JP2009211150A - Extraction device for road network - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To easily extract a road network from a remote sensing image at high speed. <P>SOLUTION: An extraction device for the road network has: a selection processing part 206 selecting endpoints of a prescribed number N of road segments adjacent to each other after performing covering processing and line thinning processing based on the remote sensing image 211; a straight line generation processing part step 207 for generating straight lines in directions of ±θ of an extension direction from the endpoint of each road segment about a prescribed angle θ; an optimization processing part 208 obtaining a point (xmax, ymax) wherein a value obtained by dividing the number of pixels of a binary image corresponding to a line segment connecting each endpoint and a point (x, y) by the number of all the pixels becomes maximum about all the points inside an area surrounded by the generated 2N pieces of straight lines; and a connection processing part 209 generating the line segment connecting each endpoint and the point (xmax, ymax). <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to a technology for creating a road map by extracting a road network from remote sensing images such as satellite images and aerial photographs.

  Creating a road map from a remote sensing image is roughly divided into two processes. The first process is a road area extraction process in which coverage classification is performed on the remote sensing image to extract an area considered to be a road. Usually, in the result of this road area extraction process, some roads are cut by gaps and need to be connected. In particular, there are many cases where the road is cut off at a point where a plurality of roads intersect. The second process is a road networking process in which a road network is generated by filling a gap between a plurality of road areas.

  The shape of the gap to be filled mainly includes I-junction, Y-junction, X-junction, and T-junction. In I-junction, two roads meet at one point. On the Y-junction, three roads meet at one point. On the X-junction, four roads meet at one point. On a T-junction, one road crosses diagonally with another road.

 In the networking method described in Non-Patent Document 1 below, a plurality of templates are prepared in advance by modeling intersections, template matching processing between these templates and remote sensing images is performed, and the best matching is selected and a gap is selected. Fill. In addition, as the invention of the applicant, there are the following Patent Documents 1 to 3, which all disclose a technique for cutting out a road region from an image.

G. Koutani, K. Uchimura and Z. Hu: Road updating from high resolution arial imagery using road intersection model: http://www.photogrammetry.ethz.ch/pitsanulok_workshop/papers/25.pdf Wikipedia: K-average method: http://en.wikipedia.org/wiki/K%E5%B9%B3%E5%9D%87%E6%B3%95 Yukiichi Sakai: Introduction to Digital Image Processing: pp.68-70, pp.87-90 CQ Publishing (2002) Hisae Yanai: Excel Statistics-Practical Multivariate Analysis-: pp.115-117 OM Publishing (2004) JP 2006-23958 A JP 2007-73009 A JP 2007-33931 A

  The problem with the method described in Non-Patent Document 1 is that the amount of calculation becomes large. For example, when processing a Y-junction, it is necessary to perform template matching for each template with different angles between three roads, using the template orientation and position as parameters, and the number of templates x the number of template orientations x templates It requires a calculation amount proportional to the number of positions. In general, the conventional road networking method requires a large amount of calculation.

  An object of the present invention is to extract a road network from a remote sensing image easily and at high speed.

  The present invention is a road network extraction apparatus that performs a process of extracting a road from a remote sensing image, and applies a cover classification to the remote sensing image to obtain a pixel value indicating a road area and a pixel value that is not a road area. A covering classification processing unit that generates a binary image, a thinning processing unit that generates a road segment by thinning the binary image, an adjacent end point of the road segment, or the road segment and the road segment A selection processing unit that selects an end point of an adjacent road segment, and generates a straight line from the end point of each road segment in the direction of ± θ with respect to a predetermined angle θ, or a road segment at a predetermined angle θ. A straight line generation processing unit that generates a straight line from the end point in the direction of ± θ of the extension direction, and all points (x, y) in the region surrounded by the straight line extending in the direction of ± θ In this case, the optimum value is obtained by finding the maximum value (xmax, ymax) obtained by dividing the number of pixels indicating the road region corresponding to the line segment connecting the end point and the point (x, y) by the total number of pixels of the line segment. There is provided a road network extraction apparatus comprising: a processing unit; and a joint processing unit that generates a line segment connecting the end point and the point (xmax, ymax).

