JP5894012B2 - Method for detecting linear deformation on concrete surface - Google Patents

Description

  The present invention relates to a method for detecting a linear deformation of a concrete surface such as a tunnel wall surface. In this specification, the linear deformation is a crack generated on the concrete surface or a chalk spot drawn on the concrete surface.

In order to prevent the concrete pieces from peeling off or falling from the concrete surface of the tunnel, it is very important to correctly determine the soundness of the tunnel. The peeling and falling of the concrete pieces are detected by the inspector by visual inspection of various deformations such as closed cracks, cross cracks, and parallel cracks that have occurred on the concrete wall of the tunnel lining surface. A hitting sound diagnosis method for hitting and determining the degree of danger from the hitting sound has been widely implemented. As for deformation detection, various deformation detection systems have been developed in place of workers' visual inspection. As for the diagnosis method, various concrete internal defect diagnosis systems have been developed in place of the hammering sound diagnosis method by an inspector.

A concrete internal defect diagnosis system using a non-contact inspection method is an extremely effective method as a quantitative method in place of the conventional hammering sound diagnosis. However, the non-contact diagnosis method does not work on the concrete wall surface, but only on the concrete wall surface in a specific diagnosis area where there is a possibility that a defect exists in the concrete.

Patent Document 1 discloses a crack detection method using wavelet transform. This crack detection method is basically identification based on the luminance distribution, but the detection accuracy of cracks in the vicinity of the boundary between the bright part and the dark part decreases, and parts that are not clearly cracked, such as joints and cables, are judged as cracks. There are problems such as. In addition, there is a trouble that a threshold table must be prepared in advance.

Patent Document 2 describes a sub-pixel processing technique for measuring a fine crack width after identifying a background and a crack part while comparing a background color density database with a color density of a target point. This crack detection method has a problem that the detection accuracy of a crack portion is lowered under a complicated background such as dirt or water leakage.

In Patent Document 3, a plurality of types of two-dimensional edge filters having different directions and widths are individually applied to an image of a concrete surface, and pixel values after application of a plurality of types of edge filters are compared for each pixel. Although a crack detection device including a crack feature amount extraction unit that selects the maximum pixel value as a crack feature amount is disclosed, there is a problem in practicality. In other words, there is almost no case where information such as the distance from the camera, the number of pixels of the photographing camera, and the assumed crack width is known in advance, so the optimum filter size must be examined each time. It is because there is a troublesome work. In addition, since the edge filter extracts feature amounts as cracks in noise edges such as backgrounds and plates, there is a problem in the accuracy of crack detection.

By the way, although cracks are the most important information in tunnel soundness evaluation, cracks generated on the concrete wall of the tunnel lining surface also have cracks and linear noise that do not affect the soundness judgment of the concrete structure. According to inspection civil engineers and field inspection workers, significant cracks that are used to judge the soundness of concrete structures are: thick cracks and thin cracks that branch off from the ends of the thick cracks. It is. Therefore, the inventor of the present application pays attention to the characteristic of the crack, which is a darker linear region than the periphery, and grasps the crack as a region constituted by the central axis and its outline. Then, as described in Non-Patent Document 1, an edge pixel which is a constituent point of a crack outline is obtained by subjecting a pre-processed image obtained by performing pre-processing such as noise removal to a photographed image of a concrete surface. We have developed a crack detection method characterized by extracting sub-pixels with sub-pixel accuracy. By this method, it was demonstrated that a thick crack, a thin crack extending from an end portion of the thick crack, and a branched thin crack can be detected. By the way, extracting crack composing points with sub-pixel accuracy by edge extraction processing by secondary differentiation simultaneously extracts edge pixels that are composing points of linear noise such as cracks. For this reason, when cracks and linear noise exist on the concrete surface, a plurality of crack candidates in which a contour line is constituted by edge pixels extracted by performing a second-order differential process on the preprocessed image are also detected. It will be. However, this method can easily and reliably identify significant cracks, non-significant cracks, and linear noise that are information for determining the soundness of concrete structures from a plurality of detected crack candidates. There is a problem that it is difficult.

In addition, there is a white line of chalk drawn on the concrete wall such as the tunnel lining surface that surrounds the deformations such as cracks and water leaks that were spotted by field workers, and this is automatically and reliably detected. A method has not been developed. In many cases, detecting the choke line or the choke portion surrounded by the choke line is directly connected to the detection of the deformation. However, when tracing the deformation periphery with a choke line, for example, it is often drawn so as to overlap with a crack part which is a black line, so that the white line may erase the black crack. The choke line is useful for visual inspection, but is an obstacle to detection of cracks by image processing. However, it is very difficult to detect a crack after removing the choke portion as noise. Therefore, it is conceivable that the choke portion is positively detected and used as one piece of information for judging the soundness level of a concrete structure such as a tunnel, that is, significant information. However, a method for detecting the chalk portion drawn on the concrete surface by image processing has not been developed yet.

