JP2007026255A - Method for extracting progress of defective point inside image - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a method for extracting a progress of a defective point such as a crack from two images different in photographed time by image processing technology. <P>SOLUTION: This method comprises: a process for extracting an edge of the image on the basis of first actual image data of a measured face; a process for extracting the edge of the image on the basis of second actual image data after a prescribed time lapses in the same measured face; a difference processing process for finding differences among a specific pixel in the second edge image, a pixel in the first edge image corresponding to the specific pixel, and a peripheral pixel of the pixel, and outputting a smallest value among the differences as a difference value; and a process for outputting the difference found in the difference processing process as the progressive defective point. A straight line part except the defective point is extracted by Hough transformation and is deleted. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a method for extracting the progress of deformation points such as cracks from two images having the same inner wall surface such as a railway tunnel and having different shooting times.

In recent years, coupled with advances in information processing technology, the processing speed of personal computers has increased, the capacity of storage devices has increased, and the speed of networks has increased, so there are more opportunities for image processing technology to be used in various businesses. Yes.
Development technology, which is one of the image processing technologies, can easily understand the internal state and secular change of civil engineering structures such as railway tunnels with a cylindrical inner structure from images. Are being widely used for maintenance work (Patent Document 1).

  Describing the development drawing order according to Patent Document 1, an apparatus in which a conical mirror and a television camera are integrated is moved along the center of the tunnel. As a result, a large number of rounding signals are obtained. When this circular belt-like image is formed on a planar CCD image pickup device and displayed on a display device, the inner wall of the tunnel is developed in a plane with equal density. This development includes information such as joints, cracks, cables, and dirt.

The development chart information is stored in the development chart storage unit, displayed on a display device as necessary, and printed out by a printer.
JP-A-6-241760

In order to effectively utilize the development map as described above for maintenance work such as a railway tunnel, it is necessary to extract and discriminate various kinds of information appearing on the development map. Since these operations are repetitive operations that require precision and patience, there are some examples of attempts to automate using computers, but the recognition speed and accuracy are often insufficient, most of which are It is a visual judgment work by hand.
Under such a background, it is desired to develop an effective method for extracting various kinds of information from the developed image.

Photographing vehicles such as the walls of tunnels were introduced in 2000, and tunnel images of all lines within the jurisdiction are acquired, and deformed development drawing work using a digitizer is being performed. However, since the input of the deformation by the digitizer is a manual operation, it takes a lot of time.
An object of the present invention is to develop a method for recognizing a deformed portion using a difference processing method in two developed images having different shooting times.
Another object of the present invention is to provide a method for removing influences such as a shift in photographing position and a change in illumination conditions during differential processing.
Still another object of the present invention is to provide a portion existing in both the first and second shot images, such as joints and conventional cracks, when performing differential processing of two images having different shooting times. The object of the present invention is to provide a processing method that removes and emphasizes only the progressing portion of the deformed portion such as a crack.

  The present invention extracts an image edge based on the first actual image data on the surface to be measured, and extracts an image edge based on the second actual image data after a predetermined time on the same surface to be measured. And calculating the difference between the specific pixel in the second edge image and the pixel in the first edge image corresponding to the specific pixel and the peripheral pixel of the pixel, and subtracting the smallest value of the difference. It is a method for extracting a progress state of an abnormal point in an image, characterized by comprising a difference processing step that outputs as a value and a step that outputs the difference obtained in the difference processing step as an advanced deformation point. .

For the second edge image, in order to remove linear components such as joints and cables, a step of extracting linear components of the image by Hough transform, and a mask processing step of masking and deleting the extracted linear components And are added.
The difference processing step uses specific pixels equal to or higher than the binarization threshold in the binarized second edge image in order to remove noise due to the ground color and dirt on the surface to be measured. In addition, even if there is a slight misalignment between the two images, the second edge image and the pixels in the first edge image corresponding to the specific pixel within a range that does not affect the extraction of the deformation point. And the difference with the surrounding 8 pixels of this corresponding pixel is calculated | required, and the smallest value of this difference is output as a difference value.
The difference processing step is performed after performing the position correction step of the captured image of the first edge image and the second edge image in order to extract the deformation point more accurately.