  Here, when two road segments are selected, it is determined whether only one of the extension lines intersects the other, and it is determined whether the road is a T-junction or other than a T-junction. .

  Further, the present invention is a road network extraction device that performs a process of extracting a road from a remote sensing image, and applies a cover classification to the remote sensing image to obtain a pixel value indicating a road area and a pixel that is not a road area When selecting two of the road segment, a cover classification processing unit that generates a binary image, a thinning processing unit that generates a road segment by thinning the binary image, and the road segment, A selection processing unit that determines whether or not only one of the extension lines intersects the other, and determines that it is a T-junction if it intersects and selects a predetermined number N of adjacent road segment end points A straight line generation processing unit that generates a straight line from the end point of each road segment in the direction of ± θ in the extension direction at a predetermined angle θ, and all the points (x, y) Optimum to find a point (xmax, ymax) that maximizes the value obtained by dividing the number of pixels indicating the road region corresponding to the line segment connecting each end point and the point (x, y) by the total number of pixels of the line segment A road network extraction apparatus comprising: a processing unit; and a joint processing unit that generates a line segment connecting each end point and the point (xmax, ymax). The thinning process includes a process of deleting a minute road segment. The amount of computation required by this method is proportional to the number of points in the region surrounded by 2N straight lines.

  The present invention is a road network extraction apparatus that performs a process of extracting a road from a remote sensing image, and applies a cover classification to the remote sensing image to obtain a pixel value indicating a road area and a pixel value that is not a road area. Both of the covering classification processing unit that generates a binary image, the thinning processing unit that generates a road segment by thinning the binary image, and two of the road segments are selected. It is determined whether or not they intersect with each other, and when intersecting, it is determined that the road segment is not a T-junction, a selection processing unit that selects a road segment and an end point of a road segment adjacent to the road segment, and a road segment The straight line generation processing unit that generates a straight line from the end point in the direction of ± θ of the extension direction, and the intersection of the two generated straight lines and the other road segment are obtained. The number of pixels indicating the road region corresponding to the point (x, y) on the line segment connecting the two intersections and the line segment connecting the end points is the total number of pixels of the line segment connecting the point (x, y) to the end points And an optimization processing unit for obtaining a point (xmax, ymax) where the value divided by the maximum is obtained, and a joint processing unit for generating a line segment connecting the end point and the point (xmax, ymax) It may be a road network extraction device. The thinning process includes a process of deleting a minute road segment. The amount of computation required by this method is proportional to the number of points on the line segment connecting the two intersections.

  The present invention may also be a method for executing the above processing steps, a program for causing a computer to execute the steps, and a computer-readable recording medium for recording the program.

  According to the present invention, the process for filling the gap between road areas can be performed at a higher speed than the process by template matching or the like because the necessary calculation amount is proportional to the number of points belonging to the definition area.

  Hereinafter, a road extraction technique according to an embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a flowchart showing an outline of road network extraction processing according to the present embodiment. FIG. 2 is a functional block diagram showing an example of a hardware system of a road network extraction device according to this embodiment.

  As illustrated in FIG. 2, the road extraction device according to the present embodiment includes an input device, an output device, a processing device 203, and a storage device 210. As an input device, a mouse 201 and a keyboard 202 are connected to the processing device 203 of the system. The mouse 201 and the keyboard 202 are used for starting commands and inputting parameters. As an output device, for example, a display device 215 is connected to a processing device of the system. The display device 215 is used to display image data, system status, input prompts, and the like.

  The processing device 203 includes a cover classification processing unit 204, a thinning processing unit 205, a selection processing unit 206, a straight line generation processing unit 207, an optimization processing unit 208, and a combination processing unit 209. The storage device 210 includes a remote sensing image storage area 211, a classification image storage area 212, a work binary image storage area 213, and an output binary image storage area 214.

