JP2007187934A - Method for creating electronic map line shape data - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To avoid deviation of line shape data of a road and the like when an electronic map and an aerial photography or the like are superposed. <P>SOLUTION: Line shape data of roads, rivers and the like to be used in an electronic map is created by carrying out Active Shape Model (hereafter referred to as ASM) processing using a dot sequence obtained from a picture such as an aerial photography through imaging processing, and a template. Thus, the burden of an operator is greatly reduced, and high-precision line shape data can be obtained. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention extracts line shape data such as roads, railways, and rivers in an electronic map from image data such as aerial photographs and satellite images, and creates an electronic map line shape data creation method that creates accurate electronic map line shape data. About.

  In recent years, aerial photographs and satellite images can easily grasp the actual position and color from the images, so their uses are general maps, city planning, distance measurement, land surveying, navigation equipment, area marketing, water supply and sewerage facilities Management, farmland / water use, learning / education, advertising, construction / construction planning, etc. Recently, with the appearance of high-resolution ortho images, etc., the accuracy of buildings, roads, railways, terrain, river shapes, colors, locations, etc. has improved dramatically. It is easily expected that applications will expand.

  On the other hand, many maps have been used for the above-mentioned purposes before aerial photographs exist. At present, with the advent of electronic maps, it is possible to have a lot of information on the map, so that it is widely spread mainly as a navigation device or a map on the Internet. For electronic maps, it is important to improve the accuracy of buildings, roads, railways, topography, river shapes and locations, and various researches are being conducted every day.

  Furthermore, aerial photographs and electronic maps are common in that they present geographical information, and each has its own different characteristics, so depending on the application, both aerial photographs and electronic maps can be used. It is desirable to require features and be able to use them simultaneously. In addition, by using high-precision GPS and the like together with these, it can be expected to be used for more applications.

  However, when the satellite image and the electronic map are compared at the most used scale 1/25000, for example, the position of the road data in the electronic map includes some error. For this reason, it is very difficult to use these aerial photographs and the electronic map at the same time in applications requiring accuracy. Therefore, further improvement in accuracy is required for the electronic map. Therefore, if it is possible to easily create highly accurate line shape data from images such as aerial photographs by simply performing image processing of these images, the problem of correcting an electronic map can be easily solved. it can. If the creation of line shape data in an electronic map can be automated, the burden on the operator can be reduced.

  Here, as a method of creating line shape data such as a road by performing image processing from an image such as an aerial photograph, a technique using a B-snakes method is known. This B-snakes method is a kind of dynamic contour model, and semi-automatically by applying a contour line of line shape data such as a road obtained by image processing such as an aerial photograph to a B-spline curve. In this method, a contour line is extracted. This method is a simple process in that line shape data can be obtained from an image such as an aerial photograph by a very simple process of repeating the step of approximating the contour obtained by image processing with a B-spline curve. It can be said that this is an excellent method in that the processing speed of calculation is fast and it can be easily put into practical use.

Non-Patent Document 1 relates to a method for creating road line shape data from an aerial image or the like using the B-snakes method. The operator designates a plurality of control points and processes them so as to pass through the control points, thereby interactively extracting the contour of the line shape data of the road and applying the contour to the B-spline curve to obtain the line shape data. Has been shown to be possible.
Armin Gruen and Haihong Li, "Road extraction from aerial and satellite images by dynamic programming", ISPRS Journal of Photogrammetry and Remote Sensing Vol.50, No.4, 1995, pp.11-20.

  As described above, when an image obtained from an aerial photograph or the like is used in the conventional B-snakes method, an outline of road shape data such as a road obtained by image processing such as an aerial photograph is obtained. It is possible to generate the line shape data semi-automatically by applying to the B-spline curve. However, when the conventional B-snakes method is used, the operator must designate an initial control point sequence, and in a case where it is desired to create line shape data having a complicated network over a wide range. Therefore, a large amount of labor is required, and there is a problem that practical implementation is difficult.

  The B-snakes method uses only the smoothness of the curve as a limiting condition. Under this limiting condition, the line shape data that does not resemble the required line shape data when the contour line is freely deformed repeatedly. May result in Therefore, in order to obtain highly accurate line shape data using the B-snakes method, it is required that the initial contour position accuracy, which is the initial control point sequence, is high. For example, in order to reduce the load of operator control point designation, when line shape data is created using an outline extracted from an existing electronic map as an initial outline, an initial outline with low position accuracy must be used. Become. Then, in order to improve the position accuracy of the line shape data to be created, it must be repeatedly repeatedly deformed, and as a result, the created shape will be different from the required shape, eventually aerial photography etc. There is a problem that the line shape of the electronic map and the line shape data of the electronic map are shifted. For this reason, it is impossible to achieve simultaneously the original purpose of aerial photography and the like and a highly accurate electronic map.