Japanese Patent No. 4006007 Japanese Patent No. 4186117 JP 2011-242365 A

"Development of high-precision image processing method for extracting tunnel deformation", Proceedings of Railway Cybernet Symposium, December 1, 2011

A first problem to be solved by the present invention is a concrete in which a linear deformation is reliably extracted by image processing a photographed image of the concrete surface from a concrete surface where linear deformation and linear noise exist. It is to provide a surface linear deformation detection method.
A second problem to be solved by the present invention is to process a captured image of the concrete surface from a concrete surface where cracks and linear noise exist, and to extend or branch from a thick crack and a tip of the thick crack. It is intended to provide a method for detecting a linear deformation of a concrete surface that reliably extracts a thin crack.
A third problem to be solved by the present invention is to provide a method for detecting a linear deformation on a concrete surface, which reliably extracts a thick chalk line drawn on the concrete surface by subjecting a photographed image of the concrete surface to image processing. That is.

The concrete surface linear deformation detection method for solving the first problem includes a preprocessing step of obtaining a preprocessed image by performing correction and noise removal on an original image that is a captured image of the concrete surface, the preprocessing A step of applying a second-order differential edge filter to extract a linear deformation candidate after smoothing the processed image, performing a first-order differentiation on the pre-processed image, It consists of a step of removing linear noise and extracting a thick linear deformation and a thin linear deformation, and a step of extracting a significant linear deformation that serves as information for determining the soundness of a concrete structure. Features.

The concrete surface linear deformation detection method for solving the second problem includes a preprocessing step of obtaining a preprocessed image by performing correction and noise removal on an original image that is a captured image of the concrete surface, the preprocessing A step of extracting a crack candidate by applying a second-order differential edge filter after smoothing the processed image, and performing a first-order differentiation on the pre-processed image to remove linear noise from the crack candidate And a step of extracting a thick crack and a thin crack, and a step of extracting a significant crack which is information for soundness judgment of a concrete structure.

The linear deformation detection method for a concrete surface that solves the third problem is a preprocessing step of obtaining a preprocessed image by performing correction and noise removal on an original image that is a captured image of the concrete surface,
Applying a second-order differential edge filter to extract a choke line candidate after performing a smoothing process on the pre-processed image; performing a first-order differentiation on the pre-processed image; And a step of extracting a thick choke line and a thin choke line, and a step of extracting a significant choke line which is information for judging the soundness of a concrete structure.

According to the present invention, linear deformation of a concrete surface with dirt or unevenness can be easily and reliably extracted from a photographed image of the concrete surface, and significant information for determining the soundness of a concrete structure can be provided.
According to the present invention, it is possible to easily and reliably extract a crack on a concrete surface having dirt or unevenness from a photographed image of the concrete surface, and to provide significant information on the soundness judgment of the concrete structure.
According to the present invention, the chalk line drawn on the concrete surface with dirt or unevenness can be easily and reliably extracted from the photographed image of the concrete surface, and can provide significant information on the soundness judgment of the concrete structure. .

It is a basic flowchart of the crack extraction process of the concrete surface which concerns on this invention. It is a flowchart which shows an example of the crack extraction process of the concrete surface which concerns on this invention. It is explanatory drawing of a thick crack, the significant thin crack couple | bonded with this, and the non-significant thin crack which is not couple | bonded. It is a figure which shows a thick crack and the thin crack couple | bonded with this. It is a figure which shows an example of the differentiation operator for a primary differentiation. It is a flowchart which shows an example of the process of the chalk line extraction drawn on the concrete surface based on this invention. It is an example of the pre-processed image of the picked-up image of the concrete surface where the chalk line was drawn. It is an example of the image which gave the Gaussian process to the pre-processed image, and also gave the Laplacian. It is an example of the concrete surface image containing the extracted chalk wire. It is a block diagram of the detection apparatus of the linear deformation of the concrete surface which concerns on this invention.

A smoothing process is performed on a preprocessed image obtained by performing correction and noise removal on an original image that is a captured image of a concrete surface using a Gaussian smoothing filter, and then a Laplacian filter is applied to extract crack candidates. Next, first-order differentiation is performed on the preprocessed image to remove linear noise from the crack candidates and to extract a thick crack and a thin crack. Finally, a significant thick crack and a thin crack that is a branch of the significant crack are combined by a connection process to extract a crack that is significant information for determining the soundness of a concrete structure.

