refers to the two-dimensional map data 32, obtains a correct answer label of a building that is surely a building in a part of the polygon data, and surely determines that the non-building is surely non-building. Find the correct label. In the present embodiment, the correct answer label obtained is set to a plane node. However, the present invention is not limited to this, and it may be set for each surface of the three-dimensional map information 31.

  The data term calculation unit 44 (second weight setting unit) refers to the altitude data 33, and calculates a weight (hereinafter referred to as “data term D”) from the height of a certain surface and the altitude (height) of the ground at the point of this surface. Or “second weight”). A data term D can be set for each of the second edges. Data term D indicates how much the surface represented by the surface node connected to the second edge is likely to be a building or non-building.

  More specifically, the data term calculation unit 44 obtains the difference between the height of a certain surface and the altitude of the ground at the point of this surface. The data term calculation unit 44 assumes that the smaller the difference is, the more likely the surface is to be ground (that is, a non-building), and the greater the difference is, the more likely this surface is to be a building. Regarding the second edge connecting the building node B and the plane node, the data term D has a larger value as the possibility of being a building is higher. On the other hand, regarding the second edge connecting the non-building node O and the plane node, the data term D becomes a larger value as the possibility of being a non-building is higher. The data term calculation unit 44 refers to the correct answer label set by the correct answer label setting unit 43 to further obtain the data term D of each face node. More specifically, the plane of the plane node on which the correct answer label of the building is set is sufficiently large to the second edge connecting the plane node and the building node B because the possibility of being a building is quite high. The value data term D is set. In addition, since the surface of the surface node to which the correct answer label for the non-building is set is highly likely to be a non-building, a sufficiently large value for the second edge connecting the surface node and the non-building node O. Data term D is set.

  The segmentation unit 45 performs graph cut processing on the graph structure including the first edge in which the smooth term S is set and the second edge in which the data term D is set, and each surface is connected to the building node B The node group is separated into a plane node group connected to the non-building node O.

FIG. 2 is a flowchart showing the processing operation of the 3D map separation apparatus 100.
First, the graph structure generation unit 41 generates a graph structure from polygon data in the three-dimensional map information, and stores it as graph data 34 in the storage unit 3 (step S1). Specifically:

  The graph structure generation unit 41 sets one surface node for each surface defined in each polygon data of the 3D map information 31. For example, the graph structure generation unit 41 may use the center of gravity of each surface as a surface node. Subsequently, the graph structure generation unit 41 connects the surface nodes representing adjacent surfaces to form a first edge. Further, the graph structure generation unit 41 connects the virtual building node B representing the building and the virtual non-building node O representing the other and each of all the surface nodes to form the second edge. Thus, a graph structure is generated. The building node B and the non-building node O need not have position information.

  FIG. 3 is a diagram schematically illustrating a graph structure generated by the graph structure generation unit 41. In the figure, one black circle represents one surface node, that is, one surface. A solid line connecting the surface nodes indicates the first edge. Further, a broken line connecting each face node and the building node B and a broken line connecting each face node and the non-building node O indicate the second edge. As will be described later, the smooth term S is set for the first edge, and the data term D is set for the second edge.

Returning to FIG. 2, after the graph structure is generated, the smooth term calculation unit 42 sets a smooth term S for each of the first edges in the graph structure (step S2). The smooth term S is a value indicating the likelihood of a boundary between a building and a non-building. For example, the smaller the difference between the angle φ1 facing one surface connected to each first edge and the angle φ2 facing the other surface is smaller ( The higher the possibility that it is not a boundary, the larger the value.
FIG. 4 is a diagram for explaining the outline of the smooth term S.

  In FIG. 4A, the two surfaces F1 and F2 are directed in substantially the same direction. In this case, it means that there is a high possibility that the two faces F1 and F2 belong to the same building or both belong to the same non-building. In other words, there is a high possibility that the boundary between the surface F1 and the surface F2 is not a boundary between a building and a non-building. Accordingly, the smooth term S set at the first edge connecting the surface node representing the surface F1 and the surface node representing the surface F2 is a large value.

  On the other hand, in FIG.4 (b), two surface F1, F2 has faced the direction from which it differs greatly. In this case, it means that one of the two surfaces F1 and F2 belongs to a certain building, and the other is likely to belong to a non-building. In other words, there is a high possibility that the space between the surface F1 and the surface F2 is a boundary between a building and a non-building. Therefore, the smooth term S set at the first edge connecting the surface node representing the surface F1 and the surface node representing the surface F2 is a small value.