  The cover classification processing unit 204 performs a classification process of an image such as an image classified as a road using the remote sensing image 211 as an input, and outputs the obtained result to the classification image storage area 212. The thinning processing unit 205 applies the thinning process to the area classified as a road, and outputs the road segment to the work binary image storage area 213 and the output binary image storage area 214. The selection processing unit 206 scans the work binary image stored in the work binary image storage area 213, and clusters the end points of road segments adjacent to each other or road segments adjacent to each other. And a pair of road segment endpoints.

  Here, when two nearby road segments are selected, it is determined whether or not only one of the extension lines intersects the other. If Yes, it is determined that the road is a T-junction. If it is No (when both extension lines of the road segment intersect), it is determined that it is other than the T-shaped road (Y-shaped road, X-shaped road, I-shaped road).

  The straight line generation processing unit 207 calculates information on a straight line drawn in the direction of ± θ of the extension direction from the end point of each selected road segment for a predetermined angle θ. The optimization processing unit 208 obtains an optimization definition area on the work binary image 213 from the straight line information obtained by the straight line generation processing unit 207. Next, for all points in the definition area, a value obtained by dividing the number of pixels indicating the road region of the classification image 212 that is a binary image corresponding to the line segment connecting each end point by the total number of pixels of the line segment is Find the maximum point.

  In the present embodiment, the “line-like road-likeness” is a ratio of pixels indicating a road area having a value of 1 to all pixels on the line segment. Generally, based on the sum of pixel values, a line segment passes through a road area, but it is not suitable because it allows an area other than the road to pass through indefinitely.

That is, for example, consider the following two line segment patterns.
L1: 000111010010000
L2: 0011101000
In the sum of the pixel values, L1 is 5 and L2 is 4, and the value of L1 is larger. However, in the “road likeness”, that is, the ratio, L1 is 5/15 = 0.333. 4/10 = 0.4, and the value of L2 is larger. Therefore, it is reasonable to think in proportion.

  The combination processing unit 209 generates a line segment connecting each end point and the maximum point in the output binary image 214. The display processing unit performs control to display on the display 215.

As shown in FIG. 1, in the above processing apparatus, when processing is started (start), the covering classification in step 101 is processed as follows. First, the pixels of the remote sensing image are mapped to the RGB color space. Next, clustering is performed based on the K-average method in the RGB color space (see Non-Patent Document 2 above). That is,
(1) A cluster is randomly assigned to each element xi.
(2) Calculate the center Vj in each cluster based on the allocated elements. The calculation typically uses the average of each assigned element.
(3) Find the Euclidean distance between each xi and each Vj, and reassign xi to the nearest central cluster.
(4) If the allocation of all xi clusters has not changed in the above processing, the processing is terminated. Otherwise, Vj is recalculated from the newly allocated cluster and the above process is repeated.

  Separately, road color information is obtained from the remote sensing image. Based on this color information, what is determined to be a road area from the clustering result is manually extracted using a mouse or the like, the pixel value corresponding to the road area is set to “1”, and the other pixel values A classification image with “0” is generated. This operation can also be performed automatically using image recognition or the like (see Non-Patent Document 2 above).

  In the thinning process in step 102, the binary image obtained in step 101 is raster-scanned sequentially from the upper left, for example, and the surrounding eight pixels surrounding the pixel having the value “1” are examined, and several patterns are obtained. If at least one of the conditions is met, the pixel having the center value “1” is changed to the value “0”. This is repeated until no change occurs (see Non-Patent Document 3 above).

In the selection process in step 103 of FIG. 1, the entire image is searched for a cluster (cluster) having close endpoints. Perform hierarchical cluster analysis with the endpoints of all thin lines as elements. The hierarchical cluster analysis is performed according to the following procedure (see Non-Patent Document 4 above).
(1) Generate a cluster consisting of one element.
(2) One cluster is generated by fusing two shortest clusters that are equal to or less than the threshold.
(3) If fusion does not occur, the process ends. If fusion occurs, return to (2) above.

  Here, the distance is determined based on the longest distance. That is, the distance between two clusters is defined as the largest value among the Euclidean distances between elements selected one by one from each cluster. By the above-described hierarchical cluster analysis, a pair of end points adjacent to each other can be generated (see the above-mentioned comparative patent document 4). If the selection can be made (YES), the process proceeds to step 105. If the selection cannot be made, the process is ended (end).

