JP2003323640A - Method, system and program for preparing highly precise city model using laser scanner data and aerial photographic image - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To construct a highly precise three-dimensional city model by combining the advantages of the altitude precision of laser scanner data, the advantages of the high resolution of an aerial photograph, and a conventional variable density DEM (Digital Elevation Model). <P>SOLUTION: This system is provided with a coordinate calculating module 10, a ground height calculating module 11, a region division processing part 15, a horizontal face stereo matching processing part 16, a region division processing part 17, a polygon generation processing part 18, and a three-dimensional modeling processing part 19. Each of stereo aerial photographic pictures is divided into regions to obtain an image region whose hue is uniform. Stereo matching is carried out by using the altitude value of the three-dimensional point group of the laser scanner overlapping each color region as an initial value. When the calculation result of the obtained altitude value represents almost the same altitude as the initial value, the altitude value being the calculation result is applied to the region to generate the DSM. Then, the horizontal region of the obtained DSM is recognized by region division, and linear extraction is carried out in a peripheral region on the stereo image corresponding to each region, and the results are polygonzied. <P>COPYRIGHT: (C)2004,JPO

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a high-accuracy city model generation apparatus that can easily and accurately convert a building surface of an aerial photograph image into three dimensions by using laser scanner data and an aerial photograph image.

[0002]

2. Description of the Related Art A method for generating a highly accurate city model is
Generally, there are a method using laser scanner data and a method using stereo matching.

For example, laser scanner data is obtained by scanning laser light from an aircraft such as an airplane or a helicopter onto the surface of the ground and analyzing the received light data to obtain laser data as shown in FIG. This Figure 21
Target area (shown in aerial photograph) in Figure 21 (b)
FIG. 21A shows the laser data. As shown in FIG. 20A, the laser data is a series of fine dots, each of which has three-dimensional coordinates (x, y, z).

Therefore, since the laser scanner data has three-dimensional coordinates, the DSM (Digital) shown in FIG. 22 has a high altitude accuracy as compared with stereo matching of aerial photographs.
It is an abbreviation for tal Surface Model, and can be a model that represents the shape of the ground surface including features such as buildings.

On the other hand, there is also a stereo matching method (see FIG. 20) in which automatic elevation measurement is performed using aerial photographs. In this stereo matching, the correspondence relationship of each pixel between the images is obtained, and the three-dimensional coordinates at each corresponding point position are obtained by the principle of triangulation, and the altitude data can be measured at high density.

[0006]

However, although the method using the laser scanner data can obtain highly accurate data regarding the elevation, when generating data having many linear structures such as a city model from a three-dimensional point cloud. , It is necessary to further densify the data in order to accurately take the boundary part.

That is, the laser data is a fine point sequence (three-dimensional point group), and it is not easy to obtain a resolution higher than this.

Further, when the density of the laser scanner data is increased, the cost for acquiring the data and the data capacity become enormous, and there are problems in terms of both cost and data handling.

On the other hand, the image used for the stereo matching method has a high resolution and is effective for identifying the boundary or area of the artificial structure. However, the stereo matching has a low altitude measurement accuracy in a place where the altitude changes drastically due to an influence such as hiding, which causes a problem in directly measuring the boundary three-dimensionally.

The present invention has the advantages of elevation accuracy of laser scanner data, the advantage of high resolution of aerial photography, and the existing sparse density DEM (Digital Eleva).
An abbreviation of "Tion Model", a model expressing the ground height. You can use it as the ground height because you can see the approximate ground height.)
The purpose is to build a highly accurate three-dimensional city model by combining

[0011]

A method of generating a high-accuracy urban model using laser scanner data and aerial photograph images of the present invention is a method of continuously generating the same color from aerial photograph images of a predetermined area stored in a database. The color area of the pixel is extracted, and after the extraction, horizontal plane stereo matching using the color area is performed on the digital ground surface model obtained based on the laser scanner data of the predetermined area stored in the database to perform DSM. By using the horizontal area of the DSM and the aerial photograph image, linearization is performed from the line structure information on the image corresponding to the peripheral portion of the horizontal area of the DSM, and the pixels in the area surrounded by the straight line are generated. The gist of the present invention is to convert the horizontal region into a three-dimensional dot array by converting the array into three-dimensional.

