Crack detection method, crack detection apparatus, and program therefor
The method enhances crack detection in concrete surfaces by using wavelet transform and segmentation techniques to classify crack directions accurately, addressing inefficiencies in existing methods and improving infrastructure assessment efficiency.
Patent Information
- Application Number
- JP2023077406
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-05-09
- Publication Date
- 2024-11-21
AI Technical Summary
Existing crack detection methods for concrete surfaces struggle to accurately classify the direction of each crack, especially in complex scenarios with varying directions and conditions, leading to inefficiencies in assessing the condition of infrastructure facilities.
A method involving wavelet transform processing, binarization, and segmentation to create a crack image, followed by dividing cracks into split segments for accurate direction determination, using machine learning to enhance crack detection precision and reduce analysis time.
Enables high-accuracy classification of crack directions, reducing analysis time and improving the efficiency of infrastructure assessments by accurately identifying and categorizing cracks in concrete structures.
Smart Images

Figure 2024162145000001_ABST
Abstract
Description
[Technical field]
[0001] The present invention relates to a crack detection method, a crack detection device, and a program. [Background technology]
[0002] Currently, various infrastructure facilities such as roads, bridges, and tunnels are aging, and repair work is being carried out in various places, and repair plans are being promoted. Most infrastructure facilities are reinforced concrete structures (RC (Reinforced Concrete) structures) and steel structures, but various damaged areas exist on the surface of RC structures that have been in operation for more than 40 years. Specific examples of such damaged areas on the concrete surface include cracks, efflorescence, free lime, and rust. When carrying out repair work or planning repairs to RC structures, etc., the surface of the RC structure is first inspected by technical personnel of the infrastructure facility or by technical personnel of the outsourced survey company or construction company. In this inspection, the width and length of cracks, the shape and area of free lime, etc. are quantitatively evaluated, and based on this quantitative evaluation, a decision is made as to whether repairs or maintenance should be performed on the structure.
[0003] Meanwhile, various methods have been proposed for quantitatively detecting cracks on the concrete surface of RC structures (see Patent Documents 1 to 7). All of the crack detection methods proposed in Patent Documents 1 to 7 have a common process of performing wavelet transform processing and binarization processing, and then perform noise removal processing as necessary to detect cracks. These crack detection methods make it possible to detect cracks with high accuracy even when it is difficult to detect cracks due to dirt on the concrete surface or lighting conditions.
[0004] On the other hand, Patent Document 8 proposes a method for identifying the direction of cracks on a concrete surface, which makes it possible to quantitatively classify the numerous cracks present on the surface of the concrete being inspected as either unidirectional or bidirectional cracks as a whole. This method for identifying the direction of a crack on a concrete surface includes an input image creation process in which a photographed image of the concrete surface containing cracks is input into a computer to create an input image; a wavelet image creation process in which a wavelet transform process is performed on the input image to create a wavelet image in which each pixel has a wavelet coefficient; a binary image creation process in which each pixel of the wavelet image is binarized using a wavelet coefficient table that serves as a threshold for the wavelet coefficients to create a binary image; and a crack direction classification process in which each crack in the binary image is classified into a plurality of directional categories. [Prior art documents] [Patent documents]
[0005] [Patent Document 1] JP 2006-162583 A [Patent Document 2] JP 2008-267943 A [Patent Document 3] JP 2008-185510 A [Patent Document 4] JP 2010-121992 A [Patent Document 5] JP 2012-002531 A [Patent Document 6] JP 2013-117409 A [Patent Document 7] JP 2013-002839 A [Patent Document 8] JP 2019-102031 A Summary of the Invention [Problem to be solved by the invention]
[0006] According to the method for identifying the direction of cracks on a concrete surface described in Patent Document 8, it is possible to quantitatively classify a large number of cracks present on the concrete surface to be inspected as either unidirectional cracks or bidirectional cracks as a whole. In other words, this crack direction identification method identifies whether a large number of cracks present on the concrete surface to be inspected are unidirectional cracks or bidirectional cracks as a whole, but does not disclose a means for identifying the direction of each crack with high accuracy. In understanding the condition of concrete structures, it is desirable to be able to classify the direction of each crack with high accuracy.
[0007] The present invention aims to provide a crack detection method, a crack detection device, and a program that are capable of classifying the direction of each crack with high accuracy. [Means for solving the problem]
[0008] In order to achieve the above object, one aspect of the crack detection method according to the present invention includes: an input image creation step of inputting a photographed image of a concrete surface containing cracks into a computer to create an input image; a wavelet image creation process for performing a wavelet transform process on the input image, or for extracting assumed cracks from the input image based on learning information obtained by having a computer learn the characteristics of cracks on a concrete surface, drawing along the assumed cracks to create assumed crack drawing lines, and for performing a wavelet transform process on a drawing-containing image including the assumed crack drawing lines, or for performing a wavelet transform process on a path image including paths added along assumed crack lines that are assumed to be cracks to the input image, to create a wavelet image in which each pixel has a wavelet coefficient; a binary image creation step of binarizing each pixel of the wavelet image by using a wavelet coefficient table that is a threshold value of the wavelet coefficient to create a binary image; A crack image creation process for labeling each crack with a block-like figure having geometric connectivity in the binarized image and creating a crack image including a plurality of labeled cracks; A segmentation image creation process for creating a segmentation image by connecting end points of cracks that are determined to be essentially continuous cracks among the multiple separate cracks in the crack image; A crack direction classification step of classifying cracks of each label in the segmentation image into a plurality of direction categories, The crack direction classification step includes: A split crack creation process for creating a plurality of split cracks by dividing the total length of each crack of each label in the segmentation image by a predetermined number; The method is characterized by having a crack direction determination process in which the direction of each dividing crack that makes up each label crack is determined using the coordinates of the starting point and ending point of the dividing crack, and the direction of the crack is determined based on the directions of all the dividing cracks.
[0009] According to this embodiment, a wavelet image is created from an input image, a binary image is created from the wavelet image, a crack image is created from the binary image, and a segmentation image is created from the crack image.After that, for each crack with each label in the segmentation image, the total length of the crack is divided by a predetermined number to create multiple split cracks, and the direction of each split crack that makes up the crack with each label is determined using the coordinates of its start and end points.By determining the direction of the crack based on the directions of all the split cracks, it becomes possible to classify the direction of each crack with high accuracy. By dividing the total length of each crack for each label in the segmentation image by a predetermined number, for example 2 to 10, to create multiple divided cracks and identifying the direction of each divided crack, even if a single crack has a linear shape that meanders in various directions, the various directions can be identified as the directions of multiple divided cracks, and the direction of a single crack can be identified based on the directions of the identified multiple divided cracks. A method for identifying the direction of a single crack based on the directions of multiple dividing cracks is to set a threshold value for the main direction, and if any of the identified directions of multiple dividing cracks exceeds the threshold value, this direction is identified as the direction of a single crack. On the other hand, if there is no direction exceeding the threshold value, it may be determined that the crack is indistinguishable and does not have a main direction.
[0010] In the segmentation image creation process for creating a segmentation image, the distance between the end points of the cracks is specified, and the end point distance is compared with a threshold value for the end point distance to determine whether or not the two cracks whose end point distance has been specified are cracks of the same label. In addition, by image analysis, cracks whose end points are separated from each other can be accurately specified and output as separate cracks on a pixel-by-pixel basis. In addition, in the segmentation image, cracks that meet the conditions can be specified and output as cracks of the same label, not on a pixel-by-pixel basis, as in the inspection results by an inspector. When outputting, it is also possible to selectively output either one or both of the pixel-by-pixel output method and the segmentation image output method.
[0011] In addition, in the wavelet image creation process, the assumed cracks that are assumed to be cracks are extracted from the input image, an assumed crack drawing line is created by drawing along the assumed crack, and a wavelet transform process is performed on the drawing-containing image including the assumed crack drawing line, thereby narrowing the crack image analysis range (or analysis target) as much as possible. This method makes it possible to shorten the analysis time. In addition, when drawing along the assumed cracks to create the assumed crack drawing line, instead of drawing by an analyst, a computer that has learned the characteristics of cracks on the concrete surface draws based on this learning information, thereby eliminating the need for an analyst to draw, and making it possible to create the assumed crack drawing line in a significantly shorter time than drawing by an analyst. Furthermore, by performing a wavelet transform process on the drawing-containing image created in this way to create a binary image, cracks can be detected with high accuracy.
