JP2017072517A - Detection method of surface crack of concrete - Google Patents

Abstract

PROBLEM TO BE SOLVED: To detect an isolated crack of a surface of concrete.SOLUTION: An image of a surface of concrete is created, the image is converted to a gray scale, a noise image is corrected from the gray scale image, a pixel which is high in a possibility of a crack is detected from the corrected image, a pixel adjoining the detected pixel being a pixel which exists in a range in which the luminance of the pixel is defined in advance is detected, the pixel is expanded as a crack region, the expansion of the region is finished when the pixel in a preset range is not detected, an isolated crack existing in an integrated region around the expanded crack region is detected, it is determined that the isolated crack in the integrated region is integrated with a region of the expanded crack, after the integration, the expansion of the region, the detection of the isolated crack and the integration are sequentially repeated, when a finish condition is satisfied at the integration, the integration is finished, the repetition is finished, and a group of the integrated cracks is defined as a final crack.SELECTED DRAWING: Figure 1

Description

The concrete structure is gradually damaged due to the use of the structure and aging, and if left untreated, it may lead to fatal destruction of the concrete structure.
For this reason, it is necessary to maintain and manage concrete structures continuously and systematically.
In general, the beginning of damage to a concrete structure often appears as a crack on the surface of the concrete.

Currently, in the maintenance and management of concrete structures, inspections are conducted with the aim of grasping the damage state of concrete structures, and cracks formed on the concrete surface are important for grasping the damage state. It is one of the indicators.
The confirmation of this crack has traditionally relied on the visual inspection of the inspection worker, but the evaluation of the crack changes depending on the skill level and physical condition of the inspection worker, resulting in large variations in the inspection results, high altitude and narrowness. There was a problem that visual inspection itself could not be performed in places where inspection workers such as departments were difficult to approach.

For this reason, in recent years, a method for automatically detecting cracks included in an image by photographing the surface of concrete with a camera and analyzing an image obtained by the photographing has been studied and proposed.
Crack detection through such image analysis can quantitatively evaluate the occurrence of cracks, evaluate places where it is difficult for inspectors to approach, and simplify work at the inspection site just to take images. There are a number of advantages such as greatly reduced inspection time.

  By the way, the concrete structure to be inspected is often set outdoors, and the environment at the time of inspection, for example, shadow on the concrete surface, dirt adhering to the concrete surface, formed on the concrete surface There is a problem that an image such as unevenness (hereinafter referred to as a noise image) prevents accurate detection of cracks.

Furthermore, in addition to continuous cracks, there are cracks that are isolated at a distance from the concrete surface.
Such isolated cracks are likely to be continuous in the concrete with other consecutive cracks, and the skilled worker evaluates them as part of the continuous cracks.

However, with the conventional detection method, it has been difficult to select and detect isolated cracks in the noise image in a series of steps.
In addition, even if an isolated crack can be detected, it is determined as a part of another continuous crack and is not reflected in the quantitative determination of the crack.

JP2013-250058A

  The present invention has been made to solve the above-described problem, and is a crack generated on the concrete surface of a concrete structure, and even if it is a crack that is isolated on the concrete surface. It is possible to identify multiple cracks that may be continuous inside as a single continuous crack, so that the isolated cracks on the surface can be recognized as part of a large crack that spreads inside the concrete. An object of the present invention is to make it possible to reflect a continuous crack including a crack to a quantitative determination of a damaged state of a concrete structure.

In order to achieve the above object, a method for detecting a surface crack in concrete according to the first invention of the present invention is as follows.
A shooting step of shooting the surface of the concrete and generating an image;
A conversion step for converting the generated image to grayscale;
A correction step for correcting a noise image from the converted grayscale image;
A crack detection step for detecting pixels that are likely to be cracked from the corrected image in which the noise image is corrected;
An area expansion step for detecting a pixel whose pixel brightness is in a predetermined range among pixels adjacent to the pixel detected in the crack detection step, and extending the detected pixel as a crack area;
Repeating the region expansion step, and when no pixels in the predetermined range are detected, a region expansion end step for ending the region expansion step;
After the region expansion end step, an integrated region in which cracks are highly likely to exist is set around the expanded crack region, and an isolated crack detection step for detecting an isolated crack existing in the integrated region;
An integration step for determining that an isolated crack present in the integrated region set according to the expanded crack region is integral with the expanded crack region;
When it is determined in the integration step that they are integrated, after the integration step, the region expansion step is executed again, and the region expansion end step, the isolated crack detection step, and the integration step are sequentially performed. An iterative step to perform,
In the integration step, when a predetermined end condition is satisfied, an iterative end step that ends the integration step; and
A crack determination step for determining a group of integrated cracks as a final crack;
It is characterized by including.