Further, the high-precision city model generation system of the present invention comprises means for continuously extracting color regions of pixels of the same color from aerial photograph images of a predetermined area stored in the database, and after the extraction, A means for generating a DSM area by performing horizontal plane stereo matching using the color area on a digital ground surface model obtained based on the laser scanner data of the predetermined area stored in a database; and a horizontal area of the DSM. Means for linearizing from the line structure information on the image corresponding to the peripheral portion of the horizontal area of the DSM, and a pixel row in the area surrounded by the straight line using the And a means for converting the horizontal region into a three-dimensional point sequence.

Further, the program for generating a high-precision city model of the present invention is a means for causing a computer to read laser scanner data of a predetermined area stored in a database, and an aerial photograph image of the predetermined area stored in the database. Means for sequentially extracting a color region of a group of pixels having a uniform color, and causing the digital stereo model of a predetermined region stored in the database after the extraction to perform the horizontal plane stereo matching assuming that the color region is horizontal. Means, the calculation result of the elevation value obtained as a result of the horizontal plane stereo matching means for determining whether the elevation value is similar to the laser scanner data of the digital ground surface model, when determining the elevation value of the same, Means for applying an elevation value to the color area to generate a DSM and storing the DSM; A flat area is used as a mask area, an expanded image and a contracted image are generated, means for synthesizing these areas to obtain a line extraction area around the mask area, the mask area of the aerial photograph image, and A means for obtaining an edge image of a portion corresponding to the periphery, superimposing the edge extraction image and the line extraction area, and extracting an edge image around the mask area, and performing Hough conversion on the peripheral edge image. Means for linearizing each edge of the peripheral edge image, and the linearized linearized edge image is compared with the contraction mask image to obtain a line segment from which unnecessary edges are removed from the linearized edge image. Means for generating a polygon obtained by tracing the line segment, and means for extracting a polygon coordinate sequence that most overlaps the mask area from the polygon, the polygon coordinate sequence And summarized in that with a program for functioning as a means for three-dimensional the extracted color region giving said elevation value for.

[0014]

BEST MODE FOR CARRYING OUT THE INVENTION In the present embodiment, by incorporating area information and line structure information obtained from stereoscopic aerial photograph image data into laser scanner data,
The boundaries of artificial structures are extracted accurately, and automatic building polygonalization and three-dimensional modeling are performed.

First, by dividing each of the stereo aerial photographic images into regions, an image region having a uniform color is obtained. Such image areas are considered to correspond to the rooftop of the building, the road surface, the ground, and other large flat areas.

Stereo matching is performed with the elevation value of the three-dimensional point group of the laser scanner overlapping each color region as an initial value. When the calculation result of the elevation value obtained as a result becomes almost the same as the initial value, the elevation value of the calculation result is given to the area, and the DSM (Digital Surface Mo
Abbreviation of del, a model expressing the shape of the ground surface including features such as buildings is generated. As a result, the feature boundary, which was ambiguous in the laser scanner data, can be expressed with higher accuracy.

Next, the horizontal regions are sequentially recognized by region division from the highest DSM obtained above, and straight lines are extracted in the peripheral region on the stereo image corresponding to each region, and the result is polygonalized. . An area in which the difference in DSM with respect to the existing DEM or the average ground height of the area is equal to or less than a threshold value that is arbitrarily set is an area other than the building and is estimated to be the ground height.
Thereby, the building area can be made into a three-dimensional polygon. Furthermore, a three-dimensional model can be obtained by forming the polygonal building area into a prismatic shape.

FIG. 1 shows the configuration of a three-dimensional measuring apparatus that realizes the present invention. The three-dimensional measurement system 1 comprises a general personal computer including a personal computer main body 2, an external storage device 3 for storing stereo images and the like, and a display 4. In addition, the following is given as a premise.

(1) Input data It is assumed that the following various data are stored in the external storage device 3.

[0020] (a) and stereo aerial photograph image (I _L and I _R (FIG. 20)) and its orientation parameters are saved (b) laser scanner data. This laser scanner data is a set of three-dimensional coordinates. In the following, LP _i (i =
1 ... n _LP ) (FIG. 21 (a)).

(C) Digital Terrain Model (DEM)).
Representing the ground height, given data such as national numerical information may be used (FIG. 22).

(2) Calculation Module It is assumed that the following calculation module is provided. This calculation module includes a coordinate calculation module 10, a ground height calculation module 11 and the like.