  In view of these problems, the present invention creates a line shape data from an image such as an aerial photograph using Active Shape Model (hereinafter, ASM), which is one of the active contour creation methods, thereby greatly reducing the burden on the operator. It is an object of the present invention to provide an electronic map line shape data creation technique that can obtain highly accurate line shape data in both position and shape.

The first solving means of the present invention is:
A line shape data creation processing device for creating line shape data in an electronic map from image data,
An image database storing the image data including an image obtained by photographing the ground from above and position coordinate information of the image;
Among the image data, a receiving unit that receives an input range of the image data that is a target of the line shape data creation process;
A target image acquisition unit that acquires the range received by the reception unit from the image data stored in the image database;
A template acquisition unit that acquires the template represented by the position coordinates used for the line shape data creation processing at the corresponding position coordinates on the target image;
An extraction contour creating unit that creates an extraction contour using the point sequence obtained by image processing the target image obtained by the target image obtaining unit, and the template obtained by the template obtaining unit;
An ASM processing unit that creates a line-shaped contour by ASM processing using the extracted contour created by the extracted contour creating unit and the template obtained by the template obtaining unit;
It can be configured as a line shape data creation processing device having a line shape data extraction unit that extracts the line shape data from the line shape outline obtained by the ASM processing unit.

The second solution of the present invention is as follows:
Using an image database storage device that stores image data including an image taken from above the ground and position coordinate information of the image,
A line shape data creation processing method for creating line shape data in an electronic map from the image data,
An accepting step of accepting an input range of the image data to be subjected to the line shape data creating process among the image data;
A target image acquisition step of acquiring the range received in the reception step from the image data stored in the image database;
A template acquisition step of acquiring a template represented by the position coordinates used for the line shape data creation processing at the corresponding position coordinates on the target image;
An extraction contour creating step for creating an extraction contour using the point sequence obtained by image processing the target image obtained in the target image obtaining step, and the template obtained in the template obtaining step;
An ASM processing step of creating a line shape contour by ASM processing using the extracted contour created in the extracted contour creation step and the template obtained in the template acquisition step;
It can be configured as a line shape data creation processing method including a line shape data extraction step for extracting the line shape data from the line shape outline obtained in the ASM processing step.

The third solution of the present invention is:
Using an image database storage device that stores image data including an image taken from above the ground and position coordinate information of the image,
A line shape data creation processing program for creating line shape data in an electronic map from the image data,
Among the image data, a reception subprogram that receives an input range of the image data that is a target of the line shape data creation process;
A target image acquisition subprogram for acquiring the range received by the reception subprogram from the image data stored in the image database;
A template acquisition subprogram for acquiring a template represented by the position coordinates used for the line shape data creation processing at the corresponding position coordinates on the target image;
An extraction contour creation subprogram for creating an extraction contour using a point sequence obtained by image processing the target image obtained by the target image acquisition subprogram and the template obtained by the template acquisition subprogram When,
An ASM processing subprogram for creating a line shape contour by ASM processing using the extracted contour created by the extracted contour creation subprogram and the template obtained by the template acquisition subprogram;
A line shape data creation processing program having a line shape data extraction subprogram for extracting the line shape data from the line shape contour obtained by the ASM processing subprogram can be configured.

  “ASM” is a kind of dynamic contour model as in the B-snakes method, but highly accurate line shape data can be created by using a contour created based on an image such as an aerial photograph and a template. Since ASM uses the template as the initial contour, it is not necessary for the operator to specify each control point sequence. In addition, since the template plays a supervising role in position, highly accurate line shape data can be obtained even if it is repeatedly freely deformed.

  As described above, ASM is characterized in that highly accurate line shape data in both position and shape can be dynamically obtained even if a template as an initial contour is repeatedly deformed until it is fitted to an image. In addition to translation, rotation, and scaling of a given template contour to the required position, the control points can be freely deformed to flexibly fit the same way as the B-snakes method. Can also be done.

  Here, “line shape data” desired to be generated in the present invention refers to data represented by lines used in an electronic map. For example, it means outer frame line shape data and center line shape data of roads, railways, rivers, polygon data of buildings and houses, outer shape data of land, and the like.

  The “template” used in the present invention is, for example, a sequence of points controlling a B-spline when linear data is approximated by a B-spline curve, or a sequence of points prepared in advance. A point sequence having position coordinate information indicated by latitude and longitude. It also plays a supervising role in the shape of the line shape data desired to be created by ASM processing. This template can be said to be one of the most characteristic of the differences between the ASM and the B-snakes method.

  In the present invention, when considering what to use for the template used for ASM processing, using the line shape data in the existing electronic map stored in the map database, It can be used as it is. Therefore, the embodiment of the present invention shows a method of creating line shape data by fitting existing electronic map data to an image such as an aerial photograph using ASM.