As a specific example of the smooth term S in the present embodiment, a normal vector of a certain surface (for example, surface F1 in FIGS. 4A and 4B) and another surface (for example, FIGS. 4A and 4B). When the angle formed by the normal vector of the surface F2) is φ, the smooth term calculation unit 42 calculates the smooth term S based on the following equation.
S = (1 + cos2 | φ |) / 2
In this case, the smooth term S can take a value between 0 and 1.

  Returning to FIG. 2, the correct label setting unit 43 reliably sets the correct label area of the building in the area that is the building based on the outer peripheral line of the building represented by the two-dimensional map data 32, and is the area that is surely non-building. Set the correct label area for non-buildings. Then, the correct label of the building is set to the plane node at the position corresponding to the correct label area of the building, and the correct label of the non-building is set to the plane node at the position corresponding to the correct label area of the non-building (step S3). ). Note that since the graph structure is not used for setting the correct label area, a process for setting the correct label area may be performed prior to generation of the graph structure.

  Here, the correct answer label setting unit 43 may set the correct answer label area of the building in the area inside the outer periphery of the building in the two-dimensional map data 32 and set the correct answer label area of the non-building in the outer area. However, since the outer periphery of the building represented by the two-dimensional map data 32 is not always accurate, it is more desirable to do the following.

  FIG. 5 is a diagram illustrating a method for setting a correct label area of a building. The correct label setting unit 43 reduces the area formed by the outer perimeter line (solid line in the figure) of each building represented by the two-dimensional map data 32 by a predetermined distance d1 (for example, 2 m) to form a reduced outer perimeter line (broken line in the figure). To do. Then, the correct answer label setting unit 43 sets the correct answer label area of the building in the area inside the reduced outer peripheral line. Thereby, even when the two-dimensional map data 32 includes some errors, it is possible to reliably set the correct label area of the building only in the area that is the building.

  6A and 6B are diagrams illustrating a method of setting a correct label area of a non-building. As shown in FIG. 6A, the correct label setting unit 43 enlarges the area formed by the outer peripheral line (solid line in the figure) of each building represented by the two-dimensional map data 32 by a predetermined distance d2 (for example, 2 m), and expands the outer peripheral line ( A broken line in FIG. And the correct label setting part 43 sets the correct label area | region of a non-building in the area | region which is not contained in the expansion outer peripheral line of any building.

  However, as shown to FIG. 6B (a), the expansion which expanded the area | region which one part of the area | region which extended outer peripheral line A1 which expanded the area | region which the outer periphery line of a certain building B1 makes, and the outer periphery line of another building B2 expanded. A part of the area formed by the outer peripheral line A2 may overlap (the shaded area in FIG. 6B (a)). Even in such a case, in order to separate the two buildings B1 and B2, it is desirable to set a non-building correct label region as an overlapping region (at least a part thereof).

  Therefore, in such a case, the correct label setting unit 43 enlarges the area formed by the outer peripheral lines of the buildings B1 and B2 by k * d2 (0 <k <1, that is, k * d2 <d2), and expands the outer peripheral line A1. ', A2' (one-dot chain line in the figure) is formed. Then, the correct label setting unit 43 sets a non-building correct label area inside the area where the areas formed by the enlarged outer peripheral lines A1 and A2 overlap and outside the enlarged outer peripheral line areas A1 ′ and A2 ′. That's fine.

  If the value of k is too large, a non-building correct label area cannot be set between the buildings B1 and B2. On the other hand, if the value of k is too small, the correct label area of the building may be erroneously set in the non-building area. Therefore, it is desirable to adjust the value of k in accordance with the two-dimensional map data 32 to be used and the density of buildings on the data. As an example, it is preferable to reduce the value of k as the density of a building is larger.

  In addition, when enlarging or reducing the outer periphery of the building, a predetermined reduction rate or enlargement rate may be applied instead of using the distances d1 and d2 as a reference.

  Returning to FIG. 2, the data term calculation unit 44 calculates the data term D based on the correct answer label region for some of the second edges and based on the altitude data 33 for the other second edges (step S <b> 4). ). More specifically, when the horizontal coordinate of a certain face node (or a face corresponding to a certain face node) is included in the correct label area of the building, the correct answer label of the building is set for the face node. When the horizontal coordinate of a certain surface node (or a surface corresponding to a certain surface node) is included in the non-building correct label area, a non-building correct label is set for the surface node. And the data term calculation part 44 is a data term based on a correct label with respect to the 2nd edge which connects the surface node and virtual node (building node B, non-building node O) in which the correct label of the building or the non-building was set. D is calculated. The data term calculation unit 44 calculates a data term D based on the altitude data 33 for the other second edges.