The coordinate calculation module 10 reads the orientation elements of the stereo aerial photographic image and performs conversion between the three-dimensional coordinates and the coordinates of the stereo images I ₁ and I ₂ . This is the same as that given as the coordinate conversion function / reverse coordinate conversion function in Japanese Patent Application No. 2000-251456.

That is, for the brightness images IL and IR, the coordinate conversion function FX and the inverse coordinate conversion function function GX (collectively referred to as orientation information) for calculating the image coordinates of each image on which the three-dimensional coordinates on the ground are projected. To use.

For example, a point (x, y, z) in a three-dimensional space
Pixel coordinates (x _1pix , y
_1pix ) and ( _x2pix , _y2pix ).

That is, the following four coordinate conversion functions FX1
(X, y, z), FY1 (x, y, z), FX2 (x,
y, z) and FY2 (x, y, z) are determined analytically or procedurally.

X _1pix = FX1 (x, y, z) Formula (1) y _1pix = FY1 (x, y, z) Formula (2) x _2pix = FX2 (x, y, z) Formula (3) y _2pix = FY2 (x, y, z) Equation (4) Further, when Equation (1) and Equation (2) are solved analytically or procedurally with z = constant, it is found that a point on each image is at an altitude z. It is possible to obtain the x-coordinate and the y-coordinate on the ground when it is assumed.

X = GX1 (x _1pix , y _1pix , z) Expression (5) y = GY1 (x _1pix , y _1pix , z) Expression (6) Similarly, solving Expressions (3) and (4) gives: x = GX2 ( _x2pix , _y2pix , z) Formula (7) y = GY2 ( _x2pix , _y2pix , z) Formula (8).

Such a function is used, for example, when the irregular window shown in FIG. 2 is used. In FIG. 2, an image I1 obtained by an aircraft is displayed on the screen, and a closed region B1 corresponding to the contour of the feature Bu is given on the image by mouse operation or the like. Then, the irregular window Plh is generated at a position where the projection coincides with this closed region, and the altitude H (z) is changed.

For example, when the altitude is changed in the order of 30 m, 80 m, and 150 m, the projection positions of the brightness images I1 and I2 of the irregular window Plh in the brightness images I1 and I2 when they change to 30 m, 80 m, and 150 m are obtained each time. While
The height (elevation value) of the irregular window P1h when it is determined by the evaluation function that the luminance values in the luminance images I1 and I2 when the irregular window Plh is projected at this position are compared (estimated) by the evaluation function I am trying.

The ground height calculation module 11 reads the DEM and calculates the ground height at a given horizontal coordinate position by using an appropriate interpolation calculation.

H = GElv (x, y) For example, if the DEM is a square mesh data, an interpolation method such as the bilinear method is used. If the DEM is a three-dimensional point group having a random arrangement, an interpolation method using a triangular mesh is used. To use.

That is, in the present embodiment, the coordinate calculation module 10, the ground height calculation module 11, etc.
Using the area information and line structure information obtained from the aerial photograph image data captured in stereo with the laser scanner data, for example, the boundaries of artificial structures can be accurately extracted, and automatic building polygonalization and three-dimensional modeling can be performed. To do.

In the generation of this three-dimensional model, first, each of the stereo aerial photographic images is divided into regions to obtain image regions of uniform color.

It is considered that such an image area corresponds to the roof of a building, the road surface, the ground, and other wide flat ground.

Stereo matching is performed with the elevation value of the three-dimensional point group of the laser scanner overlapping each color region as the initial value, and when the resulting elevation value calculation result is almost the same as the initial value, the stereo matching is performed. The elevation value of the calculation result is given to the area, and the DSM is generated. As a result, the feature boundary, which was ambiguous in the laser scanner data, can be expressed with higher accuracy.

Subsequently, the horizontal area of the DSM obtained above is recognized by area division, straight line extraction is performed in the peripheral area on the stereo image corresponding to each area, and the result is polygonalized.

Areas where the difference of DSM with respect to the existing DEM or the average ground height of the area is less than or equal to a threshold value set arbitrarily,
It is an area other than buildings and is estimated to be the ground height. Thereby, the building area can be made into a three-dimensional polygon. Furthermore, a three-dimensional model can be obtained by forming the polygonal building area into a prismatic shape.

The outline of the embodiment of the present invention will be briefly described with reference to FIG. As shown in FIG. 1, it consists of the following five parts.