Further, the method for detecting surface cracks in concrete according to the second invention of the present invention, in the first invention,
In the correction step, the correction value specified by the following formula (1) is used for correcting the shading difference.

In the above formula (1), i and j are the positions of the pixels of interest, I _a is an image after correction, I _b is an image before correction, and I _m is a median filter applied to the corrected image. The image b _m is the maximum luminance value.

In addition, the method for detecting surface cracks in concrete according to the third invention of the present invention is the first invention or the second invention,
The crack detection step
Assuming that the cracked pixels and the non-cracked background pixels each follow a normal distribution, the method includes a threshold generation step for obtaining a threshold that minimizes the average misclassification rate.
The threshold generation step is
An entropy minimization step to minimize the average conditional entropy;
Based on the minimum value of entropy obtained in the entropy minimization step, after classifying the luminance value distribution of the analysis target image into two classes, a pixel classification step for obtaining an average value and a standard deviation of the crack portion and the background portion, respectively,
Threshold calculation step for identifying a threshold for detecting a pixel having a very high possibility of cracking using the average value and standard deviation of the crack portion obtained by the pixel classification step and the average value and standard deviation value of the background portion When,
A connected element specifying step for specifying a connected element based on the threshold value obtained by the threshold value calculating step;
It is characterized by including.

In addition, the method for detecting surface cracks in concrete according to the fourth invention of the present invention is the third invention,
The step of minimizing the average conditional entropy is characterized by minimizing the average conditional entropy using the following equation (2).

In Eq. (2), E is the average conditional entropy, C _i is the two classes to be classified, i = 1 is the class in which the cracked pixels are gathered, i = 2 is the class in which the background pixels are gathered, and j is the discretization The luminance value p (j) is a probability that a pixel having the luminance value j appears in the analysis image.

Further, the method for detecting surface cracks in concrete according to the fifth invention of the present invention is the fourth invention,
The threshold value calculating step is characterized in that the threshold value is specified using the following formula (3).

In Equation (3), μ _c is the average value of the cracked portion, σ _c is the standard deviation of the cracked portion, μ _b is the average value of the cracked portion, σ _b is the standard deviation of the cracked portion, and t ₁ is the threshold value. To do.

Moreover, the method for detecting surface cracks in concrete according to the sixth invention of the present invention is the fifth invention,
The connected element specifying step uses the following formula (4), omits from the detection result those that do not satisfy the condition of this formula, and specifies the connected elements from the pixel group and narrows down crack candidates.

In the equation (4), the eccentricity is an eccentricity of an ellipse having the same second moment as the connection element region, and A _min is a value that changes according to the shooting distance and the number of pixels of the camera.

Moreover, the detection method of the surface crack of the concrete concerning the 7th invention of the present invention, in the 6th invention,
The region expansion step
Among the pixels adjacent to the pixels detected in the connected element specifying step, those having a luminance value smaller than a predetermined value are detected, and the detected pixels are integrated into the connected elements specified in the connected element specifying step. It is characterized by doing.

Moreover, the method for detecting surface cracks in concrete according to the eighth invention of the present invention is any one of the first to seventh inventions.
The isolated crack detection step
An ellipse having a major axis l _l , a minor axis l _s , and an angle _α having the same second moment as the expanded crack region is set as an integrated region,
A region that satisfies the condition of the following formula (5) is set as an isolated crack candidate,
Of the isolated crack candidates, only those satisfying the geometrical condition of the following formula (6) are detected as isolated cracks.

  In the equation (5), β is a parameter representing the weight ratio between the long axis and the short axis, and d is a parameter for detecting a discontinuous crack region regardless of the long axis or short axis. It is.

In the above formula _{(6), P (t 1} ) and P _{(t r)} is the cumulative probability of the luminance values _{t 1} and _{t r} in the analysis image.

Moreover, the method for detecting cracks in the concrete surface according to the ninth invention of the present invention is the eighth invention,
An end condition of the iteration step is the following expression (7).

Note that n _p is the number of pixels determined to be cracked when the number of repetitions is p, and when this termination condition is satisfied by the pth repetition, the final result is the p−1th result.