The data term D based on the correct answer label is as follows. The data term calculation unit 44 defines the data term K as the following equation (1).

Here, P represents a set of plane nodes, and E represents a set of first edges connecting the plane nodes to the plane nodes. {P, q} represents a first edge connecting the plane node p and the plane node q. S _{[p, q]} represents the smooth term S set at the first edge connecting the plane node p and the plane node q. This data term K is larger than the sum of the weights of the connected smooth terms for the surface node having the largest sum of the weights of the edges of the connected smooth terms among all the nodes. .

  On the other hand, the data terms based on the altitude data 33 are as follows. The data term calculation unit 44 calculates the height of the surface represented by the surface node connected to the second edge (obtained from the three-dimensional map information 31) and the altitude of the ground at the point of the surface (obtained from the altitude data 33). And a predetermined threshold value δ are compared to set a data term.

More specifically, a point higher than a threshold δ (for example, 0.5 m) with respect to the altitude from the ground obtained from the altitude data 33 is assumed to be a building, and a point that is not so is assumed to be a non-building. More specifically, the value of the z coordinate at a certain coordinate (x, y) in the polygon data of the three-dimensional map information 31 is Z (x, y), and the altitude at the same coordinate (x, y) in the elevation data 33 is set. Is Z ₀ (x, y). At this time, a data term D _B (x, y) representing the uniqueness of the building is determined according to the altitude difference from the ground. That is, the value D _B (x, y) is defined as the following equation (2) using the threshold δ.

On the other hand, a data term D _O (x, y) representing non-building-likeness is defined as the following equation (3).
D _O (x, y) = 1−D (x, y) (3)

The data terms D _B (x, y) and D _O (x, y) based on the altitude data 33 can take 0 or 1. That is, the maximum and minimum values of the data terms D _B (x, y) and D _O (x, y) can be made equal to the maximum and minimum values of the smooth term S.

FIG. 7 is a table summarizing data terms set in each second edge.
Of the second edges {p, B} connecting the arbitrary plane node p and the building node B, when the correct label of the building is set to the plane node p, the value of K in the above equation (1) is The data term D of the second edge is a value sufficiently larger than 1. On the other hand, in the second edge {p, B}, when a non-building correct answer label is set to the plane node p, the data term D of the second edge is 0. When no correct label is set for the plane node p among the second edges {p, B}, the value of D _B (x, y) in the above equation (2) is the value of the second edge. The data term D becomes 0 or 1.

Of the second edges {p, O} connecting the arbitrary plane node p and the non-building node O, when the correct label for the non-building is set for the plane node p, the value of K in the above equation (1) Becomes the data term D of the second edge, which is a value sufficiently larger than 1. On the other hand, when the correct answer label of the building is set to the plane node p among the second edges {p, O}, the data term D of the second edge is 0. If no correct label is set for the plane node p among the second edges {p, O}, the value of D _O (x, y) in the above equation (3) is the value of the second edge. The data term D becomes 0 or 1.

  In the above description, the data term calculation unit 44 sets the data term D at the second edge using both the correct answer label and the elevation data 33. However, the data term calculation unit 44 may set the data term D by another method. For example, the data term calculation unit 44 can generate the altitude from the correct answer label and set the data term D without using the altitude data 33 stored in advance in the storage unit 3. Specifically, the ground height indicated by the non-building correct answer label area can be used as the altitude, and more specifically, as follows.

  All the surfaces constituting the three-dimensional map information 31 have three-dimensional coordinates. When the horizontal coordinates of the gravity center positions of these surfaces are included in the correct label area of the non-building, it is assumed that the height of the gravity center position represents an approximate altitude. When the altitude at a certain position is obtained, the median height of the center of gravity representing the altitude existing around the position is set as the height of the ground. The search range is expanded until a certain number of points representing the altitude are found (for example, 300 points or more).

  In this case, since the correct answer label is generated based on the 2D map data 32 and the 3D map information 31, it can be said that the data term D is generated based on the 2D map data 32 and the 3D map information 31.

  Returning to FIG. 2, when the calculation of the smooth term S for the first edge and the calculation of the data term D for the second edge are completed, the segmentation unit 45 converts the graph structure including the graph including the building node B representing the building and the non-building. The edge is cut so as to be divided into two graphs including the non-building node O to be represented (step S5). At this time, it is desirable to minimize the sum of the weights of edges to be cut (smooth term S at the first edge and data term D at the second edge). As a specific example, the Mincut algorithm (Andrew V Goldberg and Robert E Tarjan. A New Approach to the Maximum-flow Problem. J. ACM, 35 (4): 921-940, 1988.).