First, a stereo aerial photograph image (IR, IL) stored in a database (not shown) is read, and each of the stereo aerial photograph images is divided into regions (regions having the same color as a group: the same). The area segmentation processing unit 15 based on the color of the stereo image for obtaining the image area (which is considered to correspond to the rooftop of the building, the road surface, the ground, and other large flat grounds) by making the color pixels continuous) And a horizontal plane stereo matching processing unit 16 that performs stereo matching with the elevation value of the three-dimensional point group of the laser scanner data overlapping each divided area (also referred to as a divided color area) as an initial value, and the result after this horizontal plane stereo matching processing. If the calculated altitude value is almost the same as the initial value, the calculated altitude value is given to that area, and the DSM (Digi al Surface Mod
Abbreviation of el, a DSM area division processing unit 17 that generates a model that represents the shape of the ground surface including features such as buildings.
It includes a polygon generation processing unit 18, a three-dimensional modeling processing unit 19 and the like.

Details of each of these processes (region dividing process by color of stereo image, horizontal plane stereo matching process, DSM dividing process, polygon generating process, three-dimensional modeling process) will be described below with reference to flowcharts.

(Stereo Image Color Segmentation Processing Unit 15) The stereo image color segmentation will be described in detail with reference to the flowchart of FIG.

First, parameters required for area division are set (S10). For example, when performing area division by the iterative area expansion method, it is assumed that the following three parameters are specified by the user.

Initial threshold value, threshold value increment, threshold value for repetitive truncation Next, the left image I _L of the stereo aerial photograph image is read in from the external storage device 3 (S11).

Next, area division processing is performed using the luminance information and color information of the image (see FIG. 4). In this area division processing, first, small isolated points are removed by a median filter or the like, and then an identification number is given to each continuous area in which the luminance and color of the removed image are uniform.

As the area dividing method, a widely used area dividing method such as an iterative area expanding method may be used.

Then, as a result of the area division (area painted in a single color), a label number for identifying to which area of the identification number each pixel belongs is given. The array that stores the label number of the left image is LL (i, j).

That is, if LL (i, j) = n, it means that the pixel of the image i column and the j line belongs to the area of the label number n.

Next, a region division result storage process is performed in which the label storage array LL (i, j) obtained by the region division is stored in the external storage device 3 as the label of the left image (S1).
3).

Then, the right image reading process (S14) and the region dividing process (S15), which are the same as the left image reading process (S11) to the region dividing result saving process (S13), are performed on the right image I _R. , Right image label storage array LR
(I, j) is stored in the external storage device 3 (S16).

That is, in S14 to S16, the right image is read, the area is divided, and the area division result is saved.

Next, the horizontal plane stereo matching process will be described.

(Horizontal Plane Stereo Matching) It is assumed that each of the divided image areas is horizontal, and the area-divided portion is subjected to the "inter-image extended image matching method using an irregular window (Patent application 2000-25145
6) ”to generate a digital ground surface model (DSM). Here, the DSM expresses the altitude of the ground surface on mesh points. For example, when the origin of the mesh is (x0, y0) and the mesh interval is d, the DSM array h (ix, iy) has coordinates (x0 + d · ix, y0 +).
d. iy) altitude shall be given. Also, DSM
The origin and size of the array are set according to the target range given by the user. In the horizontal plane stereo matching described above, each divided area is used as an irregular window, the projection position on both images of the extended stereo model is obtained by changing this, and the brightness value of the area separated by the irregular window at this projected position. Then, the brightness values are compared with each other, and the altitude imparting process is performed by the extended stereo matching in which the height of the irregular window when the comparison result is the best match is the height of the feature in the extended stereo model.

FIG. 5 is a flow chart for explaining the plane stereo matching. First, the left and right images are read in the memory (S20).

Next, the DSM array for storing the DSM array is initialized (S21). The DSM array to be initialized is the following 3
Let's do it.

HL (ix, iy): DSM generated from the left image area division result hR (ix, iy): DSM generated from the right image area division result hLR (ix, iy): Left and right image area DSM generated from division result The origin and range of the DSM array are given by the user in advance. Upon initialization, the values of all the arrays are set to a constant value hinit. hinit is a small value (-9999, etc.) that is impossible in reality,
The fact that the DSM sequence is hinit indicates that the DSM has not been calculated by plane stereo matching.

Next, the result of dividing the image areas of the left and right images is read (S22). That is, the image area division result is read into the array L (i, j) on the memory.

