JPH03160349A - Device for detecting crack - Google Patents

Abstract

PURPOSE:To detect automatically and efficiently a crack in the surface of a substance to be inspected, on the basis of prescribed criteria of determination, by subjecting the crack in the surface to an image processing, by subjecting image data to a pattern recognition processing and by executing thereby extraction of specific features of the shape of the crack and determination thereof. CONSTITUTION:A two-dimensional image information on a digital gray scale image of a surface 9 to be inspected, which is obtained from an image pickup camera 3, is inputted to an image processing device 4 and smoothed 41, and it is subjected to density conversion in a half threshold processing element 43 according to a density threshold M determined by a density threshold determining element 42 and then subjected to edge extraction 44. An edge thus extracted is binary-coded 45 according to a threshold value K set beforehand, and isolated pixel data in binary-coded edge image data are removed 46. Then, the data are inputted to a determining/extracting device 6 and subjected therein to a series of processings for blank-filling 61, thinning 62, minute noise removal 63, joining 64 and short-line noise removal 65, and each segment image of the data thus subjected to extraction of specific features is judged in a continuous line figure determining element 66 as to whether it is a crack image or not. According to this constitution, a crack in the surface can be detected automatically and efficiently on the basis of prescribed criteria of determination.

Description

[Detailed Description of the Invention] [Industrial Application Field] This invention automatically detects cracks on the surface of objects to be inspected, such as paved roads and slabs or boards made of various materials, by using image processing and pattern recognition technology. -Relates to a crack detection device for determining cracks.

[Prior art] As a method for evaluating the degree of damage to paved surfaces such as existing roads and runways, for example, 50
Measure the presence or absence of cracks with a width of 1 mm or more for each cm square unit area, calculate the crack rate (%) from the ratio of the number of cracked surfaces to the total number of measurement unit surfaces, and use this as a guide for road surface repair. is being carried out. Conventionally, cracks have been measured for each unit area by visual inspection at the site or by photographing each unit area and visually determining the cracks from the photographic image, and the main method for data output is sketching. On the other hand, the detection of surface flaws using the eddy current deep flaw method is well known for steel and other metal products, and the ultrasonic deep flaw method is also used for materials with relatively dense and uniform structures. [Problems to be Solved by the Invention] First, conventional crack detection on paved roads requires an enormous amount of time for on-site visual observation or photography by workers, visual inspection of the photographic images, and creation of sketches. ,
In addition, the criteria for crack detection differ depending on the inspector.
There has been a problem in that it is extremely difficult to efficiently detect a certain base type.

Eddy current flaw detection and ultrasonic deep flaw detection are methods suitable for automation, but they are limited by measurement conditions such as the material and shape of the object to be inspected, noise in the measurement environment, and the presence or absence of fillers in cracks. However, there are many problems, which limits its use.

Therefore, an object of the present invention is to provide a versatile crack detection device that can efficiently and automatically detect surface cracks based on certain criteria without much restriction on the material structure and measurement conditions of the object to be inspected. It is.