[0012] In addition, in the wavelet image creation process, when a path image is created for an input image, a path is created along the assumed crack line, and a surface is created to cover the crack on the path, thereby allowing the crack position to be specified in the input image. Specifically, a path consisting of, for example, a line or surface of 1 pixel is created using commercially available image editing software. The path image is, for example, a surface consisting of a 3-pixel width or 5-pixel width obtained by adding 1 pixel width or 2 pixel width to both sides of a path consisting of a line of 1 pixel width. In this way, by specifying the crack position in the input image and then performing wavelet transform processing to create a wavelet image, the crack image analysis range can be reduced as much as possible, and therefore the analysis area can be significantly reduced, which makes it possible to significantly reduce the analysis time, and furthermore, continuous analysis is possible, and the data reading and writing times can also be reduced.
[0013] In another aspect of the crack detection method according to the present invention, The dividing crack creating process is characterized in that an operation of dividing the crack by the predetermined number is sequentially performed until the length of the dividing crack becomes equal to or less than a predetermined target length.
[0014] According to this aspect, by sequentially performing an operation of dividing the crack by a predetermined number until the length of the dividing crack becomes equal to or less than the predetermined target length, it is possible to create dividing cracks of a plurality of target lengths and determine the direction of each dividing crack. Note that dividing one crack by a predetermined number (predetermined number of pixels) to set the target length of the dividing crack is also included in dividing by a predetermined number until the length of the dividing crack becomes equal to or less than the predetermined target length.
[0015] In another aspect of the crack detection method according to the present invention, The crack direction determination process is characterized in that, if there is a direction that accounts for more than 50% of all the directions of the dividing cracks, that direction is determined to be the direction of the crack.
[0016] According to this embodiment, the direction of a crack can be reasonably identified by determining the direction that accounts for 50% or more of all the directions of the dividing cracks as the crack direction. Here, if there is no direction that accounts for 50% or more, it can be determined that the crack is indistinguishable and does not have a main direction.
[0017] Another aspect of the crack detection method according to the present invention is to The method further includes a whole crack state classification step of classifying all the cracks of the labels as either one-way cracks or two-way cracks; In the overall crack state classification process, Based on the directions of all the split cracks included in the cracks on all the labels, if the proportion of cracks in the main crack direction exceeds a predetermined proportion, the crack is classified as a unidirectional crack, and if the proportion of cracks in the main crack direction is equal to or less than a predetermined proportion, the crack is classified as a bidirectional crack.
[0018] According to this aspect, the overall crack condition on the concrete surface can be quantitatively classified as either one-way cracks or two-way cracks overall. Here, the "predetermined ratio" regarding the ratio of cracks in the main crack direction can be set, for example, to 55%, 60%, etc., based on past literature, empirical rules, etc. That is, for example, when 60% is set as the predetermined ratio, if there are cracks with a crack direction exceeding this 60% ratio (for example, if the crack group with the bridge axis direction as the crack direction exceeds 60% of the total, etc.), the crack group to be analyzed can be considered as unidirectional cracks. On the other hand, if the crack direction of all cracks is 60% or less, the crack group to be analyzed can be considered as bidirectional cracks. In addition, in this specification, "bidirectional cracks" does not necessarily refer only to a single or multiple cracks extending in only two directions, but includes a single or multiple cracks extending in three or more directions. For example, if the ratio of cracks in the main crack direction is less than the predetermined ratio and there are multiple cracks extending in various directions of three or more directions, these cracks are also included in "bidirectional cracks".
[0019] In addition, one aspect of the crack detection device according to the present invention is an input image creation unit that creates an input image based on a photographed image of a concrete surface containing cracks that is input to a computer; a wavelet image creation unit that performs a wavelet transform process on the input image, or extracts assumed cracks that are assumed to be cracks from the input image based on learning information obtained by having a computer learn the characteristics of cracks on a concrete surface, draws along the assumed cracks to create assumed crack drawing lines, and performs a wavelet transform process on a drawing-containing image that includes the assumed crack drawing lines, or performs a wavelet transform process on a path image that includes paths added along assumed crack lines that are assumed to be cracks on the input image, and creates a wavelet image in which each pixel has a wavelet coefficient; a binary image creation unit that uses a wavelet coefficient table that is a threshold value of wavelet coefficients to binarize each pixel of the wavelet image to create a binary image; a crack image creation unit that labels each crack with a block-like figure having geometric connectivity in the binarized image and creates a crack image including the labeled cracks; a segmentation image creation unit that creates a segmentation image by connecting end points of cracks that are determined to be essentially continuous cracks among the multiple separate cracks in the crack image; A crack direction classification unit that classifies cracks of each label in the segmentation image into a plurality of direction categories; The crack direction classification unit is a split crack creation unit that creates a plurality of split cracks by dividing a total length of each crack of each label in the segmentation image by a predetermined number; The system is characterized by having a crack direction determination unit that determines the direction of each dividing crack that makes up each label crack using the coordinates of the starting point and ending point of the dividing crack, and determines the direction of the crack based on the directions of all the dividing cracks.
[0020] According to this embodiment, a wavelet image is created from an input image, a binary image is created from the wavelet image, a crack image is created from the binary image, and a segmentation image is created from the crack image.After that, for each crack with each label in the segmentation image, the total length of the crack is divided by a predetermined number to create multiple split cracks, and the direction of each split crack that makes up the crack with each label is determined using the coordinates of its start and end points.By determining the direction of the crack based on the directions of all the split cracks, it becomes possible to classify the direction of each crack with high accuracy.
[0021] In another aspect of the crack detection device according to the present invention, The split crack creation unit is characterized in that it sequentially performs an operation to divide the split crack by the predetermined number until the length of the split crack becomes equal to or less than a predetermined target length.
[0022] According to this embodiment, by sequentially performing calculations to divide the crack by a predetermined number until the length of the dividing crack becomes equal to or less than a predetermined target length, dividing cracks of multiple target lengths can be created and the direction of each dividing crack can be determined.
[0023] In another aspect of the crack detection device according to the present invention, The crack direction determination unit is characterized in that if there is a direction that accounts for more than 50% of the directions of all the dividing cracks, the crack direction determination unit determines that direction as the direction of the crack.
[0024] According to this embodiment, the direction of a single crack can be rationally identified by determining the direction that accounts for more than 50% of all the directions of the dividing cracks as the crack direction.
[0025] Moreover, one aspect of the program according to the present invention is A program for causing a computer to execute the following processes for detecting cracks on a concrete surface, an input image creation step of inputting a photographed image of a concrete surface containing cracks into a computer to create an input image; a wavelet image creation process for performing a wavelet transform process on the input image, or for extracting assumed cracks from the input image based on learning information obtained by having a computer learn the characteristics of cracks on a concrete surface, drawing along the assumed cracks to create assumed crack drawing lines, and for performing a wavelet transform process on a drawing-containing image including the assumed crack drawing lines, or for performing a wavelet transform process on a path image including paths added along assumed crack lines that are assumed to be cracks to the input image, to create a wavelet image in which each pixel has a wavelet coefficient; a binary image creation step of binarizing each pixel of the wavelet image by using a wavelet coefficient table that is a threshold value of the wavelet coefficient to create a binary image; A crack image creation process for labeling each crack with a block-like figure having geometric connectivity in the binarized image and creating a crack image including a plurality of labeled cracks; A segmentation image creation process for creating a segmentation image by connecting end points of cracks that are determined to be essentially continuous cracks among the multiple separate cracks in the crack image; A crack direction classification step of classifying cracks of each label in the segmentation image into a plurality of direction categories, The crack direction classification step includes: A split crack creation process for creating a plurality of split cracks by dividing the total length of each crack of each label in the segmentation image by a predetermined number; The method is characterized by having a crack direction determination process in which the direction of each dividing crack that makes up each label crack is determined using the coordinates of the starting point and ending point of the dividing crack, and the direction of the crack is determined based on the directions of all the dividing cracks.