  In the above invention, a crack that appears on the concrete surface of a concrete structure and that appears to be isolated on the surface of the concrete is determined as a continuous crack inside the concrete. Quantitative determination of the failure state of a concrete structure allows continuous cracks including isolated cracks to be identified so that isolated cracks on the surface of a concrete wall can be recognized as part of a large crack extending inside the concrete It becomes possible.

It is a flowchart which shows the flow of the detection method of the surface crack of the concrete concerning this invention. FIG. 2 is an example of a captured image generated in the flow of FIG. FIG. 3 is a grayscale image of the captured image of FIG. FIG. 4 is a grayscale image before correction of a noise image for an image having a strong influence of shadow. FIG. 5 is a corrected image after correcting the noise image for an image having a strong influence of shadows. FIG. 6 is a grayscale image before correction of a noise image for an image having a strong influence of dirt. FIG. 7 is a corrected image after correcting the noise image for an image that is strongly influenced by dirt. FIG. 8 is a gray scale image before correction of a noise image for an image having many thin cracks. FIG. 9 is a corrected image after correcting the noise image for an image having a strong influence of shadows. FIG. 10 is an image near the center of FIG. FIG. 11 is a flowchart showing the flow from the start to the end of crack detection in FIG. FIG. 12 is an image obtained by binarizing the image of FIG. FIG. 13 is an image obtained as a result of detecting connected elements from the image of FIG. FIG. 14 is an enlarged image of the isolated crack shown in the image of FIG. FIG. 15 is a schematic diagram showing a region where there is a high possibility that a crack exists in the image of FIG. FIG. 16 is a processed image showing a state of p = 5 in the repetition of the processing on the image of FIG. FIG. 17 is a processed image showing a state of p = 10 in the repetition of the processing on the image of FIG. FIG. 18 is a processed image showing a state of p = 15 in the repetition of the processing on the image of FIG. FIG. 19 is a comparison image of the result of the detection method according to the present invention and the result of the conventional detection method. FIG. 20 is a comparison image of the result of the detection method according to the present invention and the result of the conventional detection method. FIG. 21 is a comparison image of the result of the detection method according to the present invention and the result of the conventional detection method. FIG. 22 is a comparison image of the result of the detection method according to the present invention and the result of the conventional detection method. FIG. 23 is a comparison image of the result of the detection method according to the present invention and the result of the conventional detection method. FIG. 24 is a comparison image of the result of the detection method according to the present invention and the result of the conventional detection method.

Hereinafter, the present invention will be described with reference to the drawings.
1 is a flowchart showing a flow of a method for detecting surface cracks in concrete according to the present invention, FIG. 2 is an example of a photographed image generated in the flow of FIG. 1, and FIG. 3 is a gray of the photographed image of FIG. 4 is a grayscale image before correction of a noise image for an image with a strong influence of shadow, FIG. 5 is a correction image after correction of the noise image for an image with a strong influence of shadow, and FIG. FIG. 7 shows a corrected image after correction of a noise image for an image having a strong influence of dirt, and FIG. 8 shows an image before correction of the noise image for an image having many thin cracks. 9 is a corrected image after correcting a noise image for an image having a strong influence of shadow, FIG. 10 is an image near the center of FIG. 8, and FIG. 11 is from the start of crack detection in FIG. I was at the end FIG. 12 is an image obtained by binarizing the image of FIG. 10, FIG. 13 is an image obtained by detecting a connected element from the image of FIG. 12, and FIG. 14 is an image of FIG. FIG. 15 is a schematic diagram showing an area where there is a high possibility that cracks are present in the image of FIG. 14, and FIG. 16 is a state where p = 5 in the repetition of the processing for the image of FIG. FIG. 17 shows a processed image showing a state of p = 10 in the repetition of the processing of the image of FIG. 12, and FIG. 18 shows a state of p = 15 in the repetition of the processing of the image of FIG. Processed image, FIG. 19 is a comparison image of the result of the detection method according to the present invention and the result of the conventional detection method, and FIG. 20 is the result of the detection method according to the present invention and the result of the conventional detection method. Comparative image, Fig. 21 FIG. 22 is a comparison image of the result of the detection method according to the present invention and the result of the conventional detection method. FIG. 22 is a comparison image of the result of the detection method according to the present invention and the result of the conventional detection method. FIG. 24 is a comparison image of the result of the detection method according to the present invention and the result of the conventional detection method. FIG. 24 is a comparison image of the result of the detection method according to the present invention and the result of the conventional detection method.