[A Thousand Steps to Solve the Problem] The crack detection device of the present invention detects cracks on the surface of objects to be inspected, such as paved roads and slabs or boards made of various materials, by extracting the characteristics of the crack shape using image processing and pattern recognition technology, and using a certain standard. It is automatically detected and judged by the judgment process in . It is. That is, in order to solve the above-mentioned problem, the crack detection device according to the invention according to claim 1 includes: an image input means for imaging the surface of the object to obtain two-dimensional image information of a digital gray image; Two-dimensional image information is scanned and the density of pixels surrounding the pixel of interest is multiplied by a predetermined weighting coefficient and added using multiple weighting matrix operators for each direction. calculates the difference in gradation in each direction with respect to the pixel, and saves the maximum value of the calculation results in each direction for each pixel of interest, and calculates a series corresponding to the degree of change in density of the contour edge of the gradation image. an image processing means for converting the two-dimensional image information into binary edge image data by generating multi-value data and binarizing the multi-value data using a predetermined threshold; Extraction and determination means detects cracks on the surface of the object by performing pattern recognition processing on image data to extract and determine characteristics of crack shapes. Further, in the crack detection device according to the invention as set forth in claim 2, the image processing means may be configured to perform the image processing according to the direction. Prior to the calculation of the IA light gradation difference, pixels on the high-brightness side corresponding to the background information are determined with respect to the two-dimensional image information at a density threshold value that gives a predetermined area ratio on the density histogram of the image pixels. It includes means for performing half-threshold processing for uniformly converting and fixing the density to the same value as the density threshold. Furthermore, in the crack detection device according to the invention according to claim 3,
The extraction determination means is a hole-filling processing means for converting pixels of the interval between the edge images in the binary edge image data into the same binary level as the edge image at a location where the interval between the edge images is less than or equal to a predetermined number of pixels. , a thinning processing means for converting the edge image data after the hole filling process into a line figure; and a thinning processing means for converting the edge image data after the hole filling process into a line figure; a combination processing means for converting pixels in between to the same binary level as the edge image, and a means for detecting a continuous line figure exceeding a predetermined two-dimensional dimension from among the edge images obtained by the combination processing as a crack image. Contains. [Function] In the crack detection device according to claim 1, the image input means includes an imaging means such as a CCD linear image sensor or an area image sensor, and an optical system, and images the surface of the object to be inspected vertically and horizontally.
xy) For example, 8 bits (25
This method obtains two-dimensional image information of a digital grayscale image (6 gradations). In this case, it is preferable to cover the imaging area with a hood and illuminate the object surface with a low incidence angle using the own illumination system, and the imaging means captures an image of the illuminated object surface from directly above in the normal direction. It is best to arrange the optical system so that it captures images of the deep shadows created by the edges of the cracks while avoiding the influence of external light.

The image processing means scans the obtained two-dimensional image information in, for example, a raster pattern, performs edge detection using, for example, a 3×3 pixel weighted matrix operator, and then performs binarized data conversion processing. In other words, the brightness (density) of cracked areas in two-dimensional image information obtained by human image processing is lower than that of the background area, but this means that it is relatively low, so simply binarization processing is performed. However, the detection of crack images is not successful.Therefore, the edge of the crack is first detected and then it is binarized.The edge detection process in the data conversion process for this purpose is based on a 3x3 pixel image. This is a modification of the edge enhancement method using the so-called Sobel operator, in which the central pixel of the weighted matrix is taken as the pixel of interest, and the value obtained by multiplying and adding the density of the surrounding pixels by a predetermined weighting coefficient is used as the point of interest data. However, especially in this invention,
Using multiple weighted matrix operators for each direction (80,000 directions considering the 4-direction Ci of the XY direction and both left and right diagonal directions between them), each pixel of interest is shaded in each direction with respect to its surrounding pixels. The gradation differences are calculated individually, and the maximum value of the calculation results for each direction is used as data for each pixel of interest. In this way, the data of the calculation results are saved for each pixel of interest, and a series of multivalued data corresponding to the degree of density change of the contour edge of the grayscale image is generated. Next, the two-dimensional image information is converted into binarized edge image data by binarizing each of the multivalued data using a predetermined threshold.