The concrete surface deformation detection method of the present invention is carried out using a deformation detection apparatus as shown in FIG. The deformation detection device includes a CPU 1, a ROM 2, a RAM 3, an input unit 4, an image data input unit 5, an output unit 6, and a display unit 5 such as a liquid crystal monitor. The input unit 4 is various input devices such as a keyboard for inputting various data for extracting deformation and a mouse for giving an operation command to the keyboard. The ROM 2 stores various programs for calculation and control in advance. The RAM 3 stores digital original image data obtained by imaging the concrete surface input from the image data input unit 5 to the CPU 1, and preprocessed image data acquired by performing correction and noise removal on the original image. The RAM 3 stores in advance various data necessary for extracting cracks and choke lines by image processing, that is, various threshold values and parameters.

(Basic flow of linear deformation extraction)
As shown in FIG. 1, the basic flow of the deformation detection method of the present invention for extracting cracks, which are typical linear deformations, is a preprocessed image acquisition step S1, a crack candidate extraction step S2, a crack. Extraction and linear noise removal step S3, a thin crack extraction step S4 combined with a thick crack, and a significant crack extraction step S5.

Step S1 is a step of obtaining a preprocessed image by performing correction and noise removal on the original image that is a captured image of the concrete surface.

In step S2, smoothing processing is applied to the preprocessed image, and then second-order differentiation is applied to extract edge points with subpixel accuracy, and crack candidates that are linear regions surrounded by the extracted edge points are extracted. It is a process to do. The crack candidates include three types of areas: thick cracks, thin cracks, and linear noise. Step S2 is a process for specifying only the edge of the region, paying attention to the feature of cracks that are darker regions that are darker than the surroundings.

Step S3 applies the first derivative to the preprocessed image to obtain the feature value of the edge point of the crack candidate and the first derivative value used for the evaluation when obtaining the feature value of the edge point. A process of identifying a thick crack, a thin crack, and a linear noise from the crack candidates by performing a hysteresis threshold process. In this process, as shown in an embodiment to be described later, a first derivative is applied to the preprocessed image to extract a pixel having a local maximum primary differential value, and a hysteresis is applied to the maximum primary differential value. A step of extracting a thick crack, a thin crack, and linear noise by performing threshold processing, and the primary differential value used for the evaluation when obtaining the feature value of the edge point is the maximum primary differential value That is. In the embodiment, the primary differential value used for the hysteresis threshold processing is the primary differential value used for evaluation when obtaining the feature value of the edge point. A thick crack is a significant crack. On the other hand, thin cracks include significant cracks and non-significant cracks. The hysteresis threshold process is a process of comparing whether or not the primary differential value exceeds the upper limit threshold High and the lower limit threshold Low, and specifying three types of regions according to the comparison result. That is, a region surrounded by edge points having a first derivative value exceeding the upper threshold value High is a thick crack, and a region surrounded by edge points having a first derivative value not exceeding the upper threshold value High but exceeding the lower threshold value Low is A region surrounded by an edge point having a fine differential and a first-order differential value not exceeding the lower threshold value Low is linear noise.

In step S3, thick cracks, thin cracks, and linear noise are different in area blackness and area edge sharpness, and area blackness and area edge sharpness are directly converted to the first derivative value. This processing is focused on being reflected in By this process, a thick linear deformation and a thin linear deformation are reliably extracted, and linear noise is reliably removed. At this stage of processing, the removed linear noise is deleted from the display screen.

Step S4 is a process of extracting significant thin cracks from the thin cracks extracted in step S3, and includes a step of calculating a thick crack observation point, a thin crack observation point, and a thick crack observation. Compare the distance between the point and the observation point of the thin crack and the direction of the center axis of the thick crack and the thin crack within the specified value. And a step of determining that the crack is not significant. The observation points for the thick cracks are the four edge points at both ends of the central axis and at both ends at the maximum width in the direction perpendicular to the central axis. In addition, the observation points for the thin cracks are the edge points at both tips in the central axis direction. At this stage of processing, insignificant thin cracks are deleted from the display screen.

Step S5 is a process of combining the thick crack identified in step S3 and the significant thin crack identified in step S4 on the display screen. Significant cracks for judging the soundness of concrete structures are extracted by combining significant cracks with thick cracks.

Hereinafter, the crack detection method of the first embodiment will be described in more detail with reference to the flowchart of FIG.