In the present invention, a concrete surface is photographed, and an image obtained by photographing is analyzed to appropriately recognize and evaluate a crack requiring maintenance.
First, based on FIG. 1, the flow of the detection method of the surface crack of the concrete concerning this invention is demonstrated.

In the present invention, first, the surface of concrete to be inspected is photographed using a photographing device to generate an image of the concrete surface (step S01).
Next, the obtained captured image is converted into a gray scale image (step S02).
This conversion to a gray scale image has the effect of shortening the calculation time required for processing the image.

As a general gray scale conversion, an average value of RGB components or a weighted average based on NTSC coefficients is generally used. In this embodiment, the maximum value of RGB components is adopted.
In general cracked parts and concrete surface parts, the RGB components are almost equal, but in graffiti etc., there is a difference in RGB components, so the use of the above maximum value is effective in eliminating elements other than cracks. is there.

Specifically, when gray scale conversion using the above maximum value is executed, the graffiti is displayed relatively brightly, and it is possible to prevent the graffiti from being mistaken as a crack.
Here, a specific example of a gray scale image employing the maximum value of RGB components will be shown.
Cracks are present in the vertical direction on the left side of the photographed image in FIG. 2, and dark stains (graffiti) are attached in the vertical direction on the right side.

When this photographed image is converted to a gray scale by the above-described processing, as shown in FIG. 3, it can be seen that the dirt becomes thin and is appropriately removed on the image.
Next, as a pre-process for correcting brain exploration due to light, shadow, dirt, etc., a median filter is adopted for the grayscale image to perform noise image correction (step S03).
Equation (1) used for correcting this noise image is shown below.

In the above formula (1), i and j are the positions of the pixels of interest, I _a is an image after correction, I _b is an image before correction, and I _m is a median filter applied to the corrected image. The image b _m is the maximum luminance value.

The maximum value of the luminance value b _m is, for example, 255 when a luminance value of 256 gradations is used.

The size of the filter affects the correction result. Here, 41 × 41 is used as the filter size.
However, the influence of this filter size does not greatly affect the final result. For example, when a 21 × 21 filter is used, the extraction result of cracks is thinner than that of 41 × 41. On the other hand, when a 61 × 61 size filter is used, the filter size is not limited to 41 × 41 in the present embodiment because it tends to be slightly thicker. .

Specifically, when the filter size is small, the luminance value of the denominator of the equation (1) becomes small near the crack, so that _{Ia on} the left side becomes large and correction is made brighter.
Here, to give an actual example, the crack width, which was 8 pixels in the case of 41 × 41, is about 6 pixels in the case of 21 × 21, and about 10 pixels in the case of 61 × 61. Often becomes.

Here, based on FIG. 4 to FIG. 9, a specific example by correcting the noise image using the filter size 41 × 41 will be described.
FIG. 4 shows a grayscale image having a large influence of shadows. When the above-described noise correction is performed on the grayscale image, as shown in FIG. 5, the noise image is appropriately removed from the image with severe shadow reflection, and cracks are generated. It can be seen that the image can be corrected while leaving

FIG. 6 shows a grayscale image that is heavily soiled, specifically, heavily soiled. When the above noise correction is performed on the grayscale image, as shown in FIG. It can be seen that the image can be corrected with the crack removed.
FIG. 8 shows a grayscale image with many thin cracks. When the above-described noise correction is performed on the grayscale image, it can be seen that the image can be corrected without losing the cracks as shown in FIG.

In the present invention, it is particularly important to properly determine cracks when a large number of fine cracks or fine cracks are included.
For this reason, below, description is continued based on the correction image of FIG. 9 including a thin crack and a fine crack.
The image in FIG. 8 is a very large image of about 5184 × 3456 pixels, and various processes executed below are difficult to discriminate on the image. Therefore, in the following description, the center of the image in FIG. An image (FIG. 10) obtained by trimming the vicinity to an area of 2592 × 1728 pixels is used.

Next, pixels that are very likely to be cracks, that is, cracks, are detected from the image of FIG. 10 (step S04).
In step S04 for detecting the cracks, threshold generation step S042 for obtaining a threshold value that minimizes the average misidentification rate, assuming that the cracked pixels and the non-cracked pixels (hereinafter referred to as background pixels) follow a normal distribution. Is included.

For this reason, when detection of a crack is started (step S041), the above-described threshold value generation step S042 is started (S043).
This threshold value generation step S042 includes an entropy minimization step S0431, a pixel classification step S0432, a threshold value calculation step S0433, and a connected element specifying step S0434.