The optimization process in step 106 in FIG. 1 includes a process for obtaining a definition area surrounded by a straight line and a process for calculation on the obtained definition area. The process for obtaining the domain is performed in the following procedure.
(1) Prepare a two-dimensional array and substitute 0 for each element.
(2) For all end points adjacent to each other, the value of the element of the two-dimensional array surrounded by a set of straight lines starting from the end points is increased by one.

  Assuming that the number of end points is N, as a result of the above processing, the elements of the two-dimensional array whose value is N correspond to the domain.

  In the calculation process on the domain, the road-likeness of a line segment connecting all the points on the domain and the adjacent end points is calculated, and the point on the domain where the road-likeness is maximized is obtained. The road-likeness of the line segment is defined by a value obtained by dividing the number of pixels of the binary image corresponding to the line segment by the total number of pixels of the line segment.

  Next, a road is generated by combining the optimized road / end points (step 107). The processing from step 103 to step 107 is continuously executed, and when there is no end point to be selected (NO in 104), the processing ends (end).

  3A and 3B are diagrams showing an example of road extraction technology according to the first embodiment of the present invention, in which the extraction method is applied to an I-junction. In FIG. 3A (a), the cover classification processing unit 204 displays a binary image obtained as a result of performing the cover classification process on the display screen 301 as shown in Non-Patent Document 2 above. It is a figure and two road area | regions 303a * 303b are shown. Since the road areas 303a and 303b include noise and the like, they are shown as thick areas. FIG. 3A (b) shows a road generated as a result of thinning processing applied to the road regions 303a and 303b subjected to the cover classification processing as described in Non-Patent Document 3 by the thinning processing unit 205. It is a figure which shows segment 305a * 305b.

  Here, when the selection processing unit 206 selects two nearby road segments 305a and 305b, it is No in this case when it is determined whether only one of the extension lines intersects the other. For this reason (corresponding to the case where both extension lines L1 and L4 of the road segment intersect), it is determined that the road segment is other than the T-shaped road (in this case, the I-shaped road). FIG. 3A (c) shows a straight line L2 generated by the straight line generation processing unit 207 from the end points P1 and P2 of the road segments 305a and 305b selected by the selection processing unit 206 for a predetermined angle θ in the direction of ± θ in the extension direction. , L3, and L5 and L6. FIG. 3A (d) shows a region R1 (α, β, γ) surrounded by four straight lines L2, L3, L5, and L6. For all the points (x, y) in the region R1, the optimization processing unit 208 calculates the number of pixels of the binary image corresponding to the line segment connecting each of the end points P1, P2 and the point (x, y). Find the point where the value divided by the total number of pixels in the line segment is maximum. The black circle Pmax in FIG. 3A (e) indicates the point where the value becomes maximum. FIG. 3A (f) shows that the combination processing unit 209 combines the point Pmax having the maximum value with the end points P1 and P2 of the two road segments, and the networking of the road is completed.

  FIG. 3B is a diagram corresponding to FIG. 3A (d) and is a diagram illustrating details of the optimization process. Although the intersection (x, y) exists in the region R1, as an example of optimization, the points P31 and P32 shown in the figure among these intersections will be compared. Comparing the line L31-L32 forming the intersection point P31 with the line L41-L42 forming the intersection point P32, the line L31-L32 indicates the number of pixels corresponding to the region R11 in which the pixel value of the binary pixel is “1”. However, it can be seen that the number of pixels in the region corresponding to the line 41-42 and the region R12 is much larger (L31-L32 >> L41-L42). In this way, for the intersection (x, y) in the region R1, the intersection at which the ratio of the pixels whose binary pixel values corresponding to the lines forming the intersection are “1” at the intersection is the maximum. Pmax can be determined. Note that the corresponding pixel does not necessarily fall within the region R1, as long as the intersection is within the region R1. The fact that the ratio of the pixels having the value “1” is the maximum is based on the idea that it is most certain that the area along the line is an actual road.