[0026] According to this embodiment, by having a computer execute each step of the program, a wavelet image is created from an input image, a binary image is created from the wavelet image, a crack image is created from the binary image, and a segmentation image is created from the crack image.After that, for each crack with each label in the segmentation image, the total length of the crack is divided by a predetermined number to create multiple split cracks, the direction of each split crack that makes up the crack with each label is determined using the coordinates of its start point and end point, and the direction of the crack is determined based on the directions of all the split cracks, making it possible to classify the direction of each crack with high accuracy. Effect of the Invention
[0027] As can be understood from the above explanation, the crack detection method, crack detection device, and program of the present invention make it possible to classify the direction of each crack with high accuracy. [Brief description of the drawings]
[0028] [Figure 1] A diagram showing an example of the overall configuration of a crack detection system including a crack detection device according to an embodiment. [Diagram 2] FIG. 2 is a diagram illustrating an example of a hardware configuration of a crack detection device. [Diagram 3] FIG. 2 is a diagram illustrating an example of a functional configuration of a crack detection device. [Figure 4] 1 is a flowchart showing an example of a crack detection method according to an embodiment. [Figure 5A] 1A and 1B are diagrams illustrating an example of a relationship between an input image and a local region in wavelet transform processing. [Figure 5B] 1A and 1B are diagrams illustrating an example of a relationship between a local region and a pixel of interest in wavelet transform processing; [Figure 6] FIG. 13 is a diagram showing an example of a pseudo image. [Figure 7] FIG. 7 is a diagram showing an example of a bird's-eye view of wavelet coefficients of the pseudo image of FIG. 6. [Figure 8] FIG. 13 is a diagram illustrating an example of a wavelet coefficient table. [Figure 9] (a) is a schematic diagram showing an example where cracks that are separated from each other are determined to be a continuous crack in nature, and (b) and (c) are both schematic diagrams showing examples where cracks that are separated from each other are determined not to be a continuous crack in nature. [Figure 10] 9(a), (b), and (c) are schematic diagrams illustrating an example of a determination that is further performed in the determination unit when the distance between the end points is equal to or smaller than a threshold value, using FIG. 9(a). [Figure 11] 9(a) and (b) are schematic diagrams illustrating an example of a determination that is further performed in the determination unit when the distance between the end points is equal to or smaller than a threshold value, using FIGS. 9(b) and 9(c), respectively. [Figure 12] 1 is a flowchart showing an example of a crack direction classification process in a crack detection method according to an embodiment. [Figure 13A] An explanatory diagram illustrating an example of a split crack creation process in the crack direction classification process. [Figure 13B] Following Figure 13A, this is an explanatory diagram illustrating an example of the dividing crack creating process in the crack direction classification process. [Figure 13C] An explanatory diagram illustrating an example of a crack direction determination process in the crack direction classification process. [Figure 13D] Following Figure 13C, this is an explanatory diagram illustrating an example of the crack direction determination process in the crack direction classification process. [Figure 14] 13 is a diagram showing an example of a list of output sheets created in the crack direction classification process, together with an image of a concrete surface containing the crack to be detected. [Figure 15] FIG. 15 is a diagram showing an enlarged image of part XV in the image of the concrete surface in FIG. 14 and the output result of the cracks in the label contained in this image. [Figure 16] A figure showing an example of a list of output sheets created in the overall crack condition classification process. [Figure 17] This figure shows the evaluation categories of the degree of damage to deck cracks, as described in the Bridge Periodic Inspection Guidelines, which are applied as an example of evaluation criteria in the overall crack condition classification process. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0029] Hereinafter, a crack detection method and a crack detection device according to an embodiment will be described with reference to the accompanying drawings. In this specification and the drawings, substantially identical components may be designated by the same reference numerals to avoid redundant description.
[0030] [Embodiment] <Crack detection system> First, a crack detection system including a crack detection device according to an embodiment will be described. FIG. 1 is a diagram showing an example of the overall configuration of a crack detection system including a crack detection device according to an embodiment. As shown in FIG. 1, the crack detection system 1000 has an imaging device 100 and a crack detection device 300. The crack detection device 300 is, for example, a server device, and the imaging device 100 and the crack detection device 300 are connected via a network 200 such as a public network such as the Internet, a wireless network such as a mobile phone network, a dedicated network such as a VPN (Virtual Private Network), or a LAN (Local Area Network).
[0031] The imaging device 100 has an imaging section such as a CCD camera, a digital camera, or a digital video camera, and a communication section that transmits image data captured by the imaging section. The imaging device 100 may have only an imaging section, and may be connected to a mobile terminal or the like, and the image data may be transmitted from the mobile terminal or the like. The image data captured by the imaging device 100 is transmitted to the crack detection device 300 via the network 200. This image data includes image information of the concrete surface. For example, the concrete surface is imaged by the imaging device 100 to determine whether or not infrastructure facilities such as road bridges and tunnels made of reinforced concrete that have been built several decades ago need repair work.
[0032] A data collection program and a data analysis program are installed in the crack detection device 300, and the crack detection device 300 functions as a data collection unit 301 and a data processing unit 302 by executing these programs.
[0033] The data collection unit 301 receives image data captured by the imaging device 100 and transmitted from the imaging device 100 or a mobile terminal or the like connected to the imaging device 100, and stores the image data in the data storage unit 303. The data processing unit 302 also executes a process of sequentially creating various images based on the image data stored in the data storage unit 303, and outputting the various created images.
[0034] <Crack detection device> Next, an example of a crack detection device according to an embodiment will be described with reference to FIGS.
[0035] (Hardware configuration of crack detection device) First, an example of the hardware configuration of the crack detection device will be described with reference to Fig. 2. As shown in Fig. 2, the crack detection device 300 has a CPU (Central Processing Unit) 401, a ROM (Read Only Memory) 402, a RAM (Random Access Memory) 403, an auxiliary storage unit 404, a display unit 405, and a communication unit 406, and each unit is connected to each other via a bus 407.
[0036] The CPU 401 executes various programs installed in the auxiliary storage unit 404. The ROM 402 is a non-volatile memory and functions as a main storage unit that stores various programs and data necessary for the CPU 401 to execute the various programs stored in the auxiliary storage unit 404. The RAM 403 is a volatile memory and functions as a main storage unit. The RAM 403 functions as a working area when the CPU 401 executes the various programs stored in the auxiliary storage unit 404. The auxiliary storage unit 404 stores various programs installed in the crack detection device 300 and data used when executing the various programs.
[0037] The display unit 405 displays various screens. For example, it displays image data transmitted from the imaging device 100 as a captured image, as well as an input image, a wavelet image, a binarized image, a crack image, a segmentation image, an output sheet showing the crack direction of each label, an output sheet showing the overall crack state, and the like.
[0038] The communication unit 406 is connected to the imaging device 100 or a mobile terminal or the like connected to the imaging device 100, and receives image data from the imaging device 100 or the like, and transmits the crack direction, crack width, etc. of each label identified and created by the crack detection device 300 to the mobile terminal or the like connected to the imaging device 100.
[0039] (Functional configuration of the crack detection device) Next, an example of the functional configuration of the crack detection device will be described with reference to Fig. 3. As shown in Fig. 3, image data transmitted from the imaging device 100 is received by the data collection unit 301, and temporarily stored in the data storage unit 303 from the data collection unit 301. When the data processing unit 302 executes an analysis, the image data stored in the data storage unit 303 is input to the input unit 502 in the data processing unit 302.
[0040] The data processing unit 302 includes an input unit 502, an input image creation unit 504, a wavelet image creation unit 506, a wavelet coefficient table creation unit 508, a binary image creation unit 510, a crack image creation unit 512, a segmentation image creation unit 514, a crack direction classification unit 520, an overall crack state classification unit 525, a machine learning unit 530, and a drawing unit 532. The segmentation image creation unit 514 includes an end point distance specification unit 515, a judgment unit 516, and a drawing unit 517. The crack direction classification unit 520 includes a split crack creation unit 521 and a crack direction judgment unit 522.