  However, the road shape is often designed in consideration of clothoid curves, etc., and by using these road design features, a template created in advance is prepared in the template database. It is also possible to use a necessary template each time. At that time, the position coordinate information of each point is set relative to each template point, and the template can be acquired at the required position coordinates in the image data by inputting the reference coordinates. It is. In addition, by preparing templates for road intersections, railway line shapes, and the like in the template database in advance, it is possible to deal with various forms of line shape data.

In the present invention, a B-spline curve is defined by a weighted sum of B-spline basis functions. The following four properties are required for the B-spline basis function.
(1) Finite support: The nth B-spline basis function B _{n, d} (s) is zero except for [n, n + d] (n is the order of the B-spline basis function).
(2) Positive value: The B-spline basis function is positive (B _{n, d} (s)> 0).
(3) Normalization: The sum of all B-spline basis functions in all s is 1 (ΣB _{n, d} (s) = 1).
(4) Invariance regarding translation: B _{n + 1, d} (s) = B _{n, d} (s-1)

In order to express the contour of an object, it is general to use a spline curve defined by a piecewise quadratic or cubic polynomial. In particular, the B-spline curve r (s) = (x (s), y (s)) ^T , 0 < s < L is the number of N basis functions Bn (n = 0,1, .., N-1). The weighted sum is expressed as the following equation (Equation 1).

The base of the default function is defined according to a sequence of nodes called knots. By using overlapping node positions called multiple knots, line shape data including differential discontinuities such as joints between straight lines and curves can be expressed. It becomes possible.

  “Extracted contour” refers to a point sequence of intermediate points obtained by linearly complementing a template with a linear center point sequence or outline point sequence such as a road obtained by image processing from an aerial image or the like. When the extracted contour is approximated by a B-spline, it is also a point sequence that controls the B-spline curve. As an image processing method, for example, a feature image of a target image is created, and a line sequence is applied to the feature image to obtain a point sequence. The image processing method is not limited to the above-described method, and various methods are conceivable. However, any method is affected by the shadow of the building in the image and other obstacles, so it is impossible to obtain highly accurate line shape data only by this method.

Next, ASM will be described in detail. First, let S be a space of a B-spline curve formed by N control points. Let S _{S be} a partial space in which only the shapes obtained by translating the template Q ₀ εS, which is the initial contour, with the rotation are collected. This deformation can be expressed linearly by _four parameters X = (x ₁ , x ₂ , x ₃ , x ₄ ) ^{T as} in the following equation (Equation 2). Here, this X is called a shape space.

The problem of applying the extracted contour Q _f εS obtained from the template, such as image processing, to the template Q ₀ so as to be the least square of the norm is to project from the space S to the subspace S _S. Actually, the difference in distortion at the position between the line shape data as a template (for example, the line shape data in the electronic map) and the line shape such as a road in an image such as an aerial image is only the rotation and translation described above. It is difficult to think that it can be corrected by. Therefore, determining the contour Q∈S as allowing some degree of deformation when fitting the template to the extracted contour Q _f. In this case, a contour that minimizes the sum of the error when converted into an invariable amount between rotation and translation and the error of the contour itself as in the following equation (Equation 3) may be obtained.

This estimated solution is expressed by the following equation (Equation 4).

  Α is a parameter for adjusting two error terms. In ASM, by adjusting this parameter to an optimal value, highly accurate line shape data can be obtained. This parameter can be set in advance as a default value for ASM processing. Various setting methods are conceivable as setting methods, such as setting a uniform maximum likelihood value, setting different values in the city center and the suburbs, and setting different values depending on the type of line shape. This parameter can also be input by the operator for each work.

  As described above, the degree of template fitting can be adjusted by parameters, and in addition to the smoothness of the curve limited by the B-snakes method, the shape can be limited by the template. From this, ASM can be said to be a flexible model having both free deformation and shape suppression by template matching.

  By creating line shape data from images such as aerial photographs using ASM, which is one of the active contour creation methods, it is possible to greatly reduce the burden on the operator and obtain highly accurate line shape data. .

Embodiments of the present invention will be described in the following order. In the present embodiment, a mode is shown in which a template is created from line shape data in an electronic map using an existing electronic map, and this is used as an initial contour to create highly accurate line shape data. However, according to the present invention, by inputting a template prepared in advance on image data such as an aerial photograph by an operator, line shape data of a new electronic map can be created without using the line shape data in the electronic map. Is also possible.
A. Device configuration:
B. Template creation process:
C. Feature image creation process:
D. ASM processing:

A. Device configuration:
FIG. 1 is an explanatory diagram showing a configuration of a line shape data creation processing apparatus. The line shape data creation processing apparatus 1 of the present embodiment is configured by connecting a server 2, a server 3, and a server 4 via a network NET. The network NET is a network using wireless communication, and may be limited such as a LAN or an intranet, or may be a wide area such as the Internet.