  According to the first aspect of the present invention, the step of extracting the edge of the image based on the first actual image data on the surface to be measured, and the second actual image data after a predetermined time elapses on the same surface to be measured. A step of extracting an edge of the image based on a difference processing step for obtaining a difference between the specific pixel in the second edge image and the pixel in the first edge image corresponding to the specific pixel, and the difference processing step Since the obtained difference is output as an advanced deformation point, it does not appear in the first image, but only the abnormal point such as a crack newly generated in the second image is processed. Output by technology and displayed.

  According to the second aspect of the present invention, the step of extracting the edge of the image based on the first actual image data on the measured surface, and the second actual image data after a predetermined time on the same measured surface. A step of extracting an edge of the image on the basis of the difference between the specific pixel in the second edge image, the pixel in the first edge image corresponding to the specific pixel and the peripheral pixel of the pixel, Since it comprises a difference processing step for outputting the smallest value as a difference value and a step for outputting the difference obtained in the difference processing step as an advanced deformation point, there are some differences in the first and second shot images. Even if there is a misalignment, only the deformation point is output accurately.

  According to the third aspect of the present invention, a step of extracting a linear component of the image by the Hough transform based on the second edge image and a mask processing step of masking and deleting the extracted linear component are added. Therefore, joints, newly laid cables and other linear edge images can be deleted by mask processing, and only deformed points can be detected. If the mask process is performed only for the second image, the first image is deleted by the difference process, so the mask process for the first image becomes unnecessary, and the processing time is greatly reduced. it can.

  According to the fourth aspect of the present invention, the difference processing step includes a specific pixel equal to or higher than the binarization threshold in the binarized second edge image, and a first edge image corresponding to the specific pixel. Since the difference between the pixel and the surrounding eight pixels of the corresponding pixel is obtained and the smallest value of the difference is output as the difference value, noise due to the ground color and dirt on the surface to be measured can be removed. Even if there is a slight misalignment between images, the deformation point can be extracted without affecting the deformation point more than necessary.

  According to the fifth aspect of the present invention, the step of extracting the edge of the image based on the actual image data of the surface to be measured is the first time extracted by setting the binarization threshold value representing the brightness of the edge image higher. The edge image and the second edge image extracted with the binarization threshold set to a low value are combined, and the threshold value is set to a high value and the extracted edge image and the high edge image are set to a low continuous value. The extracted edge image is deleted and the extracted edge image is deleted by setting it to a low level that exists by itself, so that not only clearly appearing cracks but also slightly unclear and easily overlooked cracks are ensured Can be extracted.

The present invention makes use of the first survey result to the maximum extent by using two images with different survey implementation times, and the progress of the deformation of the railway tunnel, etc. and the new deformation from the second survey result image. Extract parts efficiently.
For this reason, the present invention focuses on a method that contributes to maximizing the efficiency of information extraction / determination work relating to deformation, rather than automatically extracting abnormal portions inside the tunnel that are recognized as abnormal by performing image processing. Put.

Since the position of the captured image and the illumination condition of the two-period image are different, it is necessary to perform image position correction and shading correction processing before performing the difference processing.
The order of the image difference processing is as follows.
(1) Correct the position of the captured image of the two-period image.
(2) Perform image difference processing. In the difference processing, depending on the purpose, a simple difference method (a method of performing a difference between two images based on a corrected position) and a neighborhood minimum difference method (a first image of the second image based on a corrected position). Any one of a method of obtaining a difference between a point and a corresponding point of the first image and its neighboring points and using a minimum point of the difference as a difference value is employed. In particular, when it is assumed that the position of the two-period image is slightly shifted due to the shift of the photographing position, the neighborhood minimum difference method is adopted.

Embodiments of the present invention will be described with reference to the drawings.
As a two-time image that is a target of the method for extracting the progress of deformation points in the image of the present invention, for example, a first edge image is shown in FIG. Assume that the image is shown in FIG. In order to obtain such an edge image from the original image, first, the process of extracting the edge image will be described with reference to FIG.