  Next, an example of road extraction technology according to the second embodiment of the present invention in which the extraction method is applied to a T-junction will be described with reference to FIG. FIG. 4A (a) is a diagram in which the cover classification processing unit 204 displays a binary image obtained as a result of performing the cover classification process on the display screen 401 as shown in Non-Patent Document 2 above. Two road areas 403a and 403b are shown. FIG. 4A (b) is a diagram illustrating an example of road segments 405a and 405b generated as a result of the thinning processing unit 205 applying the thinning process to the road areas 403a and 403b.

  Here, when the selection processing unit 206 selects two nearby road segments 405a and 405b, it is Yes in this case when it is determined whether only one of the extension lines intersects the other. Therefore, it is determined to be a T-junction (corresponding to the case where one extension line L51 of the road segment intersects the other road segment 405b).

  FIG. 4A (c) shows straight lines L52 and L53 generated by the straight line generation processing unit 207 in the direction of ± θ in the extension direction from the end point (P50 in the figure) of each road segment at a predetermined angle θ. A dotted line L51 is a line indicating the extending direction from the end point P50 of the road segment.

  The thick line region R51 in FIG. 4A (d) obtains intersections P51 and P52 between the two generated straight lines L52 and L53 and the other road segment 405b, and a line segment connecting these two intersections P51 and P52. It is.

  The optimization processing unit 208 calculates the number of pixels of the binary image corresponding to the line segment connecting the point on the line segment P51-P52 (R51) and the end point P50 to all the line segments connecting the point and the end point P50. A point Pmax at which the value divided by the number of pixels is maximized is obtained. The black circle Pmax in FIG. 4A (e) indicates the point where the value becomes the maximum. FIG. 4A (f) is a diagram illustrating that the combination processing unit 209 combines the point Pmax having the maximum value with the end point P50 of the road segment, and road networking is completed.

  FIG. 4B is a diagram corresponding to FIG. 4A (d) and shows details of the optimization process. As an example of intersection optimization, when comparing the end point P50 and lines L51 to L54 forming intersections in the line segment region R51, the line L51 is a pixel corresponding to the region R21 in which the pixel value of the binary pixel is “1”. It can be seen that the number is much larger than the number of pixels corresponding to the region R23 in which the pixel value of the binary pixel in the line L54 is “1” (L51 >> L54). In this way, for the intersection (x, y) in the region R21, the intersection at which the ratio of the pixels whose binary pixel values corresponding to the lines forming the intersection are “1” at the respective intersections is the maximum. Pmax can be determined. The fact that the ratio of the pixels having the value “1” is the maximum is based on the idea that it is most certain that the area along the line is an actual road.

  FIG. 5 is a diagram illustrating an example in which road extraction processing is performed on a satellite image using the road network extraction device according to the present embodiment. FIG. 5A is a satellite image obtained by photographing the display area 301 to be processed. This image is 200 × 160 pixels in size. It can be seen that the area indicated by the reference symbol R101 appears to be a road, but is almost indistinguishable. FIG. 5B is an image obtained as a result of applying the cover classification to the satellite image. The road area R102 is represented as a substantially white area R102 in the display area 301, and can be distinguished from the other areas R103. FIG. 5C is a diagram showing road segments L101 and L102 obtained as a result of the thinning process in the region R104. FIG. 5 (d) is a diagram showing roads L101-L103-L104-L102 obtained by combining the points P101 searched by the optimization processing and the end points of the road segments L101 and L102 with line segments L103 and L104. It can be seen that the road network is clearly extracted. The road shown in FIG. 5 is obtained by using processing corresponding to the first embodiment as shown in FIG.

  As described above, according to the road extraction technique according to the embodiment of the present invention, the process of filling the gap between road areas is proportional to the number of points belonging to the definition area. The burden can be reduced and high-speed road extraction processing can be performed.

  The present invention can be used for a road extraction device.