Then, the following processing is performed in the order of numbers with respect to the respective areas having the label numbers recorded in L (i, j) for each area.

In this process, the laser scanner data is overlaid and the approximate altitude is calculated (S23).

For example, the projection position of the laser scanner data L _{pi on} the image is calculated by the coordinate calculation module, and the one that overlaps the area being processed is searched (see FIG. 6). Next, make a histogram of the points that overlap,
The altitude H _{mode at} which the histogram becomes maximum is given as the approximate altitude of the region.

Next, the elevation value is calculated by plane stereo matching (S24). In this calculation, the optimum altitude of the image area is calculated by the irregular window method of Japanese Patent Application No. 2000-251456, using the area shape as an irregular window (see FIG. 7).

The altitude of the calculation result is H _opt .
When LL (i, j) is read in L (i, j), the image I _L is I ₁ and I _R is I ₂ . Also, LR (i,
When j) is read, the image I _R is I ₁ and I _L is I ₂ .

Then, it is tested whether or not the altitude value satisfies the following condition (S25).

| H _mode −H _opt | <ΔH where ΔH is the threshold value of the altitude test.

Next, if the conditions for the elevation value test are satisfied, the coordinate calculation module 10 is used to obtain the ground coordinates corresponding to the image area. If the processed image is the left image, the DSM array h
The altitude value H _opt is stored in L (ix, iy). If it is the right image, DSM is stored in array hR (ix, iy)). If not processed for all the regions, the process returns to S23 (S27). When the processing has been completed for all areas, the process returns to S22 to read the next image, and the same processing is performed (S28).

Next, the DSM arrays hL (x, y) and hR
A DS that fuses (x, y) to produce hLR (x, y)
Fusion processing of M is performed (S29: refer to FIG. 8).

[0067]

[Equation 1] The DSM array is stored in the external storage device (S30).

(DSM Area Division) The DSM is divided into horizontal areas. FIG. 9 shows a flow of DSM area division.

Area division of DSM is performed as shown in FIG.
The DSMhLR (x, y) is read into the DSM array h (x, y) (S31).

Then, the read DSM array h
(X, y), the area is divided for each horizontal area (S32).

More specifically, a continuous area having a constant height Dh is defined as one area. Then, an identification number is attached to each area as a label, and the label storage array LH
Store in (x, y). That is, LH (x, y) = n
Then, it means that DSMh (x, y) belongs to the area of the identification number n. The area corresponding to the identification number n is R
LH (n).

Next, this label storage array LH (x, y)
Is stored in the external storage device (S33).

(Polygonization of Horizontal Area) Next, the boundary of each horizontal area is polygonized by a straight line. Stereo images are used for polygonization. FIG. 10 shows a schematic flow of polygonization.

First, the DSM and its area division result are read (S40). Specifically, the label LH (x, y) and the corresponding DSMh (x, y) are read from the external storage device.

Next, as processing for each horizontal area, LH
The following processing is performed for each area of the label number k = 1 to n (n is the number of labels) stored in (x, y).

First, the ground height calculation module 11 is used to calculate the ground height HG at the position of each region RLH (k). At this time, the difference from the altitude H (k) of the area is a threshold value (2 m
The height from the ground surface is determined by the following formula (S41).

H (k) -HG> ΔHB Then, when the difference from the altitude H (k) of the area is larger than the threshold value, this area is regarded as a building and the following processing is performed.

An image area corresponding to the label area RLH (k) is obtained by using the coordinate calculation module 10 and a "cut-out of corresponding image area" processing for cutting out the area including the periphery is performed (S42).

At this time, as shown in FIG. 11, the clipping origin is (xc0, yc0), the clipped image is Ic, and the clipping size is sxc, syc. Further, the mask image Icm of DSM having the same size as Ic is set.

This Icm mask image is RL on Ic.
It is a binary image in which the code T (≠ 0) is in the area corresponding to H (k) and is 0 in other areas.

Next, on the image Ic, buffering processing of the line extraction area for generating the buffer Icb for line extraction is performed in order to perform straight line extraction in the peripheral portion of the mask specified by Icm (S43: (See FIG. 12).

Specifically, as shown in FIG. 12, Icb is an expansion mask Icmf obtained by expanding Icm Nfat times.
At and a contraction mask Icmthin obtained by contracting Icmfat Nth times are generated, and the following relational expression is established from these two masks.