For this reason, when the threshold generation step is started (S043), first, an entropy minimization step S0431 is executed.
This entropy minimizing step S0431 is to minimize the average conditional entropy.
Equation (2) used for this minimization is shown below.

In Eq. (2), E is the average conditional entropy, C _i is the two classes to be classified, i = 1 is the class in which the cracked pixels are gathered, i = 2 is the class in which the background pixels are gathered, and j is the discretization The luminance value p (j) is a probability that a pixel having the luminance value j appears in the analysis image.
In the above description, when using 256 gradation luminance values, j = 0. . . 255.

  Next, a pixel classification step S0433 is executed, and the luminance value distribution of the analysis target image is classified into two classes based on the above equation (2).

Next, a threshold calculation step S0433 is executed.
This threshold value calculation step S0433 uses the average value and standard deviation of the crack portion obtained in the pixel classification step S0432 and the average value and standard deviation value of the background portion to detect a pixel having a very high possibility of cracking. The threshold value is specified.
Equation (3) used for specifying this threshold is shown below.

In equation (3), μ _c is the average value of the cracked portion, σ _c is the standard deviation of the cracked portion, μ _b is the average value of the cracked portion, σ _b is the standard deviation of the cracked portion, and t ₁ is the threshold value. .

Here, after correcting the image of FIG. 10 using the filter of Expression (1),
FIG. 12 shows the results of analysis based on Equation (2) and Equation (3).
The image shown in FIG. 12 shows the result of binarization based on the obtained threshold value, and here, an area where a white display portion is detected as a crack is shown.
However, in this image, voids other than cracks are also detected in the same manner as cracks.

Therefore, in order to remove the detection pixels other than the cracks based on the geometric criteria, next, a connected element specifying step S0434 is executed.
In this connected element specifying step S0434, connected elements are specified for the pixel group detected by Expression (3), and crack candidates are further narrowed down based on the geometric characteristics of each connected element.
Here, the condition of the following formula (4) is given, and those not satisfying this condition are omitted from the detection result of the formula (3).

In the equation (4), the eccentricity is an eccentricity of an ellipse having the same second moment as the connection element region, and A _min is a value that changes according to the shooting distance and the number of pixels of the camera.
In this embodiment, A _min = 100.

As a result of the narrowing down under the above conditions, only the connecting elements that are long and have a certain area can be detected.
Here, FIG. 13 shows an image obtained as a result of selecting the pixels in the image of FIG. 12 based on Expression (4).

According to the image of FIG. 13, it can be confirmed that most of the voids often seen in FIG. 12 can be removed from the selected connected area displayed in white.
Then, when the narrowing-down process based on the condition according to the equation (4) is finished, the detection of cracks is finished (step S042).

After the above detection of cracks (step S042), next, expansion of the crack region (step S05) in FIG. 1 is executed.
Enhanced cracked area (step S05), first, among the pixels adjacent to the pixel detected in the coupling element specifying step, the luminance values, detects a smaller than a predetermined value t _r.
At this time, _{t r,} as a luminance value of a little lighter than _{t 1,} and _{t r = t 1 + ασ c} .

Next, the detected pixels are integrated into the connecting element specified in the connecting element specifying step S0434.
Next, in pixels adjacent to integrated pixel, the luminance value to detect smaller than t _r.
The pixels detected here are pixels that can be integrated and are pixels to be expanded.
When there is a pixel to be expanded, the process returns to expanding the cracked area (step S05), and the detection of the area to be expanded is executed again.

Next, when there is no longer any detection of the area to be expanded, the expansion of the cracked area ends (step S06).
In addition, in the expansion of the crack region (step S05), the value of α is important. In this embodiment, α = 0.1.
However, this value may be between α = 0.05 and α = 0.5.
This is because the value of α increases sequentially by repeating the processing described later, and no significant difference is recognized in the final result within this range.

When this expansion is completed, the luminance value in the image becomes slightly high, and connected cracks can be detected.
At this stage, an isolated crack as shown in FIG. 14 can be recognized as a continuous crack if the person confirms with the naked eye, but such an isolated crack, particularly an isolated small crack, is possible. Some cracks do not satisfy the condition of formula (4).

In order to avoid overlooking such an isolated crack, a region where there is a high possibility of the presence of a crack as shown in the schematic diagram of FIG. 15 is obtained, and if a certain condition is satisfied within the region, it is detected as a crack. There is a need.
Therefore, in the present invention, after the end of expansion (step S06), first, an ellipse having a major axis l _l , a minor axis l _s , and an angle _α having the same second moment as the crack region detected so far is integrated. Set as the conversion area.