It is a flowchart figure which shows the outline | summary of the extraction method of the road network of this invention. It is a block diagram of a hardware system showing an embodiment of a road network extraction method of the present invention. It is explanatory drawing of the Example which applied the extraction method of the road network by the 1st Embodiment of this invention to I-junction. It is the figure which showed the mode of the optimization process of FIG. 3A in detail. It is explanatory drawing of the Example which applied the extraction method of the road network by the 2nd Embodiment of this invention to the T-junction. It is the figure which showed the mode of the optimization process of FIG. 4A in detail. It is a figure which shows the example which performed the extraction process of the road from the actual satellite image by the road network extraction apparatus by this Embodiment.

Explanation of symbols

201 ... Mouse, 202 ... Keyboard, 203 ... Processing device, 204 ... Cover classification processing unit, 205 ... Thinning processing unit, 206 ... Selection processing unit, 207 ... Linear generation processing unit, 208 ... Optimization processing unit, 209 ... Combination Processing unit 210 ... Storage device 211 ... Remote sensing image 212 ... Classification image 213 ... Working binary image 214 ... Output binary image 215 ... Display

Claims (3)

A road network extraction device that performs a process of extracting a road from a remote sensing image,
A cover classification processing unit that applies a cover classification to the remote sensing image to generate a binary image of a pixel value indicating a road area and a pixel value that is not a road area;
A thinning processing unit for thinning the binary image to generate a road segment;
A selection processing unit that selects the end points of the road segments adjacent to each other, or the end points of the road segment and the adjacent road segments;
A straight line is generated from the end point of each road segment for the predetermined angle θ in the direction of ± θ in the extension direction, or a straight line is generated from the end point of the road segment in the direction of ± θ for the predetermined angle θ. A straight line generation processing unit,
Within all the points (x, y) in the area surrounded by the straight line extending in the direction of ± θ, the number of pixels indicating the road area corresponding to the line connecting the end point and the point (x, y) is the line segment. An optimization processing unit for obtaining a point (xmax, ymax) where the value divided by the total number of pixels is maximum;
A road network extraction apparatus comprising: a connection processing unit that generates a line segment connecting an end point and the point (xmax, ymax). A road network extraction device that performs a process of extracting a road from a remote sensing image,
A cover classification processing unit that applies a cover classification to the remote sensing image to generate a binary image of a pixel value indicating a road area and a pixel value that is not a road area;
A thinning processing unit for thinning the binary image to generate a road segment;
When two of the road segments are selected, it is determined whether only one of the extension lines intersects the other, and if it intersects, it is determined that the road is a T-junction, and a predetermined number N A selection processing unit for selecting end points of road segments adjacent to each other,
A straight line generation processing unit that generates a straight line in the direction of ± θ of the extension direction from the end point of each road segment for a predetermined angle θ;
For all the points (x, y) in the area surrounded by the generated 2N straight lines, the number of pixels indicating the road area corresponding to the line segment connecting each end point and the point (x, y) is determined as the line segment. An optimization processing unit for obtaining a point (xmax, ymax) where the value divided by the total number of pixels is maximum;
A road network extraction apparatus comprising: a joint processing unit that generates a line segment connecting each end point and the point (xmax, ymax). A road network extraction device that performs a process of extracting a road from a remote sensing image,
A cover classification processing unit that applies a cover classification to the remote sensing image to generate a binary image of a pixel value indicating a road area and a pixel value that is not a road area;
A thinning processing unit for thinning the binary image to generate a road segment;
When two of the road segments are selected, it is determined whether or not both intersect each other. If they intersect, it is determined that the road segment is not a T-junction, and the end points of the road segment and the adjacent road segment are determined. A selection processing unit for selecting
A straight line generation processing unit that generates a straight line in the direction of ± θ of the extension direction from the end point of the road segment for a predetermined angle θ;
The intersection of the two generated straight lines and the other road segment is obtained, and the road region corresponding to the point (x, y) on the line connecting these two intersections and the line connecting the end points is determined. An optimization processing unit for obtaining a point (xmax, ymax) having a maximum value obtained by dividing the number of pixels of the binary image shown by the total number of pixels of a line segment connecting the point (x, y) and the end point;
A road network extraction apparatus comprising: a joint processing unit that generates a line segment connecting an end point and the point (xmax, ymax).

Images