  The server 2 is a device that performs ASM processing for creating line shape data and other processing necessary for ASM processing, and each processing is performed by the CPU of the server 2. In addition, the server 2 is connected to the input device 5 and can receive various instructions for receiving a command input from the input device 5 so that the CPU of the server 2 can perform various processes for creating highly accurate line shape data. .

  The input device 5 includes an input range of image data to be subjected to line shape data creation processing, line shape data used to create a template from line shape data stored in the map database 30, parameters used in ASM processing, And other necessary parameters when inputting necessary parameters. Examples of the input device 5 include a keyboard, a mouse, and a pen tablet.

  The server 3 has a map database 30 and is accessed from the server 2 in order to acquire line shape data used for creating a template when the template is created.

  The server 4 has an image database 40, and is accessed from the server 2 in order to acquire image data to be processed for line shape data creation when creating an extracted contour.

  FIG. 2 is an explanatory diagram showing a functional configuration of the line shape data creation processing apparatus 1. The server 2 includes a reception unit 21, a target data acquisition unit 22, a template generation unit 23, a template acquisition unit 24, a target image acquisition unit 25, an extraction contour generation unit 26, an ASM processing unit 27, and a line shape data extraction unit 28. Yes. Each functional block in the drawing is configured by software by a computer program executed by the CPU of the server 2, but may be configured by hardware.

  The map database 30 is a server that stores an electronic map, and stores necessary data divided into layers such as a background layer, a feature layer, and a character layer. The background layer is a layer that records the shape of a water system such as a river or land. The feature layer is a layer for recording polygon data and attribute data of features such as roads, railways, and buildings. The character layer is a layer that stores character information to be displayed on the map. Each layer is further divided into fine layers and recorded. In the present invention, at least roads and railways having a line shape in the electronic map corresponding to the line shape in the image data to be subjected to the line shape data creation process, and position coordinate information indicated by the latitude and longitude thereof, Line shape data such as rivers must be stored.

  The image database 40 stores an image obtained by photographing the ground such as an aerial image from the sky and position coordinate information indicated by the latitude and longitude of the image. In this embodiment, the case where the image is stored in the server 4 as a database is taken as an example. However, the image may be stored in a storage medium 6 such as a CD-ROM, DVD, floppy disk (registered trademark), or MO. .

  The receiving unit 21 uses the line shape data used to create a template from the input range of the image data to be processed by the input device 5 and the line shape data stored in the map database 30. , Processing for accepting input of parameters used in ASM processing and other necessary parameters.

  The target data acquisition unit 22 performs processing for acquiring the line shape data received by the reception unit 21 from the line shape data stored in the map database 30. A template used for ASM processing is created by the template creation unit 23 using the line shape data acquired here.

  The template creation unit 23 performs processing for creating a template from the line shape data obtained by the target data acquisition unit 22. The template created here must be a point sequence in which each point has position coordinate information indicated by latitude and longitude. In this embodiment, ASM processing is performed using the template created here. This template processing will be described in detail later with reference to the drawings.

  The template acquisition unit 24 performs processing for acquiring the template represented by the position coordinates used for the line shape data creation processing at the corresponding position coordinates on the target image acquired by the target image acquisition unit 25. In this embodiment, a template created from line shape data in an electronic map is used. However, a template created in advance in the template database can also be used.

  The target image acquisition unit 25 performs a process of acquiring the input range of the image data to be subjected to the line shape data creation process received by the reception unit 21 from the image data stored in the image database 40.

  The extracted contour creating unit 26 uses the point sequence obtained by image processing such as applying a filter to the feature image obtained from the target image obtained by the target image obtaining unit 25 and the template obtained by the template obtaining unit 24. A process for creating an extracted contour is performed. The extracted contour creation process will be described in detail later with reference to the drawings.

  The ASM processing unit 27 performs a process of creating a line shape contour by ASM processing using the extracted contour created by the extracted contour creating unit 26 and the template obtained by the template obtaining unit 24. The concept of the ASM processing is as described above, and the implementation procedure will be described later with reference to the drawings.

  The line shape data extraction unit 28 forms a B-spline curve from the point sequence of the line shape outline obtained by the ASM processing unit 27 and performs a process of extracting the line shape data. By this processing, highly accurate line shape data used in the electronic map can be obtained, and by correcting the line shape data in the electronic map, the electronic map and the aerial photograph can be used simultaneously.

B. Template creation process:
FIG. 3 is a flowchart of the template creation process. In the present embodiment, a method of acquiring line shape data from an electronic map and creating a template from the line shape data is shown. However, in the present invention, it is also possible to use a template prepared in advance.