Next, the u axis is set in the major axis direction of the ellipse and the v axis is set in the minor axis direction.
Next, considering that an elongated crack region has an isolated crack region in the major axis direction, it is highly possible that the isolated crack region is also a connected crack region. An area that satisfies all of the above is a candidate for an isolated crack.

In the equation (5), β is a parameter representing the weight ratio between the long axis and the short axis, and d is a parameter for detecting a discontinuous crack region regardless of the long axis or short axis. It is.
In this embodiment, d is 40 pixels.

These parameters are artificially determined based on the observation of isolated cracked areas in the image with the naked eye. However, it has been confirmed that the results do not have a significant effect if the order does not change.
For example, even if the analysis is performed with β = 4 and d = 8, the F value described later changes only about 0.01 or less in most images, and even the image with the largest change is only about 0.05 at most. It was.

However, since there is a risk of picking up a noise image existing in the major axis direction only under the condition of the equation (5), the luminance value is calculated as shown in the equation (5) from the newly detected area. t _r running region expansion on the condition that those smaller, on it, only those that meet all of the following geometrical condition of the expression (6), was detected as an isolated cracked (step S07).

In the above formula _{(6), P (t 1} ) and P _{(t r)} is the cumulative probability of the luminance values _{t 1} and _{t r} in the analysis image.
This is used to correct the area condition of Equation (4) in consideration that the noise area increases as the threshold value increases.
Other conditions are the same as those in the equation (4).

  Next, an isolated crack detected as being included in an area where there is a high possibility of a crack is determined as an integral part of the crack (step S08).

Here, when the integration is determined, the process proceeds to the repetition of the process (step S09) to return to the process of expanding the crack region (step S05), and the subsequent process, that is, the target of the expansion. Pixels are detected (step S07), and isolated cracks are integrated (step S08).
By repeating this process, the crack detection system is improved.

First, when the number of repetitions is p, region expansion is executed with α in Expression (5) set to 0.1 p.
Then, the detection of an isolated crack is executed using an elliptical formula in the same manner as the formula (6).
Although crack area detected by increasing the number of iterations increases, if the crack region by repeating the process has rapidly increased, the threshold t _r has become larger enough to determine the crack that much background and Conceivable.
Therefore, in consideration of these, the iteration termination condition shown in the following formula (7) was set.

N _p is the number of pixels determined to be cracked at the p-th iteration, and when this termination condition is met at the p-th iteration, the p-1th result is the final result, and the integration is terminated. (Step S10).

Here, FIGS. 16 to 18 show how the cracked area in the image is enlarged by repeating the process (step S09).
According to these drawings, it can be seen that the crack region is enlarged as a result of the repetition of the processing (step S09).

  When the results of the initial expansion in FIG. 16 before the region expansion in FIG. 13 and the analysis results after the expansion in FIG. 17 and FIG. 18 are compared, since the luminance value is slightly high and bright, it is initially detected as a crack. It can be seen that the pixels that were not detected are gradually detected as cracked areas.

Next, a group of integrated crack regions is determined as a final crack (step S11).
And, as a result of the present invention, since the information on the confirmed crack properly indicates the state of the crack generated in the concrete structure, if this is reflected in the quantitative determination of the damaged state of the concrete structure, the concrete It is very useful for the accuracy of inspection of structures and effective maintenance work.

Next, based on FIG. 19 to FIG. 24, the detection method according to the present invention and the conventional detection method (Otsu's method: Research on basic survey on the road edge, see National Research Institute document 381, 2007) The comparison result will be described.
Here, the detection method according to the present invention described above and the conventional detection method were used to detect cracks for six digital images taken under various conditions such as bridge decks, girders, and inner walls of buildings. .

The detection method according to the present invention is assumed to be strongly affected by shadows and dirt, and images including such shadows and dirt are also analyzed.
Among the images of FIGS. 19 to 24, the left figure shows the captured digital image, the middle figure shows the result of the detection method according to the present invention, and the right figure shows the result of the conventional detection method.
The captured images in FIGS. 19 and 20 are images each having a low luminance value and many dark portions because of the peripheral light reduction and shadows of the lens.

  In the detection method according to the present invention, these effects can be mainly removed by the filter of the expression (2), so that a good result is obtained. On the other hand, in the conventional detection method, the determination is simply made by the luminance value. The accuracy is poor.