The extraction determination means extracts a crack portion from the binarized edge image data converted by the image processing means,
In that case, features are extracted by focusing on the shape of the crack by processing it using pattern recognition techniques such as connecting line segments or thresholding two-dimensional length dimensions. A distinction (judgment) is performed, and an image as a continuous line figure is detected as a crack on the surface of the object, based on, for example, a two-dimensional dimension exceeding a predetermined value or a degree of directional convergence. The detected data is output to a blotter or display along with the position data of the crack. The invention according to claim 2 (in the crack detection device,
The image processing means includes means for performing half-threshold processing on the two-dimensional image information in order to remove noise in the two-dimensional image information and improve subsequent detection accuracy. In this case, a density threshold is set that gives a predetermined area ratio on the density histogram of image pixels of the two-dimensional image information. For this purpose, the P-tile method, which is known as a threshold value determination method for binary separation into graphic parts and background parts in feature analysis of image information, can be used. That is,
In the P-tile method, the ratio p=s (
F)/S (S) is determined, and a histogram of the pixel density of the entire screen is determined, and in this histogram, the number of pixels counting from the pixels with the lowest density is equal to (i) of the total number of pixels.
-p), but in this invention, the target figure to be detected is a relatively thin line figure crack, so the threshold determined by the P-tile method (for example, when P=80% or more ) at the beginning, a certain amount of experience points (e.g. 20-30%)
The accuracy of noise removal can be improved by setting the value estimated as low as the concentration threshold, and gradually reducing the empirical value as measured results are accumulated. In this half-threshold processing, prior to calculating the gradation difference for each direction, the pixel density of the two-dimensional image information (in contrast, the pixel density on the high-brightness side corresponding to the background information with the density threshold as the boundary) is It is converted and fixed to the same value as the density threshold. This improves the binarization accuracy in the data conversion process for the subsequent edge extraction process. In the crack detection device according to the invention as set forth in claim 3,
The extraction determination means includes several means for improving crack detection accuracy and determination accuracy. That is, the binarized edge image data converted by the image processing unit is first processed by a hole filling unit, and there is a portion in the binarized edge image data where the interval between edge images is less than or equal to a predetermined number of pixels. Then, the pixels at the interval between the edge images at that location are converted to the same binary level as the edge images. As a result, the gaps between adjacent edge images are filled in to form a thick line image, and then the edge image data after the hole filling process is thinned in the direction of the center of the line width by a thinning processing means and converted into a line figure. Furthermore, the edge image data after the line thinning process is processed by a combination processing means, and pixels on the shortest distance line connecting the end points of the edge image are The thinned edge images are converted to the same binary level as the edge images, and thinned edge images that are close to each other are combined in this way. Among the obtained edge images, continuous line figures exceeding a predetermined two-dimensional dimension are detected as crack images by the determination means. In this case, short ones are rejected as minute flaws other than cracks or noise, and short ones that are determined to be loop-shaped based on the degree of convergence of direction by analyzing the directional component of the continuous line figure are rejected as obstructions other than cracks. It is considered to be a protrusion attachment (pebbles, etc.) and is rejected. In order to better understand the features and advantages of the present invention, an embodiment applied to a crack detection device for a paved road surface will be described below with reference to the drawings. [Embodiment] FIG. 1 is a block diagram showing the configuration of a crack detection device for a paved road surface according to an embodiment of the present invention, in which a hood l mounted on a cart (not shown) covers the inspection area of the road surface 9 from external light. , an illuminator 2 using a halogen lamp or the like is held inside to illuminate the road surface inspection area with high brightness from a low angle. The imaging camera 3 constituting the image input means uses an imaging device such as a CCD image sensor to capture 8 pixels of 512 x 512 pixels.
It obtains two-dimensional image information of a digital grayscale image of bits (256 gradations), and is provided in a fixed relationship with the hood 1 so as to image the road surface inspection area from directly above. When the inspection area of the road surface 9 is imaged by a portion, the road surface portions other than the cracks become highly bright due to the low-angle illumination light, and the crack portions are emphasized by the strong shadows of their edges, so that the digital grayscale image obtained from the imaging camera 3 In this case, the pixels corresponding to strong shadow areas such as cracks (C) are dark and low-density, and the background areas other than cracks 5 are bright, uniform, and high-density, and the relative density difference between the two is greatly emphasized. The image processing device 4 that receives the input image information from the camera 3 includes a smoothing processing unit 41 for reducing noise during observation on the input two-dimensional image information, and a smoothing processing unit 41 for reducing the noise during observation on the input two-dimensional image information, and the image information after the smoothing processing. a density threshold determination unit 42 that calculates a histogram of the density of the pixel, and determines the density threshold M by subtracting a separately given empirical value from the threshold K obtained by the P-tile method according to the area of the histogram; , a half-threshold processor 43 that performs half-threshold tji processing by density-converting the two-dimensional image information after the smoothing process using the density threshold M, and a half-threshold processor 43 that performs the above-described transformation on the two-dimensional image information after the half-threshold process. an edge extraction processing unit 44 that generates a series of multivalued data corresponding to the degree of density change of the contour edge of the grayscale image by performing data conversion processing using an additive matrix operator according to Sobel method C; a binarization processing unit 45 that binarizes the multi-valued data using a threshold obtained by externally correcting the threshold K obtained by the P-tile method in the density threshold determination unit 42 and outputs the digitalized edge image data; It also includes an isolated point removal processing unit 46 that replaces the binarized information of a pixel isolated from surrounding pixels with background data.
A monitor device 5 for processing image data is connected to this image processing device 4 .