(Pre-processed image acquisition)
The CPU 1 reads an original image from the RAM 3 according to a program stored in the ROM 2, performs correction and noise removal on the original image, and acquires a preprocessed image (S101). The correction is correction of uneven illuminance due to shading. To remove noise, first, periodic line segments such as molds are removed from the original image by frequency analysis. Next, spotted unevenness is removed from the original image by frequency analysis. Finally, dirt is removed from the original image by frequency analysis. In this manner, a preprocessed image obtained by performing correction and noise removal processing on the original image is acquired and stored in the RAM 3. The frequency analysis is image processing that applies an FFT filter to an original image. The parameters of the FFT filter are selected to the optimum values according to the removal target.

(Extraction of crack candidates)
Next, the CPU 1 sets the smoothing parameter of the Gaussian filter stored in the RAM 3 to a larger value in accordance with the program stored in the ROM 2 (S102). Next, the CPU 1 reads out the preprocessed image stored in the RAM 3 according to the program stored in the ROM 2, applies a Gaussian filter, and smoothes it (S103).

Subsequently, according to the program stored in the ROM 2, the CPU 1 applies a Laplacian filter to the smoothed preprocessed image and applies a second-order differential process to extract the edge points of the cracked area with sub-pixel accuracy. (S104). The edge points extracted by this processing are the edge points constituting the contour lines of the crack candidates and the edge points constituting the contour lines of the linear noise candidates. Therefore, the crack candidate and the linear noise candidate are easily and reliably extracted from the preprocessed image by the process of step S104. At this stage, since the crack candidate and the linear noise candidate cannot be distinguished, all the regions surrounded by the extracted edge points are called crack candidates. These crack candidates are assigned an identification number and stored in the RAM 3. The secondary differentiation process is performed using a Laplacian filter, but is not necessarily limited to this. When the Laplacian filter is applied to the smoothed preprocessed image, the edges are sharpened, and there is an advantage that the edge points can be easily extracted. In the series of steps S102 to S104, crack candidates are reliably extracted from the preprocessed image.

(Crack extraction and linear noise removal)
Next, the CPU 1 identifies the feature amount of the edge point of the crack candidate (S105). That is, the CPU 1 applies the first derivative to the preprocessed image stored in the RAM 3 according to the program stored in the ROM 2 to obtain the first derivative value of the edge point of the crack candidate. The primary differentiation applied to the preprocessed image is a process of applying the primary differentiation operator in a direction perpendicular to the central axis of the choke line candidate region, not in the horizontal direction or the vertical direction. This is a process that takes into account speed and practical accuracy.

By the way, in a digital image, since data are arranged in a discrete manner at regular intervals, a true differential operation cannot be performed. For this reason, the differentiation is approximated by an operation for obtaining a difference between adjacent pixels. A set of coefficients expressing the calculation of adjacent pixels for performing a differential calculation is a differential operator. The numerical sequence of the differential operator performs an inner product calculation in which surrounding pixels are multiplied by a coefficient corresponding to the position to obtain a sum. Since the edge direction at a certain coordinate point is obtained by the vector orientation θ by calculation, in order to obtain the first-order differential in the perpendicular direction by 90 °, it is rotated in 8 directions in increments of 45 ° as shown in FIG. The first-order differential operator is obtained by applying it to the processing target image. The central axis of the crack candidate is calculated by the CPU 1 for the target crack candidate before performing the first-order differentiation. And CPU1 extracts the pixel which has the largest differential value locally from the calculated | required differential value. As a result, this pixel is a pixel at or near the edge point extracted in step S104.

Since the first derivative value represents the degree of change of the luminance value of the pixel, that is, the sharpness of the contour line, the characteristic of the crack candidate that is a black region compared to the surrounding area is quantified by the process of step S205. It was done. This is because the process S205 focuses on the fact that the luminance value indicating the darkness of the area and the sharpness of the edge of the area are directly reflected in the first-order differential value.

  Subsequent to step S105, the CPU 1 performs a hysteresis threshold process, and identifies cracks and noise from the crack candidates (S106). The hysteresis threshold process is a process of comparing whether or not the primary differential value exceeds the upper limit threshold High and the lower limit threshold Low, and specifying three types of regions based on the comparison result. That is, a region surrounded by edge points having a first derivative value exceeding the upper threshold value High is a thick crack, and a region surrounded by edge points having a first derivative value not exceeding the upper threshold value High but exceeding the lower threshold value Low is A region surrounded by an edge point having a fine differential and a first-order differential value not exceeding the lower threshold value Low is linear noise. This is a process that focuses on the fact that cracks and line noise differ in brightness values that represent the darkness of dark areas and the sharpness of contour lines compared to the surroundings. Cracks and thin cracks are reliably extracted, and linear noise is reliably removed. At this stage of processing, the removed linear noise is deleted from the display screen. FIG. 3 shows an example of a display screen before the linear noise is removed. A thick crack 11 and seven thin cracks 12, 13, 14, 15, 16, 17 are displayed. FIG. 4 shows an example of a display screen from which linear noise has been removed, in which a thick crack 11 and three significant thin cracks 12, 13, and 14 are displayed. A small circle on the outline of the crack is an edge point given for explanation.