  FIG. 8 is a diagram schematically illustrating the graph cut. As shown in the figure, the edges are cut so that all face nodes belong to one of the graph including the building node B (right side of the figure) and the graph including the non-building node O (left side of the figure). Is done. The attribute labeling that the surface represented by each surface node is a building is performed on the surface node group belonging to the graph including the building node B. On the other hand, attribute labeling that the surface represented by each surface node is a non-building is performed on the surface node group belonging to the graph including the non-building node O.

  Thereby, the building part of the three-dimensional map information 31 can be segmented. Moreover, since the attribute label of a building part is divided | segmented by the attribute label of a non-building part for every building, it can also distinguish each building.

  In the example of FIG. 8, the three planes represented by the three plane nodes included in the region C1 surrounded by the alternate long and short dash line are directly connected to each other via the first edge (that is, not via the building node B). Since the first edge is not cut by the graph cutting process, it belongs to a single building. Similarly, the three faces represented by the three face nodes included in the region C2 also belong to a single building.

  However, the plane node included in the area C1 and the plane node included in the area C2 are not directly connected (only indirectly connected via the building node B). Therefore, it can be seen that the building to which the three faces in the region C1 belong and the building to which the three faces belonging to the region C2 belong are different.

  FIG. 9A is a diagram showing polygon data in the three-dimensional map information 31. FIG. 9B is a diagram obtained by performing the above-described processing on the three-dimensional map information 31 of FIG. 9A. In FIG. 9B, it can be seen that the ground, roads, and the like that appear in FIG. 9A are removed, and only the buildings are extracted.

  As described above, in the present embodiment, a graph structure including a plane node, a building node, a non-building node, and an edge is generated from the 3D map information 31. Furthermore, a smooth term S indicating the likelihood of a boundary between a building and a non-building is set as a weight for an edge connecting surface nodes representing adjacent surfaces. Further, the data term D indicating the building-likeness is set as a weight for the edge connecting the plane node representing each plane and the building node, and the edge connecting the plane node representing each plane and the non-building node. . Then, the graph structure is separated into the building side and the non-building side so that the sum of the weights of the edges to be cut is minimized. Thereby, even if a planting, a road structure, etc. are contained in the three-dimensional map information 31, a building can be extracted accurately and adjacent buildings can also be distinguished.

  By separating buildings in this way, applications such as associating attribute labels such as building names and addresses with building parts or replacing only non-buildings (typically the ground part) with other data can be used. Is possible.

(Second Embodiment)
In the first embodiment described above, each surface is separated from a building and a non-building. In this case, planting and road structures (hereinafter referred to as planting and the like) are separated as non-buildings. On the other hand, in the second embodiment described below, each surface is separated into the ground and the rest (hereinafter referred to as non-ground). In this case, planting or the like is non-ground.

  When performing such separation, the correct label setting process using the two-dimensional map data 32 in step S3 of FIG. 2 may be omitted. That is, the data term calculation unit 44 calculates the data term D based on the altitude data 33 for all the second edges. Since the altitude data 33 is used, the ground and non-ground are distinguished. On the other hand, since the 2D map data 32 is not used, it is not distinguished between planting and the like. In the second embodiment, the building node B and the non-building node O in the first embodiment are virtual nodes corresponding to the non-ground and the ground, respectively.

  As a result, each surface is separated into the ground and the non-ground by the graph cut process in step S5 of FIG.

(Third embodiment)
In a third embodiment to be described next, each surface is separated into a building, the ground, and planting.

  According to the method described in the first embodiment, each surface can be separated into a building and a non-building, and the non-building includes the ground and planting. On the other hand, according to the method described in the second embodiment, each surface can be separated into the ground and the non-ground, and the non-ground includes buildings, planting, and the like.

  Therefore, planting or the like can be separated from the difference between the result of applying the method of the first embodiment and the result of applying the method of the second embodiment to the three-dimensional map information 31. More specifically, the surface separated as non-building in the result of applying the method of the first embodiment and separated as non-ground in the result of applying the method of the second embodiment is planting or the like. is there. Thus, each surface in the three-dimensional map information 31 can be separated into a building, the ground, and planting.

  The embodiment described above is described for the purpose of enabling the person having ordinary knowledge in the technical field to which the present invention belongs to implement the present invention. Various modifications of the above embodiment can be naturally made by those skilled in the art, and the technical idea of the present invention can be applied to other embodiments. Therefore, the present invention should not be limited to the described embodiments, but should be the widest scope according to the technical idea defined by the claims.

