JP2021092414A - Crack extraction method - Google Patents

Abstract

To provide an image processing method for accurately extracting image information to obtain digital data when extracting a crack from a photographed image obtained by photographing the surface of a concrete material.SOLUTION: A crack extraction method extracts image information of a crack part generated on the surface of a concrete material from a photographed image obtained by photographing the surface of the concrete material. The crack extraction method includes: a binarization process step of binarizing a photographed image to distinguish a shadow portion inside the crack from other portions; an isolated point removal process step of removing isolated points which are not detected as linear portions corresponding to the crack from the binarized image; and a raster vector conversion process step of detecting continuous dots from the image from which the isolated points are removed, generating line information, and displaying the line information on a display device.SELECTED DRAWING: Figure 2

Description

The present invention relates to a crack extraction method for extracting cracks generated on the surface of a concrete material from a photographed image of the surface of the concrete material.

For existing concrete structures, the occurrence of cracks should be definitely recognized in order to maintain their soundness. For example, in an existing reinforced concrete structure, it is important to verify whether the soundness of the internal reinforcing bars is maintained due to deterioration over time. Existing reinforced concrete structures deteriorate due to various causes. For example, the causes of deterioration of concrete materials are categorized as typical causes such as salt damage, neutralization, corrosion of reinforced steel materials due to frost damage, alkali silica reaction, chemical corrosion, fatigue, etc., and individual deterioration mechanisms. Are different.

Specifically, salt damage is a phenomenon in which cracks occur in a concrete material due to corrosive expansion of reinforcing bars, and structural safety is impaired. For example, salt that comes from the coast penetrates into the concrete material and corrodes the reinforcing bars. In addition, antifreeze agents sprayed on roads in the mountainous areas of cold regions may also cause salt damage. The passivation film that protects the corrosion of the reinforcing bar is destroyed when the chloride ion reaches a certain concentration. By forming a corrosive battery with the part covered with the passivation film, the corrosion of the steel material progresses violently.

Further, the corrosion of the reinforcing bar due to the neutralization is caused by the neutralization of the highly alkaline concrete material. Neutralization is a phenomenon in which carbon dioxide in the air infiltrates calcium hydroxide in hardened cement and changes it to calcium carbonate. Calcium hydroxide, which is a high alkali, is transformed into calcium carbonate, the pH is neutralized, and it deteriorates by making the reinforcing bar a corrosive environment.

Further, frost damage is a deterioration phenomenon caused by repeated freezing and thawing of water contained in a concrete material. Due to the volume expansion due to the freezing of water in the concrete material, pressure is generated inside the concrete material, and the structure loosens when the ice melts and returns to water. This repetition makes the concrete structure fragile.

Further, the alkali-silica reactivity: degradation by (ASR Alkali Sillica Reaction) is an alkali in the cement, and K ions or Na ions and of H ₂ O some silica in the aggregate (silica aggregate) concrete tissue Is a phenomenon in which the concrete material cracks due to water absorption and expansion, resulting in deterioration of performance. The water-absorbed and expanded aggregate causes cracks of a relatively large width in the concrete material, the aggregate itself loses strength, and the concrete structure becomes fragile.

Further, chemical corrosion is a phenomenon in which the surface of a concrete material is gradually weakened by sulfate generated from a sewage treatment facility or the like, strong acid used in a plating factory or the like, acidic water in a hot spring area or the like. Originally, highly alkaline cement concrete materials are sensitive to acids. Therefore, when it comes into contact with acid, it is necessary to take measures such as coating it with an acid-resistant material. Fatigue deterioration is a phenomenon in which a floor slab is gradually cracked and finally missing due to repeated action even if an external force smaller than the fracture strength is applied to the road deck or the like.

Currently, there is an attempt to recognize cracks from images of concrete materials using artificial intelligence (AI), but in order for AI to learn, the same accuracy, the same shooting conditions, and images of concrete materials of the same structure are required. It is necessary to prepare at least 1000 sets of images in which the cracks of the image are accurately specified (in units of one pixel) for deep learning, but since the concrete materials themselves are different, the conditions are set. It's very difficult to meet, and it's almost impossible.