The binarized edge image data output from the image processing device 4 is binary edge data from which isolated points have been removed, and the judgment extraction device 6 processes this edge data for pattern B recognition, thereby identifying the crack shape. Characteristic extraction and determination are performed to detect cracks on the surface of the subject and output to the output device 8. In this embodiment, the determination extraction device 6 is constituted by, for example, a microcomputer, and its functional elements include a fill-in processing section 61 and a thinning processing section 62, as shown in the figure.
and minute noise removal processing section 63) 2. Line segment combination processing section 6
4, a short line noise removal processing section 65, and a continuous line figure determination section 66.

First, feature extraction processing is performed by the elements up to the hole filling processing section 61 and the six short line noise removal processing section 65.

The hole-filling processing unit 61 determines that the interval between edge images in the binarized edge image data from the image processing device 4 is less than or equal to a predetermined number of pixels, and identifies the pixels within the interval as an edge image. The thinning processing unit 62 converts the edge image data after the hole filling process into a line figure according to the general thinning process T method. The minute noise removal processing unit 63 removes line segment data having a predetermined number of pixels or less and isolated data in the edge image data after the line thinning process by replacing them with background data C. For the edge image data from which minute noise has been removed in this way, the combination processing unit 64 converts the pixels between the end points to the same number of pixels as the edge image for locations where the interval between the end points of the edge image is less than or equal to a predetermined number of pixels. Convert to value 1 novel, and connect line figures whose end points are close to each other into a series of line segments.

This series of line segments. Among them, those smaller than the set two-dimensional size are replaced by background data and removed by the short line noise removal processing unit 65, leaving only those that appear to be cracks in the image data.

A continuous line figure determination unit 66 that determines whether each line segment in the image data after feature extraction is a crack or not determines each edge image output from the short line noise removal processing unit 65, , and those whose directional components have biased convergence are detected as crack images.

A monitor device 7 for displaying processed images of each section and an output device 8 for printing out or displaying the output of the continuous line figure judgment section 66 are connected to the judgment extraction device 6.
There is.

The crack detection operation performed by the image processing device 4 and the determination extraction device 6 in the structural crack detection device shown in FIG. 1 will be described below with reference to FIGS. 2 to 9.

First of all, the input information to the image processing device 4 is 8 bits (256
This is two-dimensional image information of a digital gradation image (gradation), and background noise such as gradation and color unevenness of the road 9 is reduced in advance by making it uniform brightness by low-angle illumination by the illuminator 2, and moreover, it reduces cracks. Since we are trying to focus on the edge part and make its shadow stronger, the density of the boundary part of the crack is emphasized in the image information compared to the part C that is not a crack on the road surface before detecting the edge part. .

The flow of the image conversion algorithm in the image processing device 4 that handles such input images is as shown in FIG.

In other words, 8 bits of input 512x512 pixels (
The two-dimensional digital image information (256 gradations) is first smoothed by the matrix operator shown in FIG. Although the input image information is density-enhanced, it still contains considerable noise such as brightness unevenness, so it is necessary to smooth this to reduce noise during observation.
The matrix operator for this smoothing process has a 3 x 3 matrix configuration as shown in Figure 3, and the density of the surrounding pixels is multiplied by a coefficient of 178 for the center pixel of interest, and these are added. Import the value as the density value of the point of interest,
This is arithmetic processed while scanning the pixels of the entire screen to smooth out isolated variations in brightness (density) between pixels.

The density threshold determining unit 42 takes a histogram of pixel density for the image information after the smoothing process, and subtracts a separately given empirical value from the threshold K obtained by the P-tile method according to the area of the histogram. Then, determine the concentration threshold M. Figure 4 shows an example of a histogram of a human image. The density threshold value M is used in the density conversion process in the half-threshold processing unit 43 to equalize the density of pixels in the luminance portion corresponding to the road background. The modal method, discriminant analysis method, P-tile method, differential histogram method, etc. are known as methods for determining the binarization threshold for graphic analysis in image processing technology, but in the case of crack detection, Even if the histogram is taken, the valleys in the histogram required for the mode method or discriminant analysis method often do not appear, so here we focus on the area of the histogram, and as shown in FIG. The threshold value K is determined from the histogram (a) using the P-tile method, and from this the threshold value K is determined by an external setting of 20 to 30%.
The density threshold value M is determined by subtracting . This half-threshold processing using the density threshold M can be said to limit the range of the absolute density of the cracked image, and as shown in FIG. 5(b), the density data of each pixel of the smoothed input image is compared with the density threshold M. ,M
The above density data is uniformly converted to density M to prevent variations in image brightness from affecting subsequent edge extraction. In other words, there is a certain degree of difference in brightness (density) between the properties of the road surface itself, such as color unevenness, and the properties of the cracks themselves, and even if there is uneven brightness on the road surface, it is relatively brighter than the cracks. be. Therefore, we take the density histogram of the pixels of the input image information, and convert the density higher than 50% to a uniform density, assuming that it originates from the road surface.This eliminates the noise in pixels/part C with a density higher than 50%. Make sure that it stays in place. This concentration threshold value of 50% may be corrected by adjusting the above-mentioned correction amount based on the empirical value from past measurement data.