The captured images in FIGS. 21 and 22 contain a lot of dirt.
This contamination can also be appropriately removed by the correction by the filter based on the formula (2) and the geometric feature, so that the detection method according to the present invention is very institutional compared to the conventional detection method. Cracks can be detected.

The photographed image in FIG. 23 is an image including many fine cracks.
The width of these fine cracks is about 1 to 3 pixels on the image, but the detection method according to the present invention can appropriately detect such fine cracks.
On the other hand, in the conventional detection method, as in FIGS. 19 and 20, in addition to erroneously detecting many shadow portions as cracks, a slightly thin line of water stains seen at the center of the image is erroneously detected as cracks. It has been detected.

  In the detection method according to the present invention, the reason why the water stain line is not erroneously detected is based on the determination criterion using the geometric feature amount considered in the processing related to Equation (4) or Equation (6). The effectiveness is remarkably exhibited.

The captured image of FIG. 24 is an example including erroneous detection, and the quantitative evaluation result of the F value shown later is the worst.
As can be seen from FIG. 24, since the shadow formed inside the P-con trace satisfies the condition of Equation (6), the periphery of the P-con trace is also detected as a crack.

  Since the area of the P-con trace is large, the quantitative evaluation described later has a bad result. However, if the result image is qualitatively judged, there are few false detections other than the P-con trace, and there is a crack. It can be seen that the parts are reliably detected.

Next, quantitative evaluation based on the detection result obtained by the detection method according to the present invention will be described.
Here, the results of the detection method according to the present invention according to the three indexes of precision, sensitivity, and specificity, and the comparison result, according to the above-mentioned conventional detection method, are modeled on the literature that discloses the method of Otsu. Evaluate the valuation result.
Table 1 shows an average value of the results obtained by analyzing 43 images obtained by the detection method according to the present invention.

In order to compare the accuracy of these indicators, you should originally create and share a benchmark set and examine the accuracy rate for them, but at present there is no such benchmark set.
Here, the accuracy was evaluated with the result of detecting cracks by the inventors' prior manual work as the correct answer.

Here, the sensitivity is the ratio of how much of the cracked pixels in the correct image are detected as cracks in the analysis result, and the specificity is the degree of analysis in the non-cracked pixels (background pixels) in the correct image. Also, the ratio of whether the pixel is detected as a background pixel and the matching ratio are defined as the ratio of how much the correct image is a cracked pixel among the cracked pixels detected in the analysis result.
In addition, in order to evaluate both sensitivity and precision at the same time, F value (pattern recognition (data science 5 learned by R), see Kyoritsu Shuppan, 2009) was also obtained.

As can be seen from Table 1, when compared in terms of sensitivity, the conventional method is higher than the method of the present invention.
This is because everything that is ambiguous is determined as a crack in the binarization by the conventional method.
For this reason, on the contrary, the relevance rate and specificity are significantly low.

In particular, paying attention to the precision, only 1.2% of the results detected by the conventional method show that the cracks were confirmed by human eyes, and it can be seen that accurate detection was not possible.
Even when looking at the F value, which is an index for simultaneously evaluating the sensitivity and precision, the method of the present invention is slightly inferior to the conventional method in sensitivity, but has a great difference in precision because it is very excellent in precision.
Thus, even in quantitative examination, it is clear that the detection method according to the present invention can detect an accuracy crazing.

Thus, in the concrete surface crack detection method according to the present invention, a detection method based on image analysis focusing on geometric and statistical parameters was proposed, and its accuracy was confirmed.
In particular, the process of determining the threshold by statistically treating the brightness values of the crack and the background, and geometric parameters should be used to connect the cracks that are isolated in the image. The detecting step has not been seen in the past, and an excellent effect is recognized in terms of proper detection of cracks.

For this reason, in the method for detecting surface cracks in concrete according to the present invention, it is possible to select isolated cracks in a captured image in a series of affirmations on a photographed image, detect them as cracks, and evaluate them appropriately. become.
In addition, the detection method according to the present invention has made it possible to detect a crack well even from a crack image that contains a lot of shadows and dirt, which has been generally difficult to analyze in the past.