  In this process, based on the input from the input device 5, the target data acquisition unit 22 acquires the line shape data received by the reception unit 21 of the server 2 from the line shape data stored in the map database 30. This is a process in which the template creation unit 23 creates a template from the shape data, and is a process executed by the CPU of the server 2 in terms of hardware.

First, an operator inputs a range or line shape data on an electronic map for which a template is to be created from the input device 5. (Step S1)
Next, the reception unit 21 receives the input. (Step S2)
The target data acquisition unit 22 acquires line shape data stored in the map database 30 based on the input received by the reception unit 21. (Step S3)

Equally dividing a line shape data obtained by the target data acquisition unit 22 at a fine interval d _1. (Step S4)
Let this coordinate sequence be {r [n]}. (Step S5)
This coordinate sequence {r [n]} is used for the process of extracting the bending point.

  Next, the degree of curve bending (κ [n]) is calculated by the following equation (Equation 6) at each point divided in step S4.

この計算により、予め設定された閾値を超えて、かつ曲線の曲がり具合（κ[n]）が極大となる点を抽出する。（ステップＳ６）
Ｓ７で抽出した点を屈曲点とする。（ステップＳ７）

By this calculation, a point that exceeds a preset threshold and has a maximum curve bending degree (κ [n]) is extracted. (Step S6)
The point extracted in S7 is set as a bending point. (Step S7)

The start point, the end point, and the bending point are divided at an interval d ₂ (> d ₁ ). (Step S8)
With the above processing, the position of the multiple knots, which are the knot positions overlapping the number of control points to be represented by the B-spline curve, are determined.

Subsequently, a B spline curve r (s) that approximates a broken line composed of the point sequence obtained in the above step, and the point sequence Q ^X = {q _n ^x } and Q ^Y = Find {q _n ^y }. (Step S9)
Generally, in order to determine a control point sequence Q as L ₂ norm is minimized for any curve f (s), it may be calculated the following equation (7).

ここで、Ｂ(s)はＢスプライン基底関数を並べたベクトル（Ｂ_０（ｓ），Ｂ_１（ｓ），…，Ｂ_Ｎ−１（ｓ））^ＴでＢはi行j列の要素が、次式（数８）で与えられる正値対称な計量行列である。

Here, B (s) vector obtained by arranging B-spline basis functions _{(B 0 (s), B} 1 (s), ..., B N-1 (s)) in ^T B is an element of row i and column j , A positive symmetric metric matrix given by the following equation (Equation 8).

  Now, since the curve to be approximated is a broken line consisting of a sequence of points, the simple method of calculating the following equation (Equation 9) is to replace the calculation with a sequence of M points sampled as follows, but the integration error is The start point and end point position of the B-spline curve that is greatly approximated may deviate greatly.

従って、近似したい点列を線形もしくは２次のスプライン曲線で補間した後に式（数７）の積分計算を行なう。
上記の結果、得られたＢスプラインを制御している点列をテンプレートＱ_０として用いる。（ステップＳ１０）

Accordingly, after the point sequence to be approximated is interpolated with a linear or quadratic spline curve, the integral calculation of Expression (7) is performed.
The above results, using the point sequence that controls the resultant B-spline as the template Q _0. (Step S10)

C. Extraction contour creation process:
4 and 5 are flowcharts from the extraction contour creation process to the ASM process. The extracted contour is created from an image such as an aerial image and a point sequence obtained by image processing and a template. For example, image processing is performed such that a feature image of the target image is created, a line filter is applied to the feature image, and a point sequence that maximizes the filter response (in this case, a linear center point sequence) is extracted. A point sequence of intermediate points obtained by linearly complementing the point sequence obtained from this image processing and the template is obtained. When the obtained point sequence is approximated by a B-spline, the point sequence that controls the B-spline curve can be used as an extraction contour. The image processing used in the process of the extraction contour creation process is not limited to the above, and various methods are conceivable. In this embodiment, a method for creating an extraction contour using the above-described method will be described.

  In this processing, based on the input from the input device 5, the input range of the image data that is the target of the input line shape data creation process received by the receiving unit 21 of the server 2 is displayed in the image database. The target range is acquired from the image data stored in 40. The extracted contour creating unit 26 is a process for creating an extracted contour using the point sequence obtained by performing image processing on the target range and the template acquired by the template acquiring unit 24, and is executed by the CPU of the server 2. It is processing to do.

  First, a method for creating a feature image for performing line fitting will be described. A feature image is obtained by creating an image having a lightness value as a Mahalanobis distance between a mean vector and a covariance matrix.