(Extraction of thin cracks connected to thick cracks)
Subsequent to step S106, the CPU 1 identifies a thin crack to be combined with the thick crack (S107). In step S106, significant thick cracks were identified, but thin cracks include those that are not significant. Even thin cracks can lead to concrete deterioration and others can not. Therefore, the present inventor made not only a thick crack, but also a thin crack branched from the thick crack, based on the knowledge of a tunnel expert, as a significant crack that becomes information on the soundness judgment of the concrete structure. . Therefore, the process of step S107 is necessary.

In step S <b> 107, the CPU 1 selects and processes a thin crack having a distance from the thick crack and a direction of the central axis within a predetermined value by threshold processing, and extracts this as a thin crack coupled to the thick crack. That is, in the threshold processing, the CPU 1 determines whether the distance between the observation points of the thick crack and the thin crack is within a predetermined value, and determines whether the angle between the center axis of the thick crack and the thin crack is within the predetermined value, If both determination results are YES, the thin crack is identified as a crack to be joined by a thick crack. The observation point for the thin crack is the end closest to the thick crack. The predetermined value of the distance between the observation points of the thick crack and the thin crack and the predetermined value of the angle between the central axis of the thick crack and the thin crack are set based on the knowledge of a tunnel expert and stored in the RAM 3 in advance. Yes.

As shown in FIG. 3, the observation points for the thick crack 11 are at the four edge points of both the leading ends 11 a and 11 b in the longitudinal direction and both the leading ends 11 c and 11 d in the direction perpendicular to the central axis at the maximum width. is there. Further, the thin cracks 12, 13, 14, 15, 16, and 17 are edge points 12a and 12b, 13a and 13b, 14a and 14b, 15a and 15b, 16a, 17a, and 18a at the longitudinal ends. These observation points are calculated by the CPU 1 prior to threshold processing.

The process of step S107 will be described in detail with reference to FIG.
The distance between the observation point 12a of the thin crack 12 and the observation point 11b of the thick crack 11 is within a predetermined value, but the angle between the central axis of the thin crack 12 and the central axis of the thick crack 11 exceeds the predetermined value. Therefore, the thin crack 12 is usually judged as an insignificant crack. However, if the width of the thin crack 12 exceeds the threshold, it is judged that the crack is highly related to the thick crack 11, and the thin crack 12 is It is determined that the crack extends from the thick crack 11b.

Since the distance between the observation point 13a of the thin crack 13 and the observation point 11c of the thick crack is within a predetermined value, the thin crack 13 is a crack branched from the tip 11c perpendicular to the central axis of the thick crack 11. judge.

Since the distance between the observation point 14a of the thin crack 14 and the observation point 11a of the thick crack is within a predetermined value, it is determined that the thin crack 14 is a crack extending from the tip 11a of the thick crack 11 in the central axis direction. .

The distance between the observation point 15a of the thin crack 15 and the observation point 11a of the thick crack exceeds a predetermined value, and the distance between the observation point 15b of the thin crack 15 and the observation point 11c of the thick crack is a predetermined value. Therefore, the thin crack 15 is not connected to the thick crack. Therefore, it is determined that the thin crack 15 is not significant.

The distance between the observation point 16a of the thin crack 16 and the observation point 11a of the thick crack 11 is within a predetermined value, but the angle between the central axis of the thin crack 16 and the central axis of the thick crack 11 exceeds the predetermined value. Therefore, the thin crack 16 is not connected to the thick crack. The width of the thin crack 16 is also below the threshold value. Therefore, it is determined that the thin crack 16 is not significant.

The distance between the observation point 17a of the thin crack 17 and the observation point 11a of the thick crack 11 exceeds a predetermined value. The distance between the observation point 17a of the thin crack 17 and the observation point 11d of the thick crack 11 also exceeds a predetermined value. Therefore, it is determined that the thin crack 17 is not significant.

The distance between the observation point 18a of the thin crack 18 and the observation point 11a of the thick crack 11 exceeds a predetermined value. Therefore, it is determined that the thin crack 18 is not significant.