Claims (9)

A shooting step of shooting the surface of the concrete and generating an image;
A conversion step for converting the generated image to grayscale;
A correction step for correcting a noise image from the converted grayscale image;
A crack detection step for detecting pixels that are likely to be cracked from the corrected image in which the noise image is corrected;
An area expansion step for detecting a pixel whose pixel brightness is in a predetermined range among pixels adjacent to the pixel detected in the crack detection step, and extending the detected pixel as a crack area;
Repeating the region expansion step, and when no pixels in the predetermined range are detected, a region expansion end step for ending the region expansion step;
After the region expansion end step, an integrated region in which cracks are highly likely to exist is set around the expanded crack region, and an isolated crack detection step for detecting an isolated crack existing in the integrated region;
An integration step for determining that an isolated crack present in the integrated region set according to the expanded crack region is integral with the expanded crack region;
When it is determined in the integration step that they are integrated, after the integration step, the region expansion step is executed again, and the region expansion end step, the isolated crack detection step, and the integration step are sequentially performed. An iterative step to perform,
In the integration step, when a predetermined end condition is satisfied, an iterative end step that ends the integration step; and
A crack determination step for determining a group of integrated cracks as a final crack;
For detecting surface cracks in concrete containing The method for detecting a surface crack in concrete according to claim 1, wherein in the correction step, the correction value specified by the following formula (1) is used for correction of the shading difference.
In the above formula (1), i and j are the positions of the pixels of interest, I _a is an image after correction, I _b is an image before correction, and I _m is a median filter applied to the corrected image. The image b _m is the maximum luminance value. The crack detection step
Assuming that the cracked pixels and the non-cracked background pixels each follow a normal distribution, the method includes a threshold generation step for obtaining a threshold that minimizes the average misclassification rate.
The threshold generation step is
An entropy minimization step to minimize the average conditional entropy;
Based on the minimum value of entropy obtained in the entropy minimization step, after classifying the luminance value distribution of the analysis target image into two classes, a pixel classification step for obtaining an average value and a standard deviation of the crack portion and the background portion, respectively,
Threshold calculation step for identifying a threshold for detecting a pixel having a very high possibility of cracking using the average value and standard deviation of the crack portion obtained by the pixel classification step and the average value and standard deviation value of the background portion When,
A connected element specifying step for specifying a connected element based on the threshold value obtained by the threshold value calculating step;
The method for detecting cracks on the surface of a concrete according to claim 1 or 2. The method for detecting surface cracks in concrete according to claim 3, wherein the step of minimizing the average conditional entropy uses Equation (2) below to minimize the average conditional entropy.
In Eq. (2), E is the average conditional entropy, C _i is the two classes to be classified, i = 1 is the class in which the cracked pixels are gathered, i = 2 is the class in which the background pixels are gathered, and j is the discretization The luminance value p (j) is a probability that a pixel having the luminance value j appears in the analysis image. The method for detecting a surface crack in concrete according to claim 4, wherein the threshold value calculating step uses the following formula (3) to specify the threshold value.
In Equation (3), μ _c is the average value of the cracked portion, σ _c is the standard deviation of the cracked portion, μ _b is the average value of the cracked portion, σ _b is the standard deviation of the cracked portion, and t ₁ is the threshold value. To do. The connected element specifying step uses the following expression (4), omits those that do not satisfy the condition of this expression from the detection result, specifies connected elements from the pixel group, and narrows down crack candidates. A method for detecting cracks in the surface of concrete.
In the equation (4), the eccentricity is an eccentricity of an ellipse having the same second moment as the connection element region, and A _min is a value that changes according to the shooting distance and the number of pixels of the camera. The region expansion step
Among the pixels adjacent to the pixels detected in the connected element specifying step, those having a luminance value smaller than a predetermined value are detected, and the detected pixels are integrated into the connected elements specified in the connected element specifying step. The method for detecting surface cracks in concrete according to claim 6. The isolated crack detection step
An ellipse having a major axis l _l , a minor axis l _s , and an angle _α having the same second moment as the expanded crack region is set as an integrated region,
A region that satisfies the condition of the following formula (5) is set as an isolated crack candidate,
The method for detecting a surface crack in concrete according to any one of claims 1 to 7, wherein only those that satisfy the geometrical condition of the following formula (6) among the isolated crack candidates are detected as isolated cracks.
In the equation (5), β is a parameter representing the weight ratio between the long axis and the short axis, and d is a parameter for detecting a discontinuous crack region regardless of the long axis or short axis. It is.
In the above formula _{(6), P (t 1} ) and P _{(t r)} is the cumulative probability of the luminance values _{t 1} and _{t r} in the analysis image. The method for detecting a surface crack in concrete according to claim 8, wherein an end condition of the repetitive step is the following formula (7).
Note that n _p is the number of pixels determined to be cracked when the number of repetitions is p, and when this termination condition is satisfied by the pth repetition, the final result is the p−1th result.