[0041] The input unit 502 takes in a photographed image based on image data stored in the data storage unit 303. Image data of various concrete surfaces is stored in the data storage unit 303, and a photographed image based on the image data to be analyzed is selected according to the designation of the analyst, and is taken in and input to the input unit 502.
[0042] The input image creation unit 504 creates an input image by correcting the luminance of the photographed image input to the input unit 502 as necessary. For example, the luminance has 256 gradations from luminance value 0 to 255, and a predetermined luminance correction value is input in advance to the input image creation unit 504 by an analyst. For example, when 150 is input as the luminance correction value to the input image creation unit 504, the input image creation unit 504 identifies the luminance of the photographed image taken from the input unit 502, performs luminance correction processing so that the luminance of the photographed image becomes 150, which is the luminance correction value, and creates an input image. Note that whether or not such luminance correction is performed is optional.
[0043] Here, based on the learning information in the machine learning unit 530, an assumed crack that is assumed to be a crack in the input image may be extracted, and an assumed crack drawing line may be created by drawing along the assumed crack, and a drawing-containing image including this assumed crack drawing line may be created in the drawing unit 532.
[0044] The drawing unit 532 extracts assumed cracks from the input image based on learning information obtained by having a computer learn the characteristics of cracks on the surface of concrete, draws along these assumed cracks to create assumed crack drawing lines, and creates a drawing-containing image that includes these assumed crack drawing lines.
[0045] Here, the learning information in the machine learning unit 530 will be outlined. First, a feature database related to various concrete surface images is stored in the data storage unit 303. Crack lines that the analyst judges to be cracks are drawn on the concrete surface images. Therefore, the stored concrete surface images contain the photographed image and the drawn crack lines. The machine learning unit 530 creates a function that machine-learns the crack features in each concrete surface image and a model that shows the crack features, based on the concrete surface images input to the data storage unit 303 and the crack lines drawn in the concrete surface images.
[0046] The function for learning the crack features by machine learning and the model showing the crack features can be created using existing technology, such as supervised learning using a neural network, but the creation method is not particularly limited. By repeating machine learning in the machine learning unit 530, various crack patterns in various concrete surface images are learned. The learning data on the crack patterns learned for the concrete surface images is stored again in the data storage unit 303.
[0047] A computer that has learned the characteristics of cracks on a concrete surface can then perform wavelet transformation on the drawing-containing image it has drawn based on this learning information to create a binary image and a crack image, making it possible to detect cracks with high accuracy.
[0048] Alternatively, a path image including a path added along an assumed crack line that is assumed to be a crack may be created for the input image. A path is created along the assumed crack line, and a surface is created to cover the crack added to the path, thereby allowing the crack position to be specified in the input image. Specifically, a path consisting of, for example, a line or surface of one pixel is created using commercially available image editing software. Then, the path image is a surface that covers the crack added to the path, which is a surface consisting of three pixels or five pixels wide, which is obtained by adding one pixel or two pixels wide on both sides of a path consisting of a one-pixel wide line.
[0049] By creating a rough path (a path several pixels wider than the width of the expected crack line) along the expected crack line to specify the crack position, problems such as the expected crack line protruding from the path and not being determined to be a crack are eliminated, and because the path is rough, it can be created efficiently and does not take a long time to create the path (identify the crack position). Furthermore, by creating a rough path on the expected crack line and performing a thinning process after wavelet transformation, it becomes possible to identify the crack with even greater accuracy.
[0050] The wavelet image creation unit 506 creates a wavelet image by performing a wavelet transform process on the input image created by the input image creation unit 504. A wavelet means a small wave, and a basic unit of a wave with localization can be expressed by an equation using a wavelet function. By expanding or contracting this wavelet function, it becomes possible to simultaneously analyze time information, space information, and frequency information. When this wavelet coefficient is applied to a concrete surface having cracks, the coefficient is characterized in that it depends on the density of the concrete surface, the density of the cracks, and the width (or extent) of the cracks. For example, the value of the wavelet coefficient tends to increase as the width of the cracks increases, and the value of the wavelet coefficient tends to increase as the density of the cracks increases (as they become darker). An algorithm for creating a binary image using the wavelet coefficient calculated by the wavelet transform process will be described in detail below.
[0051] The wavelet coefficient table creation unit 508 creates wavelet coefficients that serve as thresholds when creating a binarized image from a wavelet image. Since the wavelet coefficients vary depending on the crack width, crack density, and concrete surface density as described above, the wavelet coefficients related to the crack density and concrete surface density are calculated for each gradation using pseudo-created data, and a wavelet coefficient table is created. For example, the wavelet coefficients (thresholds) corresponding to the two contrasting densities (one density can be assumed to be the concrete surface density and the other density to be the crack density) are uniquely determined by referring to the wavelet coefficient table.
[0052] The binary image creation unit 510 executes a binary process that compares each pixel of the wavelet image created by the wavelet image creation unit 506 with a wavelet coefficient that serves as a threshold value created by the wavelet coefficient table creation unit 508. For example, if the wavelet coefficient value of each pixel that constitutes the wavelet image is greater than the threshold value of the wavelet coefficient table, the pixel is determined to be a crack and assigned a value of 1, and if the wavelet coefficient value is less than the threshold value, the pixel is determined to not be a crack and assigned a value of 0. Therefore, the binary image is an image in which each pixel is expressed as either 0 or 1.
[0053] The crack image creation unit 512 performs contour tracing processing on the binarized image and further performs thinning processing to create a crack image.
[0054] The contour tracing process starts from an arbitrary pixel (a pixel determined to be a crack), and if an adjacent pixel is a crack, it connects to the starting pixel, and if another adjacent pixel is a crack, it connects both of them, and ends when it finally closes to the starting pixel (for example, when the first pixel, the second pixel, ..., the n-1th pixel, the nth pixel, and the first pixel are connected in that order), or when there is no further crack to connect to. This contour tracing process creates appropriate crack lines, such as a crack line that closes in a loop, or a crack line that extends linearly with multiple bends. At this time, by setting a predetermined value for the minimum number of pixels that can be connected, all pixels below this set number are deemed not to be cracks and can be removed from the crack display on the screen.
[0055] This contour tracing process labels each individual crack with a blocky shape that has geometric connectivity, and large noise is removed during this labeling process.
[0056] Following the contour tracing process, a thinning process is performed, which reduces the crack to a single pixel wide, for example, along its centerline, allowing the removal of small noise.
[0057] In this way, by performing contour tracing processing on the binarized image and then performing thinning processing, a crack image having a plurality of labeled cracks is created.
[0058] The crack image thus created is a highly accurate crack image on a pixel-by-pixel basis. However, even if the end points of the cracks are physically separated, if the cracks are determined to be essentially continuous, it is desirable to output them as cracks with the same label, since this allows for the output of crack detection similar to that when an inspector actually inspects and determines the cracks. Therefore, the segmentation image creation unit 514 creates a segmentation image by connecting the end points of the cracks that are determined to be essentially continuous among the multiple cracks that are separated in the crack image.
[0059] More specifically, the segmentation image creation unit 514 includes an end point distance specification unit 515 , a determination unit 516 , and a drawing unit 517 .
[0060] The end point distance specification unit 515 specifies the distance between mutually adjacent end points of multiple separated cracks in the crack image. For example, even if an end point of one crack intersects with the middle of another adjacent crack, the end points of both are separated, so the end point distance is specified even in such a separated form.
[0061] The judgment unit 516 compares the distance between the endpoints with a threshold value for the distance between the endpoints, and if the distance between the endpoints is less than or equal to the threshold, it judges that the two cracks whose distance between the endpoints is identified are essentially continuous cracks and therefore both are cracks with the same label.
[0062] Here, the method for setting the threshold for the distance between the endpoints is to specify a distance (e.g., 3 cm) that is highly likely (e.g., 70% or higher) to be the same-label crack even if the cracks are spaced apart, based on accumulated past crack detection results on the concrete surface, and set the threshold. In addition, the width of the endpoints of the cracks that are spaced apart from each other may also be taken into consideration, and the threshold may be set based on both the width of the endpoints and the distance.