First, the operator inputs an input range of image data to be subjected to line shape data creation processing from the input device 5. (Step S11)
Next, the receiving unit 21 receives an input of an input range of image data. (Step S12)
Next, the target image acquisition unit 25 acquires the target range from the image data stored in the image database 40. (Step S13)
A sample is cut out from the obtained image data of the target range. (Step S14)
Calculate the mean vector μ and the covariance matrix Σ on the feature space. (Step S15)

  Here, the following items are considered as preconditions for creating a feature image. When image processing is performed on a road in an image such as an aerial photograph and a feature image is created, pixels on the road surface are around an average vector μ, assuming that the road occupies a large area with a uniform color. And follow a multidimensional normal distribution of the variance matrix Σ. Pixel features are represented by RGB vectors. θ = (μ, Σ) is referred to as a road pixel feature distribution. μ and Σ are set by a map creator in advance by giving a road teacher image, and are used as evaluation criteria for matching in road tracking. In the road image data, white lines and display areas are not taken into consideration, but the ratio of these areas occupying the road is sufficiently smaller than the pavement surface and can be ignored.

  Here, the “feature space” is a vector that characterizes the property of the substance distributed in the pixel. Each pixel of the optical sensor data observed in n bands can be considered as an n-dimensional vector quantity having the density value of each band as a component. When applying various classification methods, we focus on the distribution characteristics of these data in the n-dimensional space. The classification method can be applied more easily by performing pre-processing such as ratio calculation processing that facilitates detection of features such as reflectance and luminance that are of interest and converting the feature space.

“Covariance” means that when an event is observed by two variables x and y, a set of n observations (x ₁ , y ₁ ), (x ₂ , y ₂ ),. For (x _n , y _n ), the one defined by the following equation is called the covariance c _xy of x and y.

式（数１０）の(１)、(２)の差は共分散の統計的な推定方法の違いに起因するが、得られる値をそれぞれ不偏推定値、最尤推定値という。 nが十分大きい場合は両者に実質的な差はない。共分散は(x_i−m_x)と(y_i−m_y)の積の平均であるから、 xとyの変動傾向の類似性の指標と考えることができる。その場合、共分散はxとyの尺度に影響されるので、それぞれの標準偏差σ_x、σ_yで正規化した次式（数１１）の値c_xy’が使われることも多い。

Although the difference between (1) and (2) in the equation (Equation 10) is due to the difference in the statistical estimation method of the covariance, the obtained values are referred to as the unbiased estimated value and the maximum likelihood estimated value, respectively. When n is sufficiently large, there is no substantial difference between the two. Since the covariance is the average of the product of (x _i −m _x ) and (y _i −m _y ), it can be considered as an index of the similarity of the variation tendency of x and y. In this case, since the covariance is affected by the scales of x and y, the value c _xy ′ of the following equation (Equation 11) normalized by the standard deviations σ _x and σ _y is often used.

ここにc_ｘｙ’をxとyの相関係数という。

Here, c _xy ′ is called a correlation coefficient between x and y.

Next, the feature vector of each pixel of the input image, the Mahalanobis distance between the mean vector μ covariance matrix Σ creating an image I _m to brightness values. (Step S16)

  Here, the “Mahalanobis distance” is a distance scale between the normal population and the sample, and the square of the Mahalanobis distance M is defined by the following equation (Equation 12).

  If the correlation matrix is R, it is also given by the following equation (Equation 13).

  When the covariance matrix Σ in Equation (12) is considered as a unit matrix, the Mahalanobis distance M is equal to the Euclidean distance. Therefore, the Mahalanobis distance can be interpreted as a distance considering the variance of the population. In statistical classification of images, it is often used as a similarity to a classification class of a classification object (for example, pixel) having a feature vector p.

Next, scaling the image I _m obtained in step S16 in [0,255], the image subjected to negative processing the input feature image I _f. (Step S17)
The above is the process of creating a feature image.

  Next, a method for applying a filter to the characteristic image obtained above will be described.

Template acquiring unit 24 acquires the corresponding position coordinates of the input feature image I _f of the template S17. (Step S18)
FIG. 6 is a diagram illustrating a state in which a road data template is acquired as an aerial photograph. The white line in the figure represents a curve obtained by approximating the initial contour with a B-spline. Although the template is represented by a sequence of points, it is indicated by a curve approximated by a B-spline in consideration of easy viewing.
A point sequence obtained by sampling a curve r (s) obtained by approximating a template by B-spline with respect to a parameter s at an interval Δs is defined as {r _n }. (Step S19)
A search range of length l _s is provided in the normal direction. (Step S20)
Next, a line filter is applied to the image _If with each normal direction of {r _n } as an axis. (Step S21)
A point sequence {r ′ _n } that maximizes the filter response is obtained. (Step S22)
FIG. 7 is a diagram illustrating a state where the road center position is extracted by the line filter. The x mark in the figure represents a point sequence obtained from image processing. The white line in the figure represents a curve obtained by approximating the initial contour with a B-spline, and the line drawn in the normal direction represents the filter search range.