Icb (i, j) = T (Icmfat (i, j) = T∩Icmthin (i, j) = 0) = 0 (other than that) The contraction mask Icmthin is the same as or smaller than the original mask Icm. Less than. Therefore, Nfat <Nth
in.

Next, as shown in FIG. 13, edge extraction is performed on the cut image Ic (S44). The edge extraction method satisfies the following specifications, for example.

Binarization of points on the edge and other points.

The edge is thinned (the width of the edge is 1 pixel).

As such an edge extraction method, Can
There are a ny operator and a Susans operator.
Alternatively, an edge enhancement operator such as Sobel may be applied, and then binarization and thinning may be performed.

The edge obtained here is defined as Ice. By taking a common area of the mask image Icb of the line extraction buffer in this, an edge image of only the peripheral portion of the mask is Ic
Get from e. This edge image is defined as Iceb (see FIG. 1).
4).

Next, the edge extraction image Ice around the mask
Linearization processing is performed to extract a straight line by the Hough transform principle using points on the edge of b (S45: see FIG. 14).

In the Hough transform, the equation of the straight line is calculated by using the normal direction θ and the distance ρ from the origin, and xcosθ + ysinθ =
This is a method of extracting a straight line in the (θ−ρ) space by placing ρ. Hough transform is a method for extracting a figure such as a straight line or a circle from an image.

As the number of extracted straight lines n _hough ,
From the parameter pair (θ ₁ , ρ ₁ ) of the straight line passing through the points on the most edges to the pair of straight lines (θ _nhough ρ _nhough ) passing through the n _hough- th edge point are extracted (FIG. 1).
5).

Next, intersection points are obtained for all the straight lines extracted by the Hough transform, and the line segmentation processing for the line segments is performed (S46, line segmentation in FIG. 14).

Then, a process of deleting the overlapping portion with the mask for deleting the line segment overlapping the contraction mask Icmthin is performed (see FIG. 14).

Next, the line segment is made polygonal for each area (point sequence formation).
In addition, among the polygons, the polygon having the largest common area with the mask image Icm is defined as the label area RLH.
The polygon corresponding to (k) is set (S48). This polygon is set as the image coordinate sequence PlI (k) (see FIG. 16).

Next, PlI (k) given in image coordinates
Is converted into a polygon PlT (k) in the three-dimensional space, and a three-dimensional point sequence processing is performed (S49). At this time, the Z coordinate of the polygon in the three-dimensional space is the altitude H (k) of the area (see FIG. 1).
7).

Then, the obtained polygonal coordinate sequence PlI
After storing (k) in the external storage device (S50), the process moves to the next area (S51).

(3D Modeling Process) The description of the 3D modeling process will be supplemented. FIG. 18 shows a schematic flow of converting each polygon into a prismatic three-dimensional model.

In the three-dimensional modeling process, the polygon coordinate sequence PlI (k) shown in FIG. 16 is read from the external storage device (S60), and the ground surface position of the polygon is calculated (S60).
61).

In this calculation, the ground height calculation module obtains the height of the ground surface position at the horizontal coordinate position of each point forming the polygonal coordinate sequence PlT (k), and PlT
A bottom polygon PlE (k) corresponding to (k) is calculated (see FIG. 19).

Next, the polygon PlT (k) and the bottom polygon P
A three-dimensional model of a prism BLD with the sides corresponding to lE (k) and the quadrangles connecting both ends of these sides with vertical lines as side surfaces.
A polygonal building modeling process is performed by using as a building model (S62).

This prism three-dimensional model BLD has a top polygon PlT (k) and a bottom polygon PlE (k), and a side polygon group PlW _i (i = 1, ... n _{PlT (k)} ) of the _prism. (See FIG. 19). Here, n _{PlT (k)} is the number of vertices of the top polygon PlT (k).

Then, the obtained prismatic three-dimensional BLD is stored in the external storage device (S63). When storing, V
It is preferable to use a three-dimensional model notation format such as RML.

Therefore, it can be used for the accurate calculation of the building shape in the aerial photograph and the generation of the three-dimensional model.

In the above embodiment, the horizontal plane stereo matching is described as the horizontal plane stereo matching using the irregular window, but the flat plane stereo matching method described in Japanese Patent Application No. 2002-127512 may be used.