(A) Tunnel side real image creation process (Fig. 6a)
As the image data to be subjected to the difference processing according to the present invention, for example, the image data according to the method described in Patent Document 1 mentioned as the conventional example can be used. In addition, while moving the imaging device at a constant speed along the center of the tunnel, the laser is rotated toward the inner periphery of the tunnel and rotated, the inner surface of the tunnel is scanned in a spiral shape, and the spiral scan line is flattened. It is also possible to obtain a planar image by expanding the image into a shape.
The method for obtaining a planar image is not limited to the above example. Further, the image data may be binarized for each pixel in advance, or analog image data may be A / D converted.

(B) Development view image creation process (b in FIG. 6)
In the wall image created in the step (a), the coordinate point (m, n) on the development view corresponding to the coordinate point (x, y) on the actual image data is obtained by development processing, and development is performed from this coordinate data. By creating a diagram, it becomes a development view developed in the length direction of the tunnel.

(C) Adjusting and continuing process (c in FIG. 6)
Since the developed view obtained in the step (b) may be distorted by a photographing method using a video camera or the like, the tunnel depth direction correction, the center point correction, and the inclination correction are performed accurately. Create an exploded view. Assuming that the image data expansion process is performed for each frame, the more the image after the expansion process is located at the back of the tunnel, the more information is insufficient and the image is distorted. In order to remove the distorted portion after the development processing, the development images for a plurality of frames are synthesized. At the time of synthesis, the center point of the tunnel may be different for each image, so data matching is performed. The combined development view information becomes a continuous development view, which is stored in a storage unit (not shown) and displayed on a monitor as necessary.

(D) Process of reading a developed image (d in FIG. 6)
A development view image is read from the development view obtained in the step (c). In the difference processing according to the present invention described later, a straight line portion such as a joint or a cable is read in a state displayed in the image, and the straight line portion is deleted by performing mask processing, and the process will be described later.

(E) Edge extraction process (e in FIG. 6)
Edges are extracted from the image read in step (d). As an edge extraction method, the Canny method, the Derich method, the Sugiyama-Abe method, the Sobel method, and the like are known. For example, it is extracted by the Canny method.

(F) Noise removal step (f in FIG. 6)
Since the crack has a certain length, a line segment having a certain area or less is determined as noise and deleted. The edge image after removing the noise is shown in FIGS.

Next, a process of extracting the progress of the deformed point by differential processing by the method of the present invention from the edge image of the tunnel inner surface obtained as described above will be described with reference to FIG.
(G) First edge extraction process for the first image (g in FIG. 7)
In this step, for the first edge image shown in FIG. 4 (a), the binarization threshold value representing brightness is set higher and the first edge extraction is performed.

(H) Second edge extraction process for the first image (h in FIG. 7)
In this step, the second edge extraction is performed by setting a lower binarization threshold representing brightness for the first edge image shown in FIG. 4A.

(I) First and second edge image composition step (i in FIG. 7)
In this step, the first and second edge images of steps (g) and (h) are combined, and the first edge extracted by setting the threshold value higher and the lower edge continuous to this higher edge are set. The second edge line segment extracted in this manner is left, and the lower edge line segment existing alone is deleted. The edge image on the horizontal scanning line 36 with the first edge image shown in FIG. 4A is shown in FIG. 1A, where the horizontal axis represents pixels and the vertical axis represents brightness. This figure shows a joint straight portion 57, a conventional crack 58, and a cable shadow straight portion 57.
Similarly,

(J) First edge extraction process step for the second image (j in FIG. 7)
In this step, the first edge extraction is performed by setting a high binarization threshold value representing brightness for the second edge image shown in FIG. 4B.

(K) Second edge extraction process for the second image (k in FIG. 7)
In this step, the second edge extraction is performed by setting the binarization threshold value representing the brightness lower for the second edge image shown in FIG. 4B.

(L) First and second edge image synthesis steps (l in FIG. 7)
In this step, the first and second edge images of steps (j) and (k) are combined, and the first edge extracted by setting the threshold value higher and the lower edge continuous to the higher edge are set. The second edge line segment extracted in this manner is left, and the lower edge line segment existing alone is deleted. The edge image on the horizontal scanning line 36 of the second edge image shown in FIG. 4B is shown in FIG. In this figure, a new crack 76 not shown in the first image is shown.