  Image processing using various filters is called filtering, and is an extension to a two-dimensional image that is performed by one-dimensional temporal waveform processing. In the same way as high-frequency and low-frequency frequencies are emphasized by filters for audio signals, etc., high-pass filters (high-pass), band-pass filters (band-pass), and low-frequency filters are also applied to image data. Various filtering such as a pass filter (low-pass) can be performed. Here, the frequency in the case of an image is a concept called a spatial frequency that focuses on the spatial spread of grayscale information.

  Filtering in the frequency domain (u, v) is performed with the filter function H (u, v) on the Fourier transform F (u, v) of the image data f (x, y) in the spatial domain (x, y). This can be achieved by taking the product. In other words, the filtered result G (u, v) is

と表される。最終的な画像は、G(u,v)の逆フーリエ変換g(x,y)として得られる。

It is expressed. The final image is obtained as an inverse Fourier transform g (x, y) of G (u, v).

  There is a high boost filter that not only multiplies a filter function of a certain band but also can effectively enhance the high frequency component by adding the original image to the difference between the suppressed high frequency component and the original image. In addition, in order to correct an image including blur, an inverse filter that uses the inverse of the Fourier transform of the blur point spread function as a filter function, or a mean square error between the original image and the corrected image is minimized. A Wiener filter defined as follows is used.

  Filtering can also be performed in the spatial domain by a convolution operation of the inverse Fourier transform h (x, y) of the filter function. In particular, when h (x, y) has a value in a limited region, filtering is often performed in the spatial region. There are a local average filter and a Gaussian filter for smoothing as a result of local product summation such as convolution. Further, operations other than product-sum operation include a median filter for smoothing and a percentile filter for edge detection.

Normally, in the active contour model, a correlation filter or an edge filter is used to calculate the extracted contour Q _f , but in this embodiment, since the cross-sectional view of the luminance or feature value of the object is a line, A box filter h (u) defined by (Equation 15) is used.

  Here, H (u) is a step function defined as a function indicating 1 if X is positive and 0 if negative, and w is the width of the target road. Subsequently, a Gaussian function (Equation 16) is convolved with this filter for noise reduction. Smoothing is performed by the above method.

Finally, a method for obtaining the extracted contour by calculation will be described.
First, calculate the r _n and _r'n as linear interpolation with intermediate point ^{_{r * n = r n + β}} (r'n -r n). (Step S23)

FIG. 8 is a diagram illustrating an intermediate point obtained by linearly complementing a template and a road center point sequence. Circles in the figure represent intermediate points obtained by linearly complementing the template and the road center point sequence. The white line in the figure represents a curve obtained by approximating the initial contour with a B-spline, and the thin line represents a curve obtained by approximating a point sequence obtained from image processing with a B-spline. The line drawn in the normal direction represents the filter search range.
Then, the point sequence {r ^* _n } is approximated by a B-spline curve, and the point sequence controlling the B-spline curve is set as an extraction contour Q _f . (Step S24)
The above is the process for creating the extracted contour.

D. ASM processing:
The ASM processing unit is a process for applying the template obtained by the template acquisition unit 24 and the extracted contour obtained by the extracted contour creating unit 26 to ASM, and is a process executed by the CPU of the server 2.

And a template _{Q 0} and extracted contour Q _f, calculating the equation (4). (Step S25)
Q is obtained by step S25.

で置き換える。（ステップＳ２６）
探索範囲を制御するパラメータlをl_s←γl_s（０＜γ＜１）と減少させて、ステップＳ２０からステップＳ２６を指定された所定の回数繰り返す。（ステップＳ２７）
以上が、ＡＳＭ処理である。

Replace with. (Step S26)
The parameter l for controlling the search range is decreased to l _s γ 1 _s (0 <γ <1), and steps S20 to S26 are repeated a predetermined number of times. (Step S27)
The above is the ASM processing.

  The line shape data extraction unit 28 was finally obtained.

からＢスプライン曲線を構成し、線形状データを抽出する処理を行なう。以上の処理により、精度の良い線形状データを作成することができる。

A B-spline curve is constructed from the above, and processing for extracting line shape data is performed. Through the above processing, accurate line shape data can be created.

  FIG. 9 is a diagram showing the results of creating line shape data using the ASM and B-snakes methods. (1) of FIG. 9 is a result using ASM, and (2) is a result using B-snakes method. The white line in the figure represents a curve obtained by approximating the initial contour with a B-spline, and the black line represents the created line shape data. In addition, the x mark in the figure represents a contour point sequence obtained as a result of repeating each process. In this figure, it can be seen that when ASM is used, highly accurate line shape data can be obtained even in a complex network including multiple knots. Note that this figure is processed by removing some glandular shapes in consideration of easy viewing.

  As mentioned above, although the typical Example of this invention was described, it cannot be overemphasized that this invention is not limited to this Example, and can take a various structure in the range which does not deviate from the meaning.