(Extraction of significant cracks)
Subsequent to step 107, the CPU 1 performs a process of connecting the thin crack extracted in step S107 to a thick crack, that is, a significant crack is extracted by the connecting process (S108). Of course, independent thick cracks without any connected thin cracks are also significant cracks.

Following step S107, the CPU 1 determines whether or not the crack has been extracted without omission (S109). This determination process is a determination as to whether or not all crack region extraction processes that have been performed while changing the smoothing parameter from a large value to a planned minimum value have been completed. Therefore, if the extraction of cracks is completed for the preprocessed image smoothed by the Gaussian filter with the smoothing parameter set to the minimum value, the determination is YES and the detection of the cracks ends.

If the determination result in step S109 is YES, the crack extraction process ends, and if NO, the CPU 1 masks the crack extracted in step S108 on the preprocessed image (S110).

Subsequently, the CPU 1 resets the smoothing parameter of the Gaussian filter stored in the RAM 3 to be smaller than the value set in step S102 according to the program stored in the ROM 2 (S111). Next, the CPU 1 reads out the preprocessed image stored in the RAM 3 according to the program stored in the ROM 2 and applies a Gaussian filter in which the smoothing parameter is set smaller, and smoothes it (S103). Subsequently, the CPU 1 applies a Laplacian filter to the smoothed preprocessed image to perform second-order differential processing, extracts edge points of the cracked area with subpixel accuracy (S104), and newly selects crack candidates. To extract. Thereafter, the CPU 1 performs the processes of steps S105 to S108 in accordance with a program stored in the ROM 2. And if it determines with YES by the process of process S105, the detection of a crack will be complete | finished.

The second embodiment is a choke line detection method implemented using the deformation detection apparatus as shown in FIG. 10, and will be described in detail below with reference to FIG. 6 showing the flow of processing.

(Pre-processed image acquisition)
The CPU 1 reads an original image from the RAM 3 according to a program stored in the ROM 2, performs correction and noise removal on the original image, and acquires a preprocessed image (S201). The correction is correction of uneven illuminance due to shading. To remove noise, first, periodic line segments such as molds are removed from the original image by frequency analysis. Next, spotted unevenness is removed from the original image by frequency analysis. Finally, dirt is removed from the original image by frequency analysis. In this manner, a preprocessed image obtained by performing correction and noise removal processing on the original image is acquired and stored in the RAM 3. The frequency analysis is image processing that applies an FFT filter to an original image. The parameters of the FFT filter are selected to the optimum values according to the removal target. FIG. 7 shows an example of the preprocessed image.

(Extraction of chalk line candidates)
Next, the CPU 1 sets the smoothing parameter of the Gaussian filter stored in the RAM 3 to a larger value in accordance with the program stored in the ROM 2 (S202). Next, the CPU 1 reads out the preprocessed image stored in the RAM 3 according to the program stored in the ROM 2, applies a Gaussian filter, and smoothes it (S203).

Subsequently, according to the program stored in the ROM 2, the CPU 1 applies a Laplacian filter to the smoothed preprocessed image to perform a second order differentiation process, and the edge points of the choke linear area are subpixel-accurate. Extract (S204). The edge points extracted by this processing are the edge points constituting the outline of the choke line candidate and the edge points constituting the outline of the linear noise candidate. Therefore, the choke line candidate and the linear noise candidate are easily and reliably extracted from the preprocessed image by the process of step S204. At this stage, since the choke line candidate and the linear noise candidate cannot be distinguished, all the regions surrounded by the extracted edge points are called choke line candidates. These choke line candidates are assigned identification numbers and stored in the RAM 3. The secondary differentiation process is performed using a Laplacian filter, but is not necessarily limited to this. When the Laplacian filter is applied to the smoothed preprocessed image, the edges are sharpened, and there is an advantage that the edge points can be easily extracted. The choke line candidates are reliably extracted from the preprocessed image by the series of steps S202 to S204 described above. FIG. 8 shows an example of an image obtained by performing Gaussian processing on the preprocessed image and further applying Laplacian.

(Choke line extraction and linear noise removal)
Next, the CPU 1 identifies the feature amount of the edge point of the choke line candidate (S205). That is, the CPU 1 applies the first derivative to the preprocessed image stored in the RAM 3 according to the program stored in the ROM 2 and extracts a pixel having the largest first derivative value locally. The primary differentiation applied to the preprocessed image is a process of applying the primary differentiation operator in a direction perpendicular to the central axis of the choke line candidate region, not in the horizontal direction or the vertical direction. This is a process that takes into account speed and practical accuracy.