As another method of recognizing cracks, humans still manually trace and recognize cracks. However, with this method, although the crack can be pointed out, the length and width of the crack cannot be recognized accurately. In addition, in order to display cracks as digital data, the data could not be converted to CAD as it was, and humans still had to input it to CAD.

On the other hand, a method for measuring cracks by image processing has also been proposed (see, for example, Patent Document 1). After imaging the cracks generated on the surface of the concrete structure with a digital imager and binarizing the original image data for each pixel, the crack area is determined and the total brightness value of each pixel in the crack area is calculated. Therefore, this is used as the apparent area of the crack, and the crack area is obtained from the apparent area.

Japanese Unexamined Patent Publication No. 2003-214827

However, in this crack measurement method, the amount of data for expressing one crack becomes enormous, and since the Vixel data itself does not have information such as position and shape, the data can be directly converted to CAD. Instead, in order to display cracks on a display device or the like as digital data, it was still necessary for humans to input them into CAD.

According to the present invention, when extracting cracks generated on the surface of a concrete material from a photographed image of the surface of the concrete material, the image information of the crack portion can be extracted with high accuracy, and it is possible to obtain digital data. The purpose is to obtain a new image processing method.

The crack extraction method according to the invention according to claim 1 is a crack extraction method for extracting image information of a crack portion generated on the surface of the concrete material from a photographed image of the surface of the concrete material.
A binarization processing step of binarizing the photographed image in order to distinguish a shadow portion inside a crack from another portion, and a binarization processing step.
An isolated point removal processing step that removes isolated points that are not detected as linear parts corresponding to cracks from the binarized image, and
It is characterized by including a raster vector conversion processing step of detecting continuous dots from an image from which isolated points have been removed, generating line information, and displaying the line information on a display device.

The crack extraction method according to the invention according to claim 2 is characterized by further including a pretreatment step for extracting the contour in the image before the binarization treatment step according to claim 1. Is.

The crack extraction method according to the invention according to claim 3 displays a crack width that is distinguished and displayed on a display device according to the thickness of the extracted crack after the raster vector conversion processing step according to claim 1 or 2. It is characterized by further including a processing process.

The crack extraction method according to the invention according to claim 4 is based on the positional relationship between the extracted cracks and the photographed image after the raster vector conversion processing step according to any one of claims 1 to 3. It is characterized by further including a floating / peeling detection processing step for detecting a floating / peeling portion.

According to the present invention, when extracting cracks generated on the surface of a concrete material from a photographed image of the surface of the concrete material, the image information of the crack portion can be extracted with high accuracy, and it is possible to obtain digital data. It has the effect of

It is explanatory drawing which shows the difference between a crack and a groove generated in a concrete material, FIG. a shows the cross-sectional structure of a crack, and FIG. b shows the cross-sectional structure of a groove. It is a process drawing explaining the process of one Example of the crack extraction method of this invention. It is explanatory drawing of the image in each step of the crack extraction method of this invention. It is explanatory drawing of the image in each step of the crack extraction method of this invention. It is explanatory drawing explaining the situation of the detection of the float from an image.

Since concrete is a material that is strong against compressive force, weak against tensile force, and has low deformability, cracks essentially occur. That is, when tensile stress acts on the concrete material due to various factors to cause strain, cracks occur when the strain exceeds the deformation capacity of the concrete material. This crack has a structure completely different from that of the groove because of its formation.

FIG. 1 is an explanatory view showing a difference between a crack and a groove generated in a concrete material, FIG. 1 shows a cross-sectional structure of the crack, and FIG. 1B shows a cross-sectional structure of the groove. As shown in FIG. 1a, since the crack 11 formed in the concrete material 10 is a crack formed against the strain, the facing surfaces of the cracks intersect at an acute angle.