[0063] Furthermore, the threshold value can also be set by the machine learning unit 530. In the machine learning unit 530, the threshold value may be set based on a function or the like learned by machine learning using a large amount of past data on combinations of cracks (including the linear shapes of both cracks, the crack widths at the end points, etc.) when mutually separated cracks can be determined to be essentially continuous cracks, and the distance between the end points at that time.
[0064] The machine-learned function can be created using existing technology, such as supervised learning using a neural network, but the creation method is not particularly limited. By repeating machine learning in the machine learning unit 530, an optimal threshold value according to the spaced crack pattern is learned and set based on various crack patterns in various concrete surface images.
[0065] As described above, the determining unit 516 creates a segmentation image by comparing the distance between the end points of the mutually separated cracks with a threshold value.
[0066] Here, the segmentation image is created by automatically connecting cracks in the drawing unit 517. Alternatively, the analyst may create the segmentation image by connecting cracks using image editing software (handwriting software). In the latter case, as a result of the judgment by the judging unit 516, the end points of the cracks that should be connected to each other are displayed on the screen, so that the analyst can connect the displayed end points.
[0067] As described above, in addition to the judgment method in which the judgment unit 516 only compares the distance between the endpoints with a threshold value, when it is judged that the distance between the endpoints is equal to or less than the threshold value by comparing the distance between the endpoints with the threshold value, it is also possible to execute a judgment method in which the directionality (vector) near the endpoints of both cracks is verified, and if it is judged that both vectors have continuity, it is judged that the cracks are of the same label. That is, the judgment unit 516 can make a judgment by either of these methods, and the analyst can selectively select the judgment algorithm in the judgment unit 516. For example, a simple judgment can be made by a judgment based only on a comparison between the distance between the endpoints and a threshold value, and a detailed judgment can be made by a judgment that specifies continuity based on vectors near the endpoints of both cracks in addition to comparing the distance between the endpoints with the threshold value.
[0068] Even if the distance between the endpoints is equal to or less than the threshold, if it is not possible to determine that the cracks are clearly of the same label based on the vectors near the endpoints of both cracks, the inspector will generally not determine that the two cracks are of the same label when actually inspecting them. Therefore, this detailed determination makes it possible to create a segmentation image with higher accuracy. The method of determining continuity based on vectors near the endpoints of both cracks in the detailed determination will be described in detail below.
[0069] The analyst can selectively output either the created crack image or the segmentation image, or both, as desired. In addition, in the crack distribution map, which is an example of the output, the crack width (maximum crack width) of each crack can be added to the side of the crack, allowing the distribution of each crack on the concrete surface and the damage state of each crack to be confirmed on a single screen.
[0070] The crack direction classification unit 520 determines the direction of each of the multiple cracks contained in the segmentation image. When determining the crack direction, the crack is divided into multiple division cracks, and the direction of each division crack is determined. The crack direction classification unit 520 has a division crack creation unit 521 and a crack direction determination unit 522.
[0071] The split crack creation unit 521 creates multiple split cracks for each crack with each label in the segmentation image by dividing the total length of the crack by a predetermined number (a predetermined number of pixels). This predetermined number may be set according to the analyst's wishes.
[0072] For example, the accuracy of identifying the crack direction is maximized by setting the length of the split crack to the minimum length of 3 pixels, but on the other hand, since the number of split cracks will increase if a single crack is long, split cracks of various lengths may be created from the standpoints of both accuracy and analysis efficiency.
[0073] However, as explained below, in light of the algorithm that determines the direction of a splitting crack from its start and end points, if the length of the splitting crack is too long, there is a possibility that multiple vectors (directions) will be contained within a single splitting crack, which will reduce the accuracy of determining the crack direction, so it is desirable to limit the length of the splitting crack to a maximum of 10 pixels.
[0074] The crack direction determination unit 522 determines the direction of each of the multiple dividing cracks that make up the crack of each label using the coordinates of the starting point and ending point of the dividing crack, and determines the crack direction based on the directions of all the dividing cracks.
[0075] The direction of a dividing crack is determined as the direction of a straight line connecting the coordinates of its start point and end point. After determining the directions of all the dividing cracks that make up a single crack using this method, if there is a direction that accounts for more than 50% of all the dividing crack directions, this direction is determined to be the crack direction. On the other hand, if there is no direction that accounts for more than 50%, it is deemed indistinguishable and is determined to be a crack that does not have a main direction.
[0076] In this manner, the crack direction classification unit 520 determines the crack direction for each crack of each label for all cracks present in the segmentation image.
[0077] The overall crack state classification unit 525 classifies all the cracks of all the labels as either unidirectional or bidirectional cracks. Specifically, based on the directions of all the split cracks included in the cracks of all the labels, if the proportion of cracks in the main crack direction exceeds a predetermined proportion, it is classified as a unidirectional crack, and if the proportion of cracks in the main crack direction is equal to or less than a predetermined proportion, it is classified as a bidirectional crack.
[0078] This predetermined ratio may be set, for example, to 55%, 60%, etc., based on past literature, empirical rules, etc. Alternatively, classification may be performed with reference to the evaluation categories of the degree of damage to deck cracks described in the bridge periodic inspection guidelines, which are frequently used when classifying the overall crack condition. In this case, drawing data on one-way cracks and two-way cracks described in the evaluation categories is stored in the data storage unit 303, and the overall crack condition classification unit 525 can start, for example, the machine learning unit 530 to compare the drawing data with the entire segmentation image and determine whether the cracks are one-way or two-way.
[0079] <Crack detection method> Next, an example of a crack detection method according to an embodiment will be described with reference to Fig. 4 to Fig. 17. Here, Fig. 4 is a flowchart showing an example of a crack detection method according to an embodiment, and shows a process flow in the crack detection device 300.
[0080] In step S700, an input image is created in input image creation section 504 based on image data inputted into input section 502 (input image creation step).
[0081] For example, if the input image requires brightness correction processing, the brightness correction processing is performed to create an input image, and if brightness correction processing is not required, the captured image is used as is as the input image. Of the 256 brightness levels, a brightness that the analyst determines to be optimal is set within a range of about 120 to 160, including the median value of 128. If the brightness of the captured image does not match the set brightness, a brightness correction processing is performed on the captured image to create an input image with the set brightness.
[0082] In step S702, an input image is taken into the wavelet image creation unit 506, and a wavelet transform process is performed on the input image to create a wavelet image. At this time, the wavelet image may be created by extracting assumed cracks from the input image based on learning information obtained by having a computer learn the characteristics of cracks on the concrete surface, drawing along the assumed cracks to create assumed crack drawing lines, and performing wavelet transform process on the drawing-containing image including the assumed crack drawing lines. Alternatively, a path image including paths added along assumed crack lines that are assumed to be cracks may be created for the input image, and the wavelet transform process may be performed on the path image to create the wavelet image (wavelet image creation process).
[0083] Here, the wavelet transform process will be described. Fig. 5A is a diagram showing an example of the relationship between an input image and a local region in wavelet transform process. Fig. 5B is a diagram showing an example of the relationship between a local region and a pixel of interest in wavelet transform process. Wavelet transform is performed in local region 3, which is the center of wide region 2 in input image 1, and cracks are detected in the center of local region 3. Cracks in input image 1 are detected by translating wide region 2 up, down, left, and right all over input image 1.
[0084] Fig. 5B is an enlarged view of local region 3. In the illustrated embodiment, a crack is determined at the center of, for example, 3 x 3 nine pixels (eight neighboring pixels 31, 31, ... and a pixel of interest 32 located in the center). Note that the wavelet coefficient is calculated for local region 3 in Fig. 5A. The formula for calculating the wavelet coefficient by performing a wavelet transform using a wavelet function (mother wavelet function) is shown below.
[0085]
number
[0086]
number
[0087]
number
[0088] Here, f(x, y) is the input image (where x and y are arbitrary coordinates in the two-dimensional input image), Ψ is the mother wavelet function (Gabor function), and (x0, y0) is the translation amount of Ψ. k is the expansion or contraction of Ψ (where a kis the reciprocal of frequency, and the frequency width for calculations over several frequency ranges is expressed as an integer k), f0 is the center frequency, and σ is the standard deviation of the Gaussian function. Furthermore, θ is the rotation angle that indicates the direction of wave travel, and (x', y') are the coordinates obtained by rotating (x, y) by the angle θ.