By the way, in a digital image, since data are arranged in a discrete manner at regular intervals, a true differential operation cannot be performed. For this reason, the differentiation is approximated by an operation for obtaining a difference between adjacent pixels. A set of coefficients expressing the calculation of adjacent pixels for performing a differential calculation is a differential operator. The numerical sequence of the differential operator performs an inner product calculation in which surrounding pixels are multiplied by a coefficient corresponding to the position to obtain a sum. Since the edge direction at a certain coordinate point is obtained by the vector orientation θ by calculation, in order to obtain the first-order differential in the perpendicular direction by 90 °, it is rotated in 8 directions in increments of 45 ° as shown in FIG. The first-order differential operator is obtained by applying it to the processing target image. The central axis of the choke line candidate is calculated by the CPU 1 for the target choke line candidate before performing the first-order differentiation. And CPU1 extracts the pixel which has the largest differential value locally from the calculated | required differential value. As a result, this pixel is a pixel at or near the edge point extracted in step S204.

Since the first derivative value represents the degree of change in the luminance value of the pixel, that is, the sharpness of the contour line, the feature of the choke line candidate that is a white region compared to the periphery is numerically expressed by the process of step S205. It was made. This is because the process S205 focuses on the fact that the brightness value indicating the whitishness of the region and the sharpness of the edge of the region are directly reflected in the first-order differential value.

  Subsequent to step S205, the CPU 1 performs hysteresis threshold processing to identify the choke line and the linear noise from the choke line candidates (S206). The hysteresis threshold process is a process of comparing whether or not the primary differential value exceeds the upper limit threshold High and the lower limit threshold Low, and specifying three types of regions based on the comparison result. That is, a region surrounded by edge points having a primary differential value exceeding the upper threshold value High is a thick choke line, and a region surrounded by edge points having a primary differential value not exceeding the upper threshold value High but exceeding the lower threshold value Low. The area surrounded by the thin choke line and the edge point having the first differential value not exceeding the lower threshold value Low is linear noise. This is a process that pays attention to the difference between the brightness value representing the whitishness of the whitish line-shaped region and the sharpness of the contour line in the choke line and the line noise compared to the surroundings. Thick choke lines and thin choke lines are reliably extracted, and linear noise is reliably removed. At this stage of processing, the removed linear noise is deleted from the display screen.

(Extraction of thin chalk lines connected to thick chalk lines)
Subsequent to step S206, the CPU 1 specifies a thin choke line coupled to a thick choke line (S207). In step S206, a significant thick choke line is identified, but the thin choke line also includes linear noise. Here, the thick chalk line refers to a thickness that can be drawn when the inspector normally draws on the tunnel wall surface with chalk. Therefore, an infinitely thick choke line is not included in a significant thick choke line, but rather is classified as noise.

In step S207, the CPU 1 selects a thin choke line whose distance from the thick choke line and the direction of the central axis are within a predetermined value by threshold processing, and extracts this as a thin choke line coupled to the thick choke line. That is, in the threshold processing, the CPU 1 determines whether the distance between the observation point of the thick choke line and the observation point of the thin choke line is within a predetermined value, and the angle between the central axis of the thick choke line and the thin choke line is predetermined. It is determined whether or not the value is within the range, and if both determination results are YES, the thin choke line is specified to be coupled to the thick choke line. The observation points for the thick choke line are the four edge points at both ends in the longitudinal direction and at both ends at the maximum width in the direction perpendicular to the central axis. The observation point of the thin choke line is the end closest to the thick choke line. The predetermined value of the distance between the observation points of the thick choke line and the thin choke line, and the predetermined value of the angle between the central axis of the front thick choke line and the thin choke line are set based on the knowledge of the tunnel specialist and stored in the RAM 3. Stored in advance.

(Significant choke line extraction)
Subsequent to step 207, the CPU 1 extracts a significant choke line by a connecting process for connecting the thin choke line extracted in step S207 to the thick choke line (S208).

Subsequent to step S207, the CPU 1 determines whether or not the choke line has been extracted without leakage (S209). This determination process is a determination as to whether or not all the choke line extraction processes performed while changing the smoothing parameter from a large value to a planned minimum value have been completed. Therefore, if choke line extraction is completed for a preprocessed image that has been smoothed by a Gaussian filter with the smoothing parameter set to the minimum value, the determination is YES and choke line detection ends.

If the determination result in step S209 is YES, the choke line extraction process ends, and if NO, the CPU 1 masks the choke line extracted in step S208 on the preprocessed image (S210).