Therefore, the incident light from above the crack enters the inside of the crack on the facing surface of the crack, but is not reflected in the opening direction of the crack because the bottom surface itself does not exist. Therefore, since the image of the crack is detected by a black line, the difference in brightness from the surrounding surface of the other concrete material is large.

On the other hand, as shown in FIG. 1b, in the groove 12 formed in the concrete material 10, both side surfaces are substantially vertical and the bottom surface 13 exists. Therefore, when the incident light reaches the bottom surface 13, the incident light reaches the bottom surface 13. It will be reflected in the direction of the opening of the groove. Therefore, it is clear that the crack 11 is overwhelmingly larger than the groove 12 in terms of the brightness difference from the surrounding concrete material 10.

Further, the surface of the concrete material 11 has a so-called sponge-like shape in which fine bubbles and the like are formed when enlarged. Therefore, for example, even if a line imitating a crack is drawn on the surface of the concrete material by oil-based magic or the like, the contour line is not finely linear, but is a series of fine circles. As a result, the outline becomes blurry as a whole, and the difference in brightness from the surroundings does not become large.

On the other hand, since the crack 11 is a rift, its outline is a sharp straight line. Further, as described above, the difference in brightness from the surroundings is large. It has been found that the cracks 11 can be reliably distinguished by paying attention to the characteristics of these cracks 11 and performing image processing that emphasizes the difference in the characteristics, and the present invention has been made.

The present invention is a crack extraction method for extracting cracks generated on the surface of a concrete material from a photographed image of the surface of the concrete material, in order to distinguish a shadow portion inside the crack from another portion from the photographed image. A binarization process for binarizing the image, an isolation point removal process for removing isolated points that are not detected as linear parts from the binarized image, and continuous dots from the image from which the isolated points have been removed. It is provided with a raster vector conversion processing step of detecting and generating line information corresponding to a crack portion.

As a result, when extracting cracks generated on the surface of the concrete material from the photographed image of the surface of the concrete material, the image information of the crack portion can be extracted with high accuracy, and it becomes possible to obtain digital data. ..

As for the photographing of the surface of the concrete material of the present invention, since the photographed image is processed, it may be recorded as a digital image (raster image) instead of an analog image. For example, it may be taken with a so-called "digital camera". The captured image is processed to extract cracks in the captured image.

The binarization processing step of the present invention may be any one that binarizes the captured image in order to distinguish the shadow portion inside the crack from the other portion. As described above, the structure of the crack is different from the structure of the groove, and the light incident from above the crack is not reflected in the direction of the crack opening. Therefore, the image of the crack is a black linear region with a clear edge. Because it appears as, the difference in brightness from the surrounding concrete material surface is large.

Therefore, cracks can be distinguished by setting the threshold value to an extreme threshold value such as deep black or other when performing the binarization process. In addition, the crack part is clarified by performing the "inversion of gradation" before the binarization treatment and then performing the binarization treatment to show the crack part in white and the surface of other concrete materials in black. It may be transformed into.

The isolated point removing processing step of the present invention may be any one that removes isolated points that are not detected as linear portions from the binarized image. That is, for example, when there is an exposed part such as black sand on the surface of the concrete material, even if the binarization process is performed, it is converted to black (white when the gradation is reversed) like the crack part. When this is done, by removing this as an isolated point, only the crack portion can be satisfactorily extracted.

It should be noted that the dotted linear cracks may be treated as continuous cracks in the case of extremely small intermittent cracks, and therefore it is desirable to remove them from the isolated point removal treatment. Further, even if the cracks are intermittent, the directions of the cracks are calculated, and if the directions are the same and the interruptions are small, the cracks may be treated as continuous cracks. Therefore, it is also desirable to remove them from the isolated point removal treatment.