(M) Second edge image Hough transform / mask process (m in FIG. 7)
With respect to the second edge image, Hough transformation is performed for each point of the extracted line segment, and processing for picking up the straight line portion 57 and masking the line segment in the vicinity thereof is performed.
The Hough transform is known as an effective and versatile method for extracting a straight line, a circle, an arbitrary figure, etc. in a binary image. This will be described in more detail with reference to FIG. 3. In FIG. 3A, the binarized point P1 (x1, y1) in the current image is represented by polar coordinates as follows.
ρ = x1 · cos θ + y1 · sin θ
This equation is a relational expression between θ and ρ, and represents a curve L1 on the θ-ρ plane shown in FIG. When the same processing is performed on the binarized points P2 (x2, y2) and P3 (x3, y3), curves L2 and L3 on the θ-ρ plane are obtained. When these P1, P2, and P3 are on the straight line L on the xy plane, as shown in FIG. 3B, the curves L1, L2, and L3 on the θ-ρ plane intersect one point.
Thus, it can be seen that the point having the largest number of intersections on the θ-ρ plane corresponds to the straight line passing through the most point on the xy plane.

The Hough transformation for the second edge image will be further described with reference to FIG. 5. In the second image shown in FIG. 5A, the point group constituting the collinear portion 57 is represented by the horizontal axis. Is formed as the peak value of the intersection point in FIG. By performing straight line mask processing based on the result of the Hough transform, only the conventional crack 58 and the new crack 58 are obtained as shown in FIG.

(N) Small area cut process (n in FIG. 7)
Since the crack has a certain length, a line segment having a small area of a certain area or less is determined as noise and deleted.

(O) Two-time image position correction processing step (o in FIG. 7)
In this step, for example, a plurality of characteristic pixels are extracted from the two-period image by using the template method, and the difference is obtained while shifting the position to set the minimum point as the correct position.

(P) Binary image difference processing step (p in FIG. 7)
In this step, the first edge image shown in FIG. 1A is subtracted from the masked image of the second image shown in FIG. Then, as shown in FIG. 1D, a new crack 76 appears with a positive value, and a linear component 57 such as a joint or a cable shadow appears with a negative value. Further, the conventional crack 58 becomes 0 by subtracting the first image from the second image.
Further, by excluding the minus part from the result of FIG. 1D, only the new crack 58 is extracted as a deformed point as shown in FIG.

  The difference process in the step (p) will be described in more detail. Either the simple difference method or the neighborhood minimum method is adopted for the difference process. Both methods are selectively used depending on the case as follows.

(1) Simple difference method The difference between the second mask image shown in FIG. 1C and the first edge image shown in FIG.
Here, when there is no position shift between the first image and the second image, only a deformed point such as a new crack can be obtained by calculating a simple difference between the two images.
This method is simple and excellent in operability, but is somewhat unreliable if there is misalignment between the two.
For example, it is assumed that the image shown in FIG. 2A is the first edge image and the image shown in FIG. 2B is a masked image of the second image. An image shown in FIG. 2C is obtained by taking only a simple difference between these images by the simple difference method. Thus, if there is a positional shift between the two images, the conventional image portion does not completely overlap, and the remaining portion increases.

(2) Nearest neighbor method Therefore, in order to avoid this influence, the same edge image as the first edge image shown in FIG. 2 (a) is used for one pixel P in the second image shown in FIG. 2 (b). The difference between the pixel Q5 at the position and the pixels Q1, Q2, Q3, Q4, Q6, Q7, Q8, and Q9 within the 3 × 3 area around the position is obtained, and the smallest difference value is set as the difference value. Specifically, the difference values between P and Q4, Q5, and Q7 are each 0, and the difference values between P and Q1, Q2, Q3, Q6, Q8, and Q9 are each 1. Therefore, the value 0 having the smallest difference is the difference value.
Similarly, assuming that the smallest difference value for each pixel is the difference value, as shown in FIG.
Note that the difference is not limited to the surrounding 3 × 3 area, and may be obtained from a 5 × 5 pixel. However, if it is too wide, there is a risk of affecting the essential cracks, so it is desirable to determine it according to the size of the positional deviation.