[0089] The equation for calculating the cumulative value C(x0, y0) of the wavelet coefficient Ψ for multiple θ and k calculated using Equation 1 is given by Equation 4 below.
[0090]
number
[0091] The above parameters can be set arbitrarily. For example, σ is set to 0.5 to 2, a k can be set between 0 and 5, f0 can be set to 0.1, and the rotation angle can be set between 0 and 180 degrees. The translation amount (x0, y0) in Equation 4 corresponds to the position of the pixel of interest, and the continuous amount of wavelet coefficients (C(x0, y0)) can be calculated by sequentially moving the position of the pixel of interest.
[0092] After calculating the wavelet coefficients for all pixels that make up the local region 3 based on the above formula, wavelet coefficients are similarly calculated for all pixels in the wide region 2 that is created by shifting the pixel of interest one position to the left or right or up or down.
[0093] When creating a wavelet image, a wavelet coefficient table is created by the wavelet coefficient table creating unit 508. In creating the wavelet coefficient table, wavelet coefficients are calculated for a pseudo image consisting of two contrasting densities that have no relation to the input image. For example, as shown in FIG. 6, wavelet coefficients are calculated for a pseudo image consisting of a background color a assumed to be a concrete surface (for example, the background colors R, G, and B are 255, 255, 255) and line segments b1 to b5 assumed to be cracks. Here, the line segments b1 to b5 have line widths that vary from 1 pixel to 5 pixels, and each line segment has three types of densities (for example, the line segment b1 has densities that vary from b11 (black), b12 (light black), and b13 (gray) in order of darkest to lightest).
[0094] A bird's-eye view of the wavelet coefficients calculated by performing a wavelet transform on this pseudo image is shown in Fig. 7. In Fig. 7, the X-axis represents the width of the line segment, the Y-axis represents the color density of the line segment, and the Z-axis represents the wavelet coefficients. At the same time, a wavelet coefficient table as shown in Fig. 8 is created by combining two contrasting densities in 256 gradations from 0 to 255.
[0095] The wavelet coefficient table may be created at any time up to the creation of a binary image, which will be described later. For example, the wavelet coefficient table may be created after the captured image is input, or in parallel with the creation of a wavelet image, or after the wavelet image is created and before the creation of a binary image.
[0096] In step S704, the binary image creation unit 510 creates a binary image (binary image creation step).
[0097] In the wavelet coefficient table, the wavelet coefficient corresponding to the average density of the neighboring pixels in the local region and the density of the pixel of interest is set as the threshold value for the wavelet coefficient. If the wavelet coefficient of the pixel of interest is larger than the threshold, the pixel of interest is determined to be a crack (e.g., white on the screen), and if it is smaller, the pixel of interest is determined not to be a crack (e.g., black on the screen). A binary image is created by performing a comparison operation between the wavelet coefficient of the pixel of interest and the threshold while changing the local region and the pixel of interest.
[0098] In step S706, the binarized image is input into the crack image creation unit 512, blocky figures with geometric connectivity are labeled into individual cracks, and a crack image with multiple labeled cracks is created (crack image creation process).
[0099] As described above, in the crack image creation process, the binarized image is subjected to contour tracing processing and then thinning processing, thereby creating a crack image while removing large and small noise.
[0100] In step S708, the crack image is input into the segmentation image creation unit 514, and a segmentation image is created by connecting the end points of multiple cracks that are separated from each other in the crack image and that are determined to be essentially continuous cracks (segmentation image creation process).
[0101] The segmentation image creation process includes an end point distance specifying process, a determining process, and a drawing process.
[0102] In the end point distance specifying step, the end point distance between the end points of mutually separated cracks is specified by the end point distance specifying unit 515. Since a large number of mutually separated cracks (combinations) exist in a crack image, the distance between the end points of all mutually separated cracks is specified.
[0103] In the determination step, the determination unit 516 compares the distance between the endpoints with a threshold value for the distance between the endpoints, and if the distance between the endpoints is equal to or less than the threshold value, the two cracks whose distance between the endpoints is identified are essentially continuous cracks and are determined to be cracks with the same label. As described above, methods for setting this threshold include a method of setting it based on past performance, a method of setting it by having the machine learning unit 530 learn past performance by machine learning, and the like.
[0104] In the drawing process, a segmentation image is created by connecting the end points of two cracks that have been determined to have the same label. As described above, the drawing method in the drawing process is to automatically connect the cracks in the drawing unit 517. Note that the segmentation image may also be created by an analyst using handwriting software to connect the cracks.
[0105] Here, the above-mentioned judgment method in the judgment process is a simple judgment that only compares the distance between the endpoints with a threshold value, but when it is judged that the distance between the endpoints is equal to or less than the threshold value by comparing the distance between the endpoints with the threshold value, it is also possible to perform a detailed judgment that verifies the directionality (vector) near the endpoints of both cracks and judges that the cracks have the same label if it is judged that both vectors have continuity. The analyst can selectively select the judgment algorithm in the judgment unit 516.
[0106] Here, the contents of the detailed judgment will be described with reference to Figs. 9 to 11. Fig. 9(a) is a schematic diagram showing an example in which cracks separated from each other are judged to be essentially continuous cracks, and Figs. 9(b) and 9(c) are both schematic diagrams showing an example in which cracks separated from each other are judged not to be essentially continuous cracks. Moreover, Figs. 10(a), (b), and (c) are all schematic diagrams showing an example of judgment further performed in the judgment unit when the distance between the end points is equal to or less than a threshold value using Fig. 9(a). Furthermore, Figs. 11(a) and 11(b) are schematic diagrams showing an example of judgment further performed in the judgment unit when the distance between the end points is equal to or less than a threshold value using Figs. 9(b) and 9(c), respectively.
[0107] In Figures 9(a) to 9(c), the distance t1 between the end points P1, P2 of the mutually separating cracks C1 and C2, the distance t2 between the end points P3, P4 of the cracks C3 and C4, and the distance t3 between the end points P5, P6 of the cracks C5 and C6 are all assumed to be below a set threshold value.
[0108] In Fig. 9(a), vectors V1 and V2 near the end points of cracks C1 and C2 (for example, within a range of about 5 pixels from the inside of the crack to end points P1 and P2) are parallel or approximately parallel to each other (offset by about 30 to 45 degrees from parallel), and it can be judged from the outside that the end points P1 and P2 extend in a direction in which they may intersect with each other. Therefore, as a method for judging whether or not the cracks are of the same label, a method is applied in which an endless outline range is set at the end point of one crack, and if the end point of the other crack falls within this set endless outline range, both cracks are judged to be cracks of the same label, and if they do not fall within the endless outline range, they are judged not to be cracks of the same label.
[0109] Examples of endless contour shapes that have been changed in various ways are shown in Figures 10(a) to 10(c). The example shown in Figure 10(a) is an example in which an equilateral triangle (an example of a regular polygon) with one side of a predetermined length r1 is used as the endless contour range, Figure 10(b) is an example in which a square (an example of a regular polygon) with one side of a predetermined length r1 is used as the endless contour range, and Figure 10(c) is an example in which a circle with a diameter of the predetermined length r1 is used as the endless contour range.
[0110] The specified length r1 that defines the endless contour range can be set based on the performance of past crack detection methods, from the distance between the endpoints of cracks that can be determined to be cracks of the same label even if they are separated from each other, and can also be set to a value equal to a threshold value, for example.
[0111] In Figure 10(a), the apex of an equilateral triangle is positioned at the endpoint P1 of one crack C1, and an equilateral triangle is set so that a virtual extension line extending vector V1 near the endpoint P1 of one crack C1 intersects with the midpoint of the base of the equilateral triangle, and the contour range A1 of this equilateral triangle is a range of 30 degrees left and right from the extension line of vector V1. In Figure 10(a), because the endpoint P2 of the other crack C2 falls within the contour range A1 of the equilateral triangle, cracks C1 and C2 can be determined to be cracks with the same label.