The raster / vector conversion processing step of the present invention may be any one that detects continuous dots from an image from which isolated points have been removed and generates line information. That is, the raster vector conversion is a process of converting raster data into vector data, and may be any one that detects continuous dots from a binarized image (monochrome) and generates line information. As a result, the data can be digitally processed.

In the present invention, as a preferred embodiment, a pretreatment step of extracting a contour corresponding to an edge of a crack portion in an image may be further provided before the binarization treatment step. That is, filter processing such as a Laplacian filter that extracts contours from an image using quadratic differentiation is performed. As a result, the subsequent binarization process can be performed well along a clear outline.

Before the preprocessing step of extracting the contour in the image, a noise reduction processing step of smoothing the captured image to reduce noise may be performed. That is, in the Laplacian filter processing, noise may be easily emphasized because it acts as a quadratic derivative, and this is suppressed by smoothing the image in advance before the Laplacian filter. As the noise reduction processing, a Gaussian filter or the like that performs processing that weights nearby pixel values according to the distance from the pixels using a Gaussian distribution may be used.

In another preferred embodiment of the present invention, after the raster vector conversion processing step, a crack width display processing step of distinguishing and displaying the extracted cracks on the display device according to the thickness of the extracted cracks may be further provided. That is, the extracted cracks are easily detected by measuring the width of the cracks and color-coding them to distinguish them, or by marking the crack positions with the maximum width like round dots. be able to.

Specifically, as shown in the examples described later, the cracks are color-coded according to the width as a "fine display", and the total is the length total according to the width. As a "simple display", the length in crack units is displayed, and the crack width displays the maximum value and the average value in crack units. The position of the maximum width is displayed as a circle. Further, preferably, the extracted cracks are superimposed and displayed on the photographed image of the surface of the concrete material.

In yet another preferred embodiment of the present invention, after the raster vector conversion processing step, a floating / peeling detection processing step for detecting a floating / peeling portion from the extracted crack status and an image may be further provided. Good. For example, when a hollow portion that causes floating is generated in the concrete material, the height position of the concrete material surface immediately above the hollow portion is higher than that of other concrete material surfaces.

In this case, cracks are formed on both sides of the hollow portion inside, and the height positions of the concrete material surface are different across the cracks, so that a shadow is generated at a position away from the hollow portions of these cracks. It is highly possible that the float is generated at the place sandwiched between the two cracks where these shadows are formed. In addition, with regard to peeling, since the floating progresses and becomes peeling, the peeling portion matches the floating portion. Therefore, when the floating portion leading to the peeling is obtained as described above, the peeling portion appears.

FIG. 2 is a process diagram illustrating a process of an embodiment of the crack extraction method of the present invention. FIG. 3 is an explanatory diagram of images in each step of the crack extraction method of the present invention. FIG. 4 is an explanatory diagram of images in each step of the crack extraction method of the present invention. FIG. 5 is an explanatory diagram illustrating a situation in which floating is detected from an image.

As shown in FIG. 2, in the crack extraction method of the present embodiment, as the first step, as shown in the explanatory view of FIG. 3A, the concrete wall surface of the target is photographed with a digital camera to obtain the photographed image 31. The photographing step 21 was performed. In this embodiment, shooting was performed according to the setting accuracy of one pixel of the camera. Table 1 is a table showing the relationship between the crack extraction accuracy of the captured image 31 and the imaging range.

Specifically, the shooting range was set according to the number of pixels (24.1 million pixels) of the camera (Nikon D7100) used. For example, in order to be able to extract 0.2 mm cracks, the number of pixels is 6,000 x 4,000 pixels, so if the shooting range is 0.2 mm per pixel, the width is 6,000 x 0.2 mm. Since the length is 4,000 x 0.2 mm, it is shown that the shooting range should be 1.2 m x 0.8 m.

When a digital camera having a different number of pixels is used, the shooting range may be set. Regarding the setting accuracy of crack extraction, the crack extraction accuracy (that is, the vertical and horizontal lengths per pixel) is set to 0.1 mm or 0 in consideration of the number of pixels of the digital camera used for shooting and the shooting range. It may be changed to .3 mm or the like.