In the difference processing shown in FIG. 1D, the mask processing for deleting the straight portions 57 such as joints and cable shadows is performed in the second mask image, but the joints and cable shadows and the like are performed in the first edge image. The process of deleting the straight line portion 57 is not performed. However, if the difference between the second mask image and the first edge image is obtained, the straight portion 57 such as the joint of the first edge image and the cable shadow becomes negative, and it is sufficient to exclude this negative portion. There is no need to perform processing for deleting the straight line portion 57 from the image.
Also, small edges and dirt on the wall surface can be deleted by setting a binarization threshold that does not affect the cracks.
As described above, by performing the binarization difference process, as shown in FIGS. 1E and 2D, the progress point 76 of the deformed point such as a new crack is emphasized and detected. The

(Q) Small area cutting process (q in FIG. 7)
Since the crack has a certain length, a line segment having a small area equal to or smaller than a certain area is determined to be noise, and the deformed portion extraction is completed.

  In the above-described embodiment, the case where a curved wall surface such as a railway tunnel is developed into a planar image has been described. However, the present invention can be applied to a flat wall surface or an uneven structure. .

3 is a process flow diagram illustrating an embodiment of a method for extracting progress of deformation points in an image according to the present invention. (A) is an explanatory diagram of an edge image of the first imaging, (b) is an explanatory diagram of an edge image of the second imaging, and (c) is a simple difference image between the mask image of the second imaging and the edge image of the first imaging. Explanatory drawing, (d) It is explanatory drawing of the neighborhood minimum difference image of the mask image of 2nd imaging | photography, and the edge image of 1st imaging | photography. It is explanatory drawing of Hough (Hough) transformation | transformation, (a) is an xy coordinate figure of an original image, (b) is a θ-ρ polar coordinate figure. (A) is the edge image figure of the 1st imaging | photography, (b) is the edge image figure of the 2nd imaging | photography. (A) is the edge image of the 2nd imaging | photography, (b) is a Hough conversion image figure. It is a process flowchart of the method of extracting the crack in an image. 3 is a process flow diagram illustrating an embodiment of a method for extracting the progress of deformation points in an image according to the present invention.

Explanation of symbols

36: Scanning line, 57: Straight portion of joint, cable, etc., 58 ... Conventional crack, 76 ... Progression portion due to new crack.

Claims (5)

Extracting the edge of the image based on the first actual image data of the surface to be measured, extracting the edge of the image based on the second actual image data after a predetermined time on the same surface to be measured, and A difference processing step for obtaining a difference between a specific pixel in the second edge image and a pixel in the first edge image corresponding to the specific pixel, and a deformed point obtained by developing the difference obtained in the difference processing step A method for extracting the progress of deformation points in an image, comprising the step of outputting. Extracting the edge of the image based on the first actual image data of the surface to be measured, extracting the edge of the image based on the second actual image data after a predetermined time on the same surface to be measured, and A difference in which the difference between the specific pixel in the second edge image, the pixel in the first edge image corresponding to the specific pixel and the peripheral pixel of the pixel is obtained, and the smallest value of the difference is output as the difference value A method for extracting the progress of deformation points in an image, comprising a processing step and a step of outputting the difference obtained in the difference processing step as an advanced deformation point. 3. A step of extracting a linear component of the image by Hough transform based on the second edge image and a mask processing step of deleting the extracted linear component by mask processing are added. A method for extracting the progress of deformation points in the described image.   The difference processing step includes a specific pixel equal to or higher than a binarization threshold in the binarized second edge image, a pixel in the first edge image corresponding to the specific pixel, and eight pixels around the corresponding pixel. 3. A method for extracting the progress of deformed points in an image according to claim 1 or 2, wherein the difference is obtained and the smallest value of the difference is output as a difference value.   The step of extracting the edge of the image based on the actual image data of the surface to be measured includes the first edge image extracted by setting the binarization threshold representing the brightness of the edge image to a high value, and the binarization threshold being lowered. The second edge image extracted by setting to, and leaving the edge image extracted by setting the threshold value higher and the edge image extracted by setting the lower edge continuous to this higher edge image, 3. The method for extracting the progress of deformation points in an image according to claim 1 or 2, wherein the edge image extracted by setting it to be lower and existing alone is deleted.

Images