[0112] On the other hand, in Figure 10(b), one corner of the square is positioned at the endpoint P1 of one crack C1, and another corner of the square that is diagonally opposite the one corner is positioned on an imaginary extension line of vector V1 near the endpoint P1 of one crack C1, setting a square, and the contour range A2 of this equilateral triangle is a 45-degree range to the left and right of the extension line of vector V1. In Figure 10(b), because the endpoint P2 of the other crack C2 falls within the contour range A2 of the square, it can be determined that the cracks C1 and C2 are cracks with the same label.
[0113] Furthermore, in Fig. 10(c), one point on the circumference of the circle is positioned at the end point P1 of one crack C1, and another point on the circle that is at both ends of the diameter of the circle is positioned on an imaginary extension line extending vector V1 near the end point P1 of one crack C1, thereby setting the circle, and the circle of the outline range A3 can be set in the direction of the extension line of vector V1.In Fig. 10(c), because the end point P2 of the other crack C2 falls within the outline range A3 of the circle, cracks C1 and C2 can be determined to be cracks with the same label.
[0114] Furthermore, the endless contour range may be a polygon other than a square or an equilateral triangle, and may be a circle or an ellipse, etc.; however, the shape shown in the example is preferred because squares and equilateral triangles are the polygons that make it easy to clearly define the contour range, and because circles can be arranged in the vector direction as in the example shown, making it easy to clearly define the contour range by setting their radius and diameter.
[0115] In addition, if an endless contour range is set at the end point of one crack and the end point of the other crack does not fall within it, the opposite case is performed, an endless contour range is set at the end point of the other crack, and it is checked whether the end point of one crack falls within it.If it does, both cracks are determined to be cracks of the same label, and if it does not fall within it (in both cases the end points of the adjacent crack do not fall within the endless contour range), both cracks are determined to not be cracks of the same label.
[0116] On the other hand, FIG. 11(a) shows an example in which an equilateral triangular outline range is defined for the model in FIG. 9(b), and FIG. 11(b) shows an example in which a circular outline range is defined for the model in FIG. 9(c).
[0117] In Figures 11(a) and 11(b), the end points P4 and P6 of one crack, C4 and C6, do not fall within the equilateral triangle and circular contour range A1 and A3 set for the end points P3 and P5 of the other crack, C3 and C5, respectively, and in the opposite case, the end points P3 and P5 of one crack, C3 and C5, do not fall within the equilateral triangle and circular contour range A1 and A3 set for the end points P4 and P6 of the other crack, C4 and C6, respectively. Therefore, it is determined that cracks C3 and C4 and cracks C5 and C6 are not cracks with the same label.
[0118] In this way, by determining in the judgment process whether or not cracks have the same label through a detailed judgment, it is possible to create a segmentation image with even higher accuracy.
[0119] Returning to FIG. 4, in step S710, the crack direction of each crack contained in the segmentation image is classified (crack direction classification process). Here, FIG. 12 is a flowchart showing an example of the crack direction classification process in the crack detection method according to the embodiment. Also, both FIG. 13A and FIG. 13B are explanatory diagrams explaining an example of a split crack creation process in the crack direction classification process, and both FIG. 13C and FIG. 13D are explanatory diagrams explaining an example of a crack direction determination process in the crack direction classification process. Furthermore, FIG. 14 is a diagram showing an example of an output sheet list created in the crack direction classification process together with an image of the concrete surface containing the crack to be detected, and FIG. 15 is a diagram showing an image in which the XV part in the image of the concrete surface in FIG. 14 is enlarged, and an output result of the crack of the label included in this image.
[0120] In the flowchart in FIG. 12, the flow from step S800 to step S804 is included in the dividing crack creation process, and the flow from step S806 to step S814 is included in the crack direction determination process.
[0121] First, when determining the direction of cracks of each label in the segmentation image, the total length of the crack is divided by a predetermined number to set the approximate length of the divided crack for each crack of each label. In the illustrated example, this predetermined number is set to "11 (11 pixels)" and the total length of the crack is divided by 11 to set the approximate length of the divided crack (step S800).
[0122] Figure 13A shows an example of an Excel sheet in which a total of 37 pixels from segments No. 7 to No. 9 of the crack labeled "label 2" in the segmentation image are divided by 11 to calculate "3 (pixels)," and this "3" is set as the length (approximate length) of the divided crack. In the figure, the X and Y coordinates are shown next to each pixel number.
[0123] Returning to Fig. 12, when the total length of the crack is divided by 11, it is checked whether the target length satisfies the condition of 10 pixels or less (step S802), and if it exceeds 10 pixels, the divided crack is further divided by 2 to 10 (pixels) (step S804), and this operation is repeated until the target length satisfies 10 pixels or less. For example, if the total length is 220, it becomes 220 / 11=20, but since 20 exceeds 10, it further divides 20 by a predetermined value in the range of 2 to 10 (for example, "2"), and 20 / 2=10 is set as the target length, i.e., "10".
[0124] If the condition that the reference length is 10 pixels or less is satisfied, values less than the reference length at the ends of each segment are excluded (step S806). For example, as shown in Fig. 13B, No. 7 10 is a fraction when the No. 7 segment is divided into cracks of 3 pixels each, so the fraction is excluded. In Fig. 13B, the colored columns indicate the fractions that are excluded.
[0125] Next, the direction of each dividing crack is determined based on the end and start points of each dividing crack (step S808). Specifically, as shown in Fig. 13C, the direction of each dividing crack is determined based on the X and Y coordinates of the start and end points of each dividing crack. In Fig. 13C, the direction of the first dividing crack of the No. 7 segment is determined to be 90 degrees based on the coordinates of its start and end points.
[0126] In this manner, by determining the direction of each dividing crack contained in each segment, the direction of each dividing crack is determined as shown in FIG. 13D.
[0127] Returning to Fig. 12, for the cracks of each label, it is checked whether there is a direction that accounts for more than 50% of all the directions of the dividing cracks (step S810), and if there is a direction that accounts for more than 50% of the directions, that direction is determined to be the crack direction of that label, and the direction symbol is entered into the output sheet as the crack direction. Specifically, in Fig. 14, the left column shows the coordinates of each pixel related to the crack of label 1, and the right column shows a list including the total crack length, average crack width, and maximum crack width of each label, as well as the crack direction symbol. Note that the lower right column of Fig. 14 shows an overall photograph of the concrete surface that is the subject of this list.
[0128] If there is no direction that accounts for 50% or more, it is deemed undistinguishable (step S812), and "x" is entered as the crack direction symbol in FIG.
[0129] FIG. 15 shows an enlarged photograph of part XV in FIG. 14 and a list of results for labels 1-5 included in the enlarged photograph.
[0130] According to the crack direction classification process, for each crack with each label in the segmentation image, the total length of the crack is divided by a predetermined number to create multiple divided cracks, and the direction of each divided crack that makes up the crack with each label is determined using the coordinates of its start and end points.By determining the direction of the crack based on the directions of all the divided cracks, it is possible to classify the direction of each crack with high accuracy.
[0131] 4, in step S712, the overall crack state in the segmentation image is classified. More specifically, all cracks with labels in the segmentation image are classified as either one-way cracks or two-way cracks (overall crack state classification process).
[0132] In the overall crack condition classification process, based on the directions of all split cracks included in all labeled cracks, if the proportion of cracks in the main crack direction exceeds a predetermined proportion, it is classified as a unidirectional crack, and if the proportion of cracks in the main crack direction is equal to or less than a predetermined proportion, it is classified as a bidirectional crack.
[0133] Here, the "predetermined ratio" regarding the ratio of cracks in the main crack direction can be set, for example, to 55%, 60%, etc., based on past literature, empirical rules, etc. In other words, for example, when 60% is set as the predetermined ratio, if there are cracks in the crack direction that exceed this 60%, the overall crack state in the segmentation image can be determined as unidirectional cracks. On the other hand, if the crack directions of all cracks are 60% or less, the overall crack state in the segmentation image can be determined as bidirectional cracks.
[0134] Here, Fig. 16 classifies the crack directions of each label shown in Fig. 14 into the X direction, Y direction, and diagonal directions, and shows their proportions. In the illustrated example, the three directions account for almost equal proportions, and therefore, when the predetermined proportion is set to 55%, the maximum proportion: 36% for the diagonal direction is less than 55%, so it can be determined to be a two-way crack.