As shown in FIG. 2, the image information taken by the digital camera was input to an image processing means such as a PC, and the steps after the concrete wall surface photographing step 21 were performed while displaying the processed image on the display. As the next step, the pretreatment step 29 was performed. In this embodiment, before the pre-processing 29 and the binarization processing 22, as shown in the explanatory diagram of FIG. 3b, the “gradation reversal” of the photographed image 31 is performed to brighten the dark part such as the cracked portion. However, the cracked part was clarified by darkening the bright part on the surface of other concrete materials.

As a specific pretreatment 29, as shown in FIG. 3C, a process of extracting the contour corresponding to the edge of the crack in the captured image 31 was performed. That is, as a process for extracting the contour in the image, a filter process such as a Laplacian filter for extracting the contour from the image using the second derivative was performed.

In the Laplacian filter processing as the preprocessing 29, noise may be easily emphasized because it acts as a quadratic derivative. Before the Laplacian filter, noise reduction processing 28 for smoothing the image is performed in advance. May be good. As the noise reduction processing 28, a Gaussian filter or the like that performs processing that weights nearby pixel values according to the distance from the pixels using a Gaussian distribution may be used. This improves edge detection. The noise reduction processing step 28 and the pretreatment step 29 may be omitted in some cases.

As the next step, the preprocessed photographed image was binarized 22 as shown in FIG. 4d. That is, the binarization process 22 was performed in order to distinguish the shadow portion inside the crack 32 from the other portion from the captured image 31. As described above, in the structure of the crack 32, the light incident from above the crack is not reflected in the crack opening direction, so that the image of the crack is a white region with a clear edge (inversion of gradation). If not, it is displayed as "black area").

Therefore, the difference in brightness from the surface of other concrete materials is large, and when the binarization process is performed, the crack 32 is improved by setting the threshold value to pure white (dark black) or an extreme threshold value other than that. Can be distinguished into.

Next, an isolated point removing process 23 was performed to remove isolated points that were not detected as linear portions from the binarized image (Fig. 4e). In the isolated point removal treatment 23, for example, when there is an exposed portion such as dark sand on the surface of the concrete material, when the binarization treatment is performed, it is white (inverted gradation) like the crack portion. If not, when it is converted to black), the crack 32 portion can be satisfactorily extracted by removing it as an isolated point 33.

It should be noted that the dotted linear cracks are preferably treated as continuous cracks in the case of extremely small intermittent cracks, and therefore it is desirable to remove them from the isolated point removal treatment. Further, even if the cracks are intermittent, the directions of the cracks are calculated, and if the directions are the same and the interruptions are small, the cracks may be treated as continuous cracks. Therefore, it is also desirable to remove them from the isolated point removal treatment.

Therefore, the expansion / contraction treatment 24 is preferably performed. That is, an image operation for extending the end of a line segment in the direction of the line segment is performed (for example, 3 to 5 pixels), and when pixels of adjacent line segments come into contact with each other, they are treated as one line segment. Just do it. This makes it possible to remove it from the isolated point removal process.

Next, a raster vector conversion process 25 is performed to detect continuous dots from the image from which isolated points have been removed and generate line information. The raster / vector conversion process 25 is a process of converting raster data into vector data, and may be any process that detects continuous dots from a binarized image (monochrome) and generates line information. As a result, the data can be digitally processed.

By this raster vector conversion process 25, it is possible to more easily perform a crack width display process such as color-coding the extracted cracks according to the size of the width. For example, as shown in Table 2, when the setting accuracy, which is the length and width of one pixel, is 0.2 mm, the width classification is 0.2 mm, 0.4 mm, 0. It is possible to color-code the thickness of cracks of 6 mm, 0.8 mm, 1.0 mm or more into any five colors (enlarged view of FIG. 4f). As the crack width display process 26 for distinguishing and displaying according to the thickness of the extracted cracks, the position display of the maximum width may be performed as a simple display.