Subsequently, the CPU 1 resets the smoothing parameter of the Gaussian filter stored in the RAM 3 to be smaller than the value set in step S202 in accordance with the program stored in the ROM 2 (S211). Next, the CPU 1 reads out the preprocessed image stored in the RAM 3 according to the program stored in the ROM 2 and applies a Gaussian filter in which the smoothing parameter is set smaller, and smoothes it (S203). Subsequently, the CPU 1 applies a Laplacian filter to the smoothed preprocessed image to perform second order differential processing, and extracts the edge points of the choke line area with sub-pixel accuracy (S204). Is newly extracted. Thereafter, the CPU 1 performs the processes of steps S205 to S208 in accordance with a program stored in the ROM 2. And if it determines with YES by the process of process S209, the detection of a choke line will be complete | finished. FIG. 9 shows an example of a concrete surface image including the extracted chalk line.

1 CPU
2 ROM
3 RAM
4 Input part 5 Display part 6 Output part 11-18 Crack 11a-11d Observation point 12a, 12b, 13a, 13b Observation point 14a, 14b, 15a, 15b Observation point 16a, 17a, 18a Observation point

Claims (9)

A preprocessing step of obtaining a preprocessed image by performing correction and noise removal on the original image which is a captured image of the concrete surface;
Applying a second-order differential edge filter after subjecting the preprocessed image to a smoothing process to extract linear deformation candidates with sub-pixel accuracy ;
A step of obtaining a feature amount of an edge point of the linear deformation candidate by subjecting the preprocessed image to a first derivative, and a first derivative value used for evaluation in obtaining the feature amount of the edge point as an upper limit threshold value From the linear deformation candidate including a hysteresis threshold processing step for determining whether the linear deformation candidate is a thick linear deformation, a thin linear deformation, or a linear noise. Removing linear noise and extracting a thick linear deformation and a thin linear deformation; and
The step of identifying the distance and direction between the thick linear deformation and the thin linear deformation, and the thin linear deformation that cannot be connected to the thick linear deformation is extracted as a non-significant linear deformation and is thick. The thin linear deformation that can be connected to the linear deformation is extracted from the process of extracting the significant linear deformation as information for judging the soundness of the concrete structure to be extracted as a significant linear deformation as well as the thick linear deformation. For detecting linear deformation of concrete surface The pre-processing step includes a step of performing shading for correcting illuminance unevenness at the time of photographing on an original image, a step of removing a periodic line component of a mold by frequency analysis, and a step of removing an avatar by frequency analysis. The method for detecting linear deformation of a concrete surface according to claim 1. The method for detecting linear deformation of a concrete surface according to claim 1, wherein the smoothing process is performed using a Gaussian smoothing filter. The method for detecting linear deformation on a concrete surface according to claim 1, wherein the second-order differential edge filter is a Laplacian filter. The step of obtaining the feature amount of the edge point of the linear deformation candidate is to perform a first-order differentiation in a direction perpendicular to the central axis of the linear deformation candidate on the preprocessed image, and to obtain a locally maximum differential value The method for detecting linear deformation of a concrete surface according to claim 1 , wherein the pixel is a step of extracting a pixel having a line. In the step of identifying the distance and direction between the thick linear deformation and the thin linear deformation, the distance between the thick linear deformation observation point and the thin linear deformation observation point is a predetermined value. If the direction of the central axis of the thick linear deformation and the thin linear deformation is within a predetermined value, the thick linear deformation and the thin linear deformation are combined deformations A distance between the observation point of the thick linear deformation and the observation point of the thin linear deformation is a predetermined value or more, or the center of the thick linear deformation and the thin linear deformation If the direction of the axis is equal to or greater than a predetermined value, it is determined that the thin linear deformation is not combined with the thick linear deformation, and further, the observation point of the thick linear deformation and the observation point of the thin linear deformation are And the thickness of the thin linear deformation is equal to or greater than a threshold value even if the direction of the central axis of the thick linear deformation and the thin linear deformation is equal to or greater than a predetermined value. if there is Linear Deformation detection method of concrete surface according to claim 1 wherein the thick linear Deformation and thin linear Henjo is characterized in that it is determined that Deformation coupled. The observation point is the an edge point of the ends of the thick linear Deformation widest edge points at both ends of the central axis and perpendicular to said central axis for, and a central axis for the thin linear Deformation The method for detecting linear deformation of a concrete surface according to claim 1 , wherein the edge points are at both ends. The linear deformation detection method for a concrete surface according to any one of claims 1 to 7 , wherein the linear deformation is a crack on the concrete surface. Linear Deformation detection method of the concrete surface as claimed in any one of claims 1 to 7, characterized in that said linear Henjo is choke line drawn on the concrete surface.