[0135] In addition, classification may also be made by referring to the assessment categories for the degree of damage to deck cracks described in the Bridge Periodic Inspection Guidelines, which are frequently used when classifying the overall crack condition.
[0136] Here, Figure 17 shows the evaluation categories for the degree of damage to deck cracks, as described in the Bridge Periodic Inspection Guidelines (National Highway Technology Division, Road Bureau, Ministry of Land, Infrastructure, Transport and Tourism, March 2019), which are applied as an example of evaluation criteria in the overall crack condition classification process.
[0137] By comparing the ratios of each direction shown in Figure 16 with Figure 17, it can be determined that the crack corresponds to the two-way crack in the center of the right column shown in Figure 17. Note that the drawing data relating to the one-way cracks and two-way cracks described in the evaluation category can be stored in the data storage unit 303, and the overall crack condition classification unit 525 can, for example, start the machine learning unit 530 to compare the drawing data with the judgment result in Figure 16 and determine the corresponding item in the evaluation category.
[0138] The above describes the embodiment of the present invention in detail using the drawings. However, the specific configuration is not limited to this embodiment, and even if there are design changes and the like within the scope of the present invention, they are included in the present invention. [Explanation of symbols]
[0139] 100: Imaging device 200: Network 300: Crack detection device 301: Data Collection Department 302: Data processing unit 303: Data storage unit 504: Input image creation unit 506: Wavelet image creation section 508: Wavelet coefficient table creation unit 510: Binary image creation unit 512: Crack Image Creation Section 514: Segmentation image creation unit 515: Endpoint distance specification part 516: Judgment section 517: Drawing department 520: Crack direction classification section 521: Split crack creation section 522: Crack direction determination unit 525: Overall crack condition classification section 1000: Crack detection system C1~C6: Cracks P1~P6: End points t1~t3: Distance between end points V1~V6: Vectors near the end points A1 to A3: Contour range (endless contour range) r1: Predetermined length
Claims
1. an input image creation step of inputting a photographed image of a concrete surface containing cracks into a computer to create an input image; a wavelet image creation process for performing a wavelet transform process on the input image, or for extracting assumed cracks from the input image based on learning information obtained by having a computer learn the characteristics of cracks on a concrete surface, drawing along the assumed cracks to create assumed crack drawing lines, and for performing a wavelet transform process on a drawing-containing image including the assumed crack drawing lines, or for performing a wavelet transform process on a path image including paths added along assumed crack lines that are assumed to be cracks to the input image, to create a wavelet image in which each pixel has a wavelet coefficient; a binary image creation step of binarizing each pixel of the wavelet image by using a wavelet coefficient table that is a threshold value of the wavelet coefficient to create a binary image; A crack image creation process for labeling each crack with a block-like figure having geometric connectivity in the binarized image and creating a crack image including a plurality of labeled cracks; A segmentation image creation process for creating a segmentation image by connecting end points of cracks that are determined to be essentially continuous cracks among the multiple separate cracks in the crack image; A crack direction classification step of classifying cracks of each label in the segmentation image into a plurality of direction categories, The crack direction classification step includes: a split crack creation process for creating a plurality of split cracks by dividing the total length of each crack of each label in the segmentation image by a predetermined number; A crack detection method characterized by including a crack direction determination process in which the direction of each dividing crack that constitutes each label crack is determined using the coordinates of the starting point and ending point of the dividing crack, and the direction of the crack is determined based on the directions of all the dividing cracks.
2. The crack detection method described in claim 1, characterized in that in the split crack creation process, a calculation is sequentially performed to divide the crack by the predetermined number until the length of the split crack becomes equal to or less than a predetermined target length.
3. A crack detection method as described in claim 1 or 2, characterized in that in the crack direction determination process, if there is a direction that accounts for more than 50% of the directions of all the dividing cracks, that direction is determined to be the direction of the crack.
4. The method further includes a whole crack state classification step of classifying all the cracks of the labels as either one-way cracks or two-way cracks; In the overall crack state classification process, A crack detection method as described in claim 1 or 2, characterized in that, based on the directions of all the split cracks included in the cracks of all the labels, if the proportion of cracks in the main crack direction exceeds a predetermined proportion, they are classified as unidirectional cracks, and if the proportion of cracks in the main crack direction is equal to or less than a predetermined proportion, they are classified as bidirectional cracks.
5. an input image creation unit that creates an input image based on a photographed image of a concrete surface containing cracks that is input to a computer; a wavelet image creation unit that performs a wavelet transform process on the input image, or extracts assumed cracks that are assumed to be cracks from the input image based on learning information obtained by having a computer learn the characteristics of cracks on a concrete surface, draws along the assumed cracks to create assumed crack drawing lines, and performs a wavelet transform process on a drawing-containing image that includes the assumed crack drawing lines, or performs a wavelet transform process on a path image that includes paths added along assumed crack lines that are assumed to be cracks on the input image, and creates a wavelet image in which each pixel has a wavelet coefficient; a binary image creation unit that uses a wavelet coefficient table that is a threshold value of wavelet coefficients to binarize each pixel of the wavelet image to create a binary image; a crack image creation unit that labels each crack with a block-like figure having geometric connectivity in the binarized image and creates a crack image including the labeled cracks; a segmentation image creation unit that creates a segmentation image by connecting end points of cracks that are determined to be essentially continuous cracks among the multiple separate cracks in the crack image; A crack direction classification unit that classifies cracks of each label in the segmentation image into a plurality of direction categories; The crack direction classification unit includes: a split crack creation unit that creates a plurality of split cracks by dividing a total length of each crack of each label in the segmentation image by a predetermined number; A crack detection device characterized by having a crack direction determination unit that determines the direction of each dividing crack that makes up each label crack using the coordinates of the starting point and ending point of the dividing crack, and determines the direction of the crack based on the directions of all the dividing cracks.
6. The crack detection device described in claim 5, characterized in that the split crack creation unit sequentially performs an operation to divide the crack by the specified number until the length of the split crack becomes equal to or less than a specified target length.
7. The crack detection device described in claim 5 or 6, characterized in that, if there is a direction that accounts for more than 50% of the directions of all the dividing cracks, the crack direction determination unit determines that direction as the direction of the crack.
8. A program for causing a computer to execute the following processes for detecting cracks on a concrete surface, an input image creation step of inputting a photographed image of a concrete surface containing cracks into a computer to create an input image; a wavelet image creation process for performing a wavelet transform process on the input image, or for extracting assumed cracks from the input image based on learning information obtained by having a computer learn the characteristics of cracks on a concrete surface, drawing along the assumed cracks to create assumed crack drawing lines, and for performing a wavelet transform process on a drawing-containing image including the assumed crack drawing lines, or for performing a wavelet transform process on a path image including paths added along assumed crack lines that are assumed to be cracks to the input image, to create a wavelet image in which each pixel has a wavelet coefficient; a binary image creation step of binarizing each pixel of the wavelet image by using a wavelet coefficient table that is a threshold value of the wavelet coefficient to create a binary image; A crack image creation process for labeling each crack with a block-like figure having geometric connectivity in the binarized image and creating a crack image including a plurality of labeled cracks; A segmentation image creation process for creating a segmentation image by connecting end points of cracks that are determined to be essentially continuous cracks among the multiple separate cracks in the crack image; A crack direction classification step of classifying cracks of each label in the segmentation image into a plurality of direction categories, The crack direction classification step includes: a split crack creation process for creating a plurality of split cracks by dividing the total length of each crack of each label in the segmentation image by a predetermined number; The program is characterized by having a crack direction determination process that determines the direction of each dividing crack that constitutes each label crack using the coordinates of the starting point and ending point of the dividing crack, and determines the direction of the crack based on the directions of all the dividing cracks.
Citation Information
Patent Citations
Crack detection method
JP2006162583A
Crack detection method
JP2008185510A
Crack detecting method
JP2008267943A
Crack detecting method
JP2010121992A
Crack detection method
JP2012002531A
Cited By
System and method for dynamically detecting cracks of prefabricated box girder
CN120293739A