Finally, the floating / peeling detection process 27 for detecting the floating / peeling portion was performed from the extracted crack status and the image. As shown in FIG. 5, when a hollow portion 52 that causes the floating 51 is generated in the concrete material 50, the surface of the concrete material immediately above the hollow portion 52 is another concrete material. The height position rises above the surface.

In this case, cracks 53 on both sides are formed on both sides of the hollow portion 52 inward, and the height positions of the concrete material surface are different across the cracks 53, so that shadows are cast at positions away from the hollow portions 52 of these cracks 53. Occurs.

In this case, depending on the size of the hollow portion 52 and the like, a thick crack 54 is formed between the cracks 53 on both sides of the hollow portion 52 as shown in FIG. A, and a hollow portion 54 is formed as shown in FIG. The cracks 53 on both sides of the portion 52 are thickened, and thin cracks 55 are often generated at positions away from the cracks 53 according to the formation of the thick cracks.

In either case of Fig. A and Fig. B, there is a high possibility that the float is generated at the portion sandwiched between the two cracks. In particular, when the cracks are displayed by superimposing them on the captured image of the concrete material surface, if it can be recognized that shadows are generated on the side of the cracks where the cracks are separated from each other, there is a hollow space between the cracks. There is a high possibility that the portion 52 is generated, and the sandwiched region thereof becomes floating.

In addition, with regard to peeling, since the floating progresses and becomes peeling, the peeling portion matches the floating portion. Therefore, when the floating portion leading to the peeling is obtained as described above, the peeling portion appears.

It should be noted that each process of the process diagram shown in FIG. 2 can be performed by automatic processing using an application on a PC. In addition, individual settings may be set in advance, or recommended conditions may be made selectable in advance so that the operator can select them. Further, regarding the floating / peeling detection process 27 that detects the floating / peeling portion, since there is a high possibility that the floating / peeling detection process 27 is sandwiched between the two cracks, at present, the portion sandwiched between the two cracks. Is operating a discrimination means for searching in the image.

At present, the confirmer determines whether or not the identified portion is floating or peeled off. However, it is at the stage of education so that the determination of floating / peeling can be automatically performed by deep learning by AI.

10 ... Concrete material,
11 ... crack,
12 ... groove,
13 ... bottom,
21 ... Concrete wall photography,
22 ... Binarization,
23 ... Isolation point removal processing,
24 ... Expansion / contraction processing,
25 ... Raster vector conversion process,
26 ... Crack width display processing,
27 ... Floating / peeling detection processing,
28 ... Noise reduction processing,
29 ... Pretreatment,
31 ... Photographed image,
32 ... crack,
33 ... isolated point,
51 ... Float,
52 ... Hollow part,
53 ... Cracks on both sides,
54 ... Thick crack,
55 ... thin cracks,

Claims (4)

It is a crack extraction method that extracts image information of a crack portion generated on the surface of the concrete material from a photographed image of the surface of the concrete material.
A binarization processing step of binarizing the photographed image in order to distinguish a shadow portion inside a crack from another portion, and a binarization processing step.
An isolated point removal processing step that removes isolated points that are not detected as linear parts corresponding to cracks from the binarized image, and
A crack extraction method comprising a raster vector conversion processing step of detecting continuous dots from an image from which isolated points have been removed, generating line information, and displaying the line information on a display device. The crack extraction method according to claim 1, further comprising a pretreatment step of extracting the contour in the image before the binarization treatment step. The crack extraction method according to claim 1 or 2, further comprising a crack width display processing step of distinguishing and displaying on the display device according to the thickness of the extracted crack after the raster vector conversion processing step. .. Claims 1 to 1, further comprising a floating / peeling detection processing step of detecting a floating / peeling portion from the positional relationship between the extracted cracks and the photographed image after the raster / vector conversion processing step. The crack extraction method according to any one of 3.

Images