The image information thus density-converted by the half-threshold processing unit 43 is then subjected to crack detection through binarization processing.
Although the brightness of the cracked area is lower than that of the surrounding road surface area, this is a relative change in brightness, and simply performing binarization processing will not provide satisfactory results for crack detection. Therefore, here, the edges of the cracks are first detected by the edge extraction processing section 44, and then the binarization processing is performed.

Known methods for extracting the edges of figures in general image information include the first-order or second-order differential method, Sobel method, Roberts method, Prewitt method, and Labrassian method.
Edge extraction by the edge extraction processing section 44 of this device is performed using four types of weighting matrices that are different for the vertical, horizontal, and diagonal directions as shown in FIG. This is 3
An edge enhancement method using the so-called Sobel operator, in which the central pixel of a weighted matrix of ×3 pixels is set as the pixel of interest, and the value obtained by multiplying the density of the surrounding pixels by a predetermined weighting coefficient and adding the resultant value is the point of interest data. However, in this invention, in particular, each pixel of interest is calculated using a plurality of weighted matrix operators for each direction in a total of four directions (eight directions considering polarity): the XY directions and the left and right diagonal directions between them. For each pixel, the difference in gray scale in each direction with respect to surrounding pixels is calculated individually, and the maximum value of the calculation results in each direction is used as data for each pixel of interest. In Figure 6, each operator calculates the absolute value, and the operator (a
) indicates the reactivity in the vertical direction, operator (b) indicates the reactivity in the horizontal direction, operator (C) indicates the reactivity in the diagonal direction downward to the right, and operator (.j) indicates the reactivity in the diagonal direction downward to the left. It is for seeking sex. For example, when each element of the matrix is represented by the following, the density value of the pixel of interest that has undergone data conversion processing by operator (a) becomes X5=X1+2X2+X3X7-2XaX9.

In this way, in edge extraction by the edge extraction processing unit 44, calculations are performed using each operator for each pixel of interest,
Using 4 types of operators for each pixel of interest? 'Among the A calculation results, the data of the calculation result with the largest absolute value is saved as the conversion data of the pixel of interest. In this way, a series of multivalued data corresponding to the degree of density change of the outline edge of the grayscale image in the two-dimensional image information is generated together with the position data, and then each multivalued data is predetermined by the binarization processing unit 45. The two-dimensional image information is converted into binarized edge image data in which parts with the possibility of cracking and parts without the possibility of cracking are binarized by binarizing using the set threshold. As the binarization threshold in the 43-filling section 45, a threshold value K obtained from the histogram in the density threshold determining section 42, obtained by removing P-tiles, is given a correction value that can be adjusted from the outside.

The binarized edge image data obtained by converting in this way is then processed by the isolated point removal processing unit 46. For example, this is performed by scanning the 8 pixels around the pixel of interest using a 3×3 pixel matrix operator. If all pixels are background data, it is a scanning operation/A filling that converts the data of the pixel of interest into the same background data,
This removes isolated pixel data in the binarized edge image data, making it easier to determine if a crack is to be extracted.

Next, the flow of the algorithm for crack feature extraction and determination processing in the determination extraction device 6 is shown in FIG.