This plane stereo matching utilizes the fact that planes appearing in a stereo photograph are associated with each other by projective transformation based on the principle of central projection.
The projective transformation coefficient on the surface is automatically determined by setting a constraint condition that satisfies the epipolar condition with respect to the correspondence between points (none of the three points is on a straight line).

As described above, according to the present invention, the color regions of the pixels of the same color are continuously extracted from the aerial photograph image of the predetermined area stored in the database, and after the extraction, the color regions are stored in the database. Stereo matching using the color area is performed on the digital ground surface model obtained based on the laser scanner data of the predetermined area.
Generate SM.

Then, using the horizontal region and the image in the DSM, linearization is performed from the line structure information on the image corresponding to the peripheral portion of the horizontal region, and the pixel row in the region surrounded by this straight line is three-dimensionalized. By doing so, the horizontal region is made into a three-dimensional point sequence.

As a result, the effect that the feature boundary, which was ambiguous in the laser scanner data, can be expressed with higher accuracy is obtained.

[Brief description of drawings]

FIG. 1 is a schematic configuration diagram of a high-precision city model generation device using laser scanner data and aerial photograph images.

FIG. 2 is an explanatory diagram for explaining how to assign an altitude value using an irregular window.

FIG. 3 is a flowchart for explaining area division by color of a stereo image.

FIG. 4 is an explanatory diagram illustrating area division of an image.

FIG. 5 is a flowchart illustrating horizontal plane stereo matching.

FIG. 6 is an explanatory diagram illustrating superposition of laser scanner data.

FIG. 7 is an explanatory diagram illustrating horizontal stereo matching using an irregular window.

FIG. 8 is an explanatory diagram illustrating DSM synthesis.

FIG. 9 is a flowchart illustrating DSM area division.

FIG. 10 is a flowchart illustrating polygonization of a horizontal area.

FIG. 11 is an explanatory diagram illustrating clipping of an image area.

FIG. 12 is an explanatory diagram illustrating buffering of a line extraction area.

FIG. 13 is an explanatory diagram illustrating edge extraction.

FIG. 14 is an explanatory diagram of straightening and line segmentation of a straight line.

FIG. 15 is an explanatory diagram of Hough conversion.

FIG. 16 is an explanatory diagram of polygonization.

[Fig. 17] Fig. 17 is an explanatory diagram illustrating three-dimensional dot string formation.

FIG. 18 is a flowchart illustrating three-dimensional modeling.

FIG. 19 is an explanatory diagram illustrating a three-dimensional modeling of a building according to this embodiment.

FIG. 20 is an explanatory diagram illustrating a stereo aerial photograph.

FIG. 21 is an explanatory diagram illustrating laser scanner data.

FIG. 22 is an explanatory diagram illustrating a digital terrain model.

[Explanation of symbols]

10 Standard calculation module 11 Ground height calculation module 15 Area division processing unit 16 Horizontal stereo matching unit 17 Area division processing unit 18 Polygon generation processing unit 19 3D modeling processor

─────────────────────────────────────────────────── ─── Continuation of front page (51) Int.Cl. ⁷ identification code FI theme code (reference) G06T 7/00 200 G06T 7/00 200A 7/60 300 7/60 300A G09B 29/00 G09B 29/00 Z (72) Inventor Ken Doihara Shinjuku-ku, Tokyo Shinjuku-ku 4-2-18 Shinjuku Kofu Building Asia Air Survey Co., Ltd. (72) Inventor Ryosuke Shibasaki 1-5-3-408 F-term, Saroe, Koto-ku, Tokyo 2C032 HB05 HC23 5B050 AA01 BA04 BA09 BA15 DA04 EA06 EA09 EA13 EA19 EA26 FA02 FA06 5B057 AA13 CA01 CA12 CA16 CB13 CB16 CD14 CE08 CE15 CH01 DA08 DA16 DB02 DB06 DC09 DC16 DC25 5L096 AA02 GA10 FA06 CA20 FA04 FA20 FA04 FA20 FA04 FA20 FA04 FA04

Claims (11)