DESCRIPTION OF SYMBOLS 1 Input device 2 Output device 3 Memory | storage part 31 3D map information 32 2D map data 33 Elevation data 34 Graph data 4 Control part 41 Graph structure generation part 42 Smooth term calculation part 43 Correct label setting part 44 Data term calculation part 45 Segmentation Part 100 3D Map Separation Device

Claims (8)

A three-dimensional map separation device that separates a plurality of faces constituting three-dimensional map information expressed by polygon data for each face attribute,
A graph structure generation unit that generates a graph structure indicating the adjacent relationship between the plurality of surfaces using the three-dimensional map information;
A first weight setting unit for obtaining a first weight indicating a possibility that an attribute of one surface and an attribute of the other surface are different using an angle at which two adjacent surfaces face;
A second weight setting unit for obtaining a second weight indicating a certain attribute of the surface from the height of a certain surface and the height of the ground at the point of the surface;
A three-dimensional map separation device comprising: a segmentation unit that performs graph cut processing on the graph structure using the first weight and the second weight, and separates each surface for each attribute. The graph structure generation unit uses the three-dimensional map information,
A surface node representing each surface;
A first virtual node and a second virtual node which are virtual nodes corresponding to different attributes;
A first edge connecting face nodes of adjacent faces;
Generating the graph structure composed of each of the first virtual node and the surface node, and a second edge connecting each of the second virtual node and the surface node;
The first weight setting unit sets the first weight for each of the first edges,
The second weight setting unit sets the second weight for each of the second edges,
The three-dimensional map separation device according to claim 1, wherein the segmentation unit separates a plane node connected to the first virtual node and a plane node connected to the second virtual node. The second weight setting unit obtains a difference between the height of the surface and the height of the ground at the point of the surface,
Setting the second weight that the attribute of the face is more likely to be the ground as the difference is smaller to the second edge connected to the first virtual node;
The second weight that the attribute of the face is more likely to be non-ground as the difference is larger is set to the second edge connected to the second virtual node,
The segmentation unit performs a graph cut process on the graph structure, uses a surface attribute represented by a surface node connected to the first virtual node as a ground, and a surface node connected to the second virtual node The three-dimensional map separating apparatus according to claim 2, wherein the attribute of the surface represented by is separated as non-ground. A correct label for which the attribute of the surface corresponding to the area inside the reduced outer peripheral line reduced by a predetermined amount of the area formed by the outer peripheral line of the building represented by the two-dimensional map data is a building, A correct label setting unit for obtaining a correct label in which the attribute of the surface corresponding to a region outside the enlarged outer peripheral line enlarged by a predetermined amount is a non-building,
The second weight setting unit sets the second weight that the attribute of the face is likely to be a building using the correct label as the building to the second edge connected to the first virtual node, and Using the correct label as a non-building, setting the second weight that the attribute of the surface is likely to be a non-building to the second edge connected to the second virtual node;
The segmentation unit performs a graph cut process on the graph structure, sets a plane attribute represented by a plane node connected to the first virtual node as a building, and a plane node connected to the second virtual node The three-dimensional map separation apparatus according to claim 2, wherein the attribute of the surface represented by is separated as a non-building.   When the region formed by the enlarged outer peripheral line of a certain outer peripheral line overlaps with the region formed by the enlarged outer peripheral line of another outer peripheral line, the correct label setting unit is configured such that the certain outer peripheral line is included in the overlapping region. The area formed by the second predetermined outer circumference expanded by the second predetermined amount smaller than the predetermined amount and the area formed by the other outer peripheral line is formed by the region outside the second enlarged outer peripheral line expanded by the second predetermined amount. The three-dimensional map separation device according to claim 4, wherein a correct label for a non-building is set.   6. The three-dimensional map separation device according to claim 1, wherein the second weight setting unit obtains the height of the ground using elevation data indicating the height of the ground stored in advance.   The three-dimensional map according to any one of claims 1 to 6, wherein the second weight setting unit obtains the height of the ground using two-dimensional map data representing an area corresponding to the ground and the three-dimensional map information. Separation device. A three-dimensional map separation program for separating a plurality of faces constituting three-dimensional map information expressed by polygon data for each face attribute,
A graph structure generation unit that generates a graph structure indicating the adjacent relationship between the plurality of surfaces using the three-dimensional map information;
A first weight setting unit for obtaining a first weight indicating a possibility that an attribute of one surface and an attribute of the other surface are different using an angle at which two adjacent surfaces face;
A second weight setting unit for obtaining a second weight indicating a certain attribute of the surface from the height of a certain surface and the height of the ground at the point of the surface;
A three-dimensional map separation program that functions as a segmentation unit that performs graph cut processing on the graph structure using the first weight and the second weight, and separates each surface for each attribute.