It is explanatory drawing which shows the structure of a line shape data creation processing apparatus. It is explanatory drawing which shows the function structure of a line shape data creation processing apparatus. It is a flowchart of a template creation process. It is a flowchart from extraction outline creation processing to ASM processing. It is a flowchart from extraction outline creation processing to ASM processing. It is a figure which shows the state which acquired the template of road data in the aerial photograph. It is a figure which shows the state which extracted the road center position by the line filter. It is a figure which shows the intermediate point which linearly complemented the template and the road center point sequence. It is a figure which shows the result of having produced the line shape data using the ASM and B-snakes method.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Line shape data creation processing apparatus 2 Main server 3 Map database server 4 Image database server 5 Input device 6 Storage medium 21 Reception part 22 Target data acquisition part 23 Template creation part 24 Template acquisition part 25 Target image acquisition part 26 Extraction outline creation part 27 ASM processing unit 28 Line shape data extraction unit 30 Map database 40 Image database 50 Input means

Claims (5)

A line shape data creation processing device for creating line shape data in an electronic map from image data,
An image database storing the image data including an image obtained by photographing the ground from above and position coordinate information of the image;
Among the image data, a receiving unit that receives an input range of the image data that is a target of the line shape data creation process;
A target image acquisition unit that acquires the range received by the reception unit from the image data stored in the image database;
A template acquisition unit that acquires the template represented by the position coordinates used for the line shape data creation processing at the corresponding position coordinates on the target image;
An extraction contour creating unit that creates an extraction contour using the point sequence obtained by image processing the target image obtained by the target image obtaining unit, and the template obtained by the template obtaining unit;
An ASM processing unit for creating a line-shaped contour by an Active Shape Model (hereinafter, ASM) process using the extracted contour created by the extracted contour creating unit and the template obtained by the template obtaining unit;
A line shape data creation processing device comprising: a line shape data extraction unit that extracts the line shape data from the line shape outline obtained by the ASM processing unit. The line shape data creation processing device according to claim 1,
A map database storing line shape data having at least the line shape in the electronic map corresponding to the line shape in the image data to be subjected to the line shape data creation processing and the position coordinate information;
The reception unit according to claim 1,
The accepting unit according to claim 1, which accepts an input of the line shape data used for creating the template from the line shape data stored in the map database;
A target data acquisition unit that acquires the line shape data received by the reception unit from the line shape data stored in the map database;
A template creation unit that creates the template from the line shape data obtained by the target data acquisition unit,
The map database correction device according to claim 1, wherein the template according to claim 1 acquired by the template acquisition unit according to claim 1 is the template created by the template creation unit. The line shape data creation processing device according to claim 1 or 2,
The accepting unit according to claim 1 or 2 accepts an input of a parameter required for the ASM processing,
The line shape data creation processing device according to claim 1, further comprising: an ASM processing unit according to claim 1, wherein the line shape contour is created by the ASM processing based on the parameter received by the reception unit. Using an image database storage device that stores image data including an image taken from above the ground and position coordinate information of the image,
A line shape data creation processing method for creating line shape data in an electronic map from the image data,
An accepting step of accepting an input range of the image data to be subjected to the line shape data creating process among the image data;
A target image acquisition step of acquiring the range received in the reception step from the image data stored in the image database;
A template acquisition step of acquiring a template represented by the position coordinates used for the line shape data creation processing at the corresponding position coordinates on the target image;
An extraction contour creating step for creating an extraction contour using the point sequence obtained by image processing the target image obtained in the target image obtaining step, and the template obtained in the template obtaining step;
An ASM processing step of creating a line shape contour by ASM processing using the extracted contour created in the extracted contour creation step and the template obtained in the template acquisition step;
And a line shape data extracting step of extracting the line shape data from the line shape outline obtained in the ASM processing step. Using an image database storage device that stores image data including an image taken from above the ground and position coordinate information of the image,
A line shape data creation processing program for creating line shape data in an electronic map from the image data,
Among the image data, a reception subprogram that receives an input range of the image data that is a target of the line shape data creation process;
A target image acquisition subprogram for acquiring the range received by the reception subprogram from the image data stored in the image database;
A template acquisition subprogram for acquiring a template represented by the position coordinates used for the line shape data creation processing at the corresponding position coordinates on the target image;
An extraction contour creation subprogram for creating an extraction contour using a point sequence obtained by image processing the target image obtained by the target image acquisition subprogram and the template obtained by the template acquisition subprogram When,
An ASM processing subprogram for creating a line shape contour by ASM processing using the extracted contour created by the extracted contour creation subprogram and the template obtained by the template acquisition subprogram;
And a line shape data extraction subprogram for extracting the line shape data from the line shape outline obtained by the ASM processing subprogram.

Images