[Claims] 1. A color region of pixels of the same color is continuously extracted from an aerial photograph image of a predetermined region stored in a database, and after the extraction, a laser scanner data of the predetermined region is stored in the database. A horizontal plane stereo matching using the color region is performed on the obtained digital ground surface model to generate a DSM, and using the horizontal region of the DSM and the aerial photograph image, a peripheral portion of the horizontal region of the DSM is generated. Laser scanner data, characterized in that linearization is performed from the line structure information on the image corresponding to, and the horizontal region is transformed into a three-dimensional dot sequence by three-dimensionalizing a pixel column in a region surrounded by the straight line. Method for Generating High-Precision Urban Models Using Image and Aerial Photographs. 2. The horizontal plane stereo matching is performed by extracting uniform color areas by using color information of the aerial photograph image and assuming that each color area is horizontal. Method for high-precision urban model generation using laser scanner data and aerial photograph images of Japan. 3. When the calculation result of the elevation value when the horizontal plane stereo matching is performed becomes almost the same height as the laser scanner data, the elevation value of the calculation result is given to the area to obtain the laser scanner data. The horizontal area having the same elevation value as the obtained DSM is set to the DSM.
The high-precision city model generation method using the laser scanner data and the aerial photograph image according to claim 1, wherein the high-precision city model is generated in the area. 4. A horizontal area in the DSM is extracted by area division, a straight line is extracted in a peripheral area on a stereo image corresponding to each area, and the result is polygonalized so that the 3 areas of the building area are extracted. A method for generating a high-precision city model using laser scanner data and an aerial photograph image according to claim 1, 2 or 3, wherein three-dimensional polygonalization is performed. 5. The three-dimensional polygonalization of the building area is configured in a prismatic shape.
A method for generating a high-precision city model using the described laser scanner data and an aerial photograph image. 6. A means for continuously extracting color regions of pixels of the same color from an aerial photograph image of a predetermined region stored in a database, and a laser scanner of the predetermined region stored in the database after the extraction. Means for performing horizontal plane stereo matching using the color region to generate a DSM region for a digital ground surface model obtained based on data, and using the horizontal region of the DSM and the aerial photograph image, Means for linearizing from the line structure information on the image corresponding to the peripheral part of the horizontal area, and means for converting the horizontal area into a three-dimensional point array by three-dimensionalizing the pixel row in the area surrounded by the straight line A high-precision city model generation system characterized by having. 7. The horizontal plane stereo matching is performed by extracting uniform color areas using color information of the aerial photograph image and assuming that each color area is horizontal. High-precision urban model generation system. 8. When the calculation result of the elevation value when the horizontal plane stereo matching is performed becomes almost the same height as the laser scanner data, the elevation value of the calculation result is given to the area to obtain the laser scanner data. The horizontal area having the same elevation value as the DSM obtained in
9. The high-precision city model generation system according to claim 6, further comprising means for generating the area in M. 9. The horizontal area in the DSM is extracted by area division, straight lines are extracted in the peripheral area on the stereo image corresponding to each area, and the result is polygonalized so that the three areas of the building area are extracted. 9. A high-precision city model generation system using laser scanner data and aerial photographic images according to claim 6, 7 or 8 for performing dimensional polygonalization. 10. The high-precision city model generation system according to claim 6, wherein the three-dimensional polygonalization of the building area is formed in a prismatic shape. 11. A means for causing a computer to read laser scanner data of a predetermined region stored in a database, and a color region of a group of pixels having a uniform color in the aerial photograph image of the predetermined region stored in the database are sequentially arranged. Means for extracting, means for performing the horizontal stereo matching on the digital ground model of the predetermined area stored in the database after the extraction, assuming that the color region is horizontal, calculation of the elevation value obtained as a result of the stereo matching Means for determining whether or not the result is the same altitude value as the laser scanner data of the digital ground surface model, and when determining the same altitude value, apply the altitude value to the color area to generate a DSM. A means for storing, a horizontal area of the DSM is used as a mask area, and an expanded image and a deflated area are used. Means for obtaining a line extraction area around the mask area by synthesizing these areas, and obtaining an edge image of a location corresponding to the mask area and its periphery of the aerial photograph image, Means for overlapping the edge extraction image and the line extraction area to extract an edge image around the mask area; means for performing Hough transform on the peripheral edge image to linearize each edge of the peripheral edge image; A means for obtaining a line segment from which unnecessary edges are removed from the straightened edge image by comparing the straightened edge image that has been linearized with the contraction mask image, and a polygon obtained by tracing the line segment. A means for generating a polygonal coordinate sequence that most overlaps with the mask region, and giving the elevation value to the polygonal coordinate sequence, and extracting the extracted color region. Means for Motoka Precision urban model generation program to function as a.