Here, it is important to keep in mind the nature of the crack information included in the binarized edge image data manually inputted to the judgment extraction device 6. Although both ends of the crack are extracted, the middle part is a dent in the road surface, etc. Edges may become unclear due to the effect of As a result, noise caused by obstructions such as pebbles, dirt, etc. other than cracks may also be extracted as line segments. In the crack feature extraction by the judgment extraction device 6, pixels between closely extending edge images are filled in with the same data as the edge images, and then the crack images are captured in a linear manner,
Remove garbage-like minute noise, join and connect adjacent end points of line figures, and remove those that do not have a crack-like size (length) shape. The processing unit 61 performs hole-filling processing to prevent the above-mentioned situation in which the crack ends can be extracted but the middle part disappears or can only be extracted as intermittent line segments. Due to the characteristics of the image, a general hole-filling method cannot be applied to closed figures, so as shown in Figure 8, calculations are performed sequentially in the two-dimensional direction on a linear pixel area of, for example, 6 pixels in each of the X direction and Y direction. If the pixels at both ends of the area have edge image data and the pixels in between lack data, processing is performed to replace and fill in the pixels between these with edge image data. This hole filling method is extremely simple and effective for filling holes in cracked images. Needless to say, the number of interval pixels to be filled in is arbitrary depending on the setting of the linear pixel area.

The binarized edge image data subjected to hole filling processing in this manner is then subjected to line thinning processing by a thinning processing section 62 according to a general method. This is because crack images generally have line graphic properties. R
J- eggplant, ψ-h-F! 4 f-0
1 F sword nc k l. ”r/7'1
This is because it has a 7k-long shape. In this thinning process,
Both end points of each filled-in line segment are fixed, and the thick line portion between them is thinned from both sides in the width direction.

The minute noise removal processing section 63 removes line segments and isolated point data of several pixels or less, and this processing is no different from normal image processing technology.

Next, the combination processing unit 64 calculates the interval distance between the end points of each line segment of the image data, compares this with a set threshold, converts the pixels between adjacent end points within a certain distance into edge image data, and converts the pixels between the end points of each line segment of the image data into edge image data. Perform processing to connect the parts. This is to detect crack images, which are actually easily extracted as discontinuous figures, as continuous line figures as much as possible, and for this purpose, prior to the combining process, the minute noise removal processing unit 63 removes noise images other than cracks as much as possible. It is better to remove it.

The short line noise removal processing unit 65 removes line segment images with short dimensions other than crack image candidates from the image data after the combination processing. This is threshold processing based on length or area). Each line segment image of the edge data whose features have been extracted through the series of processes described above is a crack candidate, and the continuous line figure determination unit 66
Then, it is determined whether or not each of these image units is a crack image.

Although the crack image extracted from the road surface image to be inspected has already been extracted as a continuous line figure to some extent before being input to the determination unit 66, the noise present in the image data has not been completely removed. It is not something that will be done. Therefore, it is still not possible to determine whether an image unit is a small crack or noise based only on the image unit of interest, and it is necessary to make a determination based on the surrounding situation. The distinction between cracked images and noise is based on the isolation and closeness of the images. Figure 9 shows an example of a crack image. In the determination process in the determination unit 66, in FIG. 9, the sum of the length component Lx in the X direction and the length component Ly in the Y direction of the line segment of the image unit is set as A (=Lx+t,y),
First, use this as the criterion. Two threshold values A. and A2 (however, A
+ < A2 ), if A < A I, reject it as noise, if A > A +, determine it to be a crack candidate and save it, if A > A2, determine it to be a crack image. , A r < A < A
In case 2, the determination is made by focusing on the directionality of the image. That is, the data for each image is digitized in, for example, 8 directions at equal angular intervals within the XY two-dimensional plane, and a histogram for each direction is taken. It is determined by an appropriate threshold whether the ratio is high, and if the directional component has a bias exceeding a certain level, it is determined to be a line figure (that is, a crack image). If a large deviation in direction does not appear, there is a high possibility that the shape is loop-shaped, so these should be rejected as something other than cracks originating from holes or pebbles on the surface of the object to be inspected. The determining unit 66 performs the above-described determination process and labels only the image data determined to be a cracked image, and outputs each label to the output device 8 as its position coordinate value.

Figure 10 shows an example of the results of extracting a crack image using the judgment extraction device described above. FIG. 10(a) is a monitor image of the binarized edge image data provided from the image processing device 4, which is extracted by edge operator processing using the operator in FIG. 6, then binarized, and isolated points This is an image output from the removal processing unit 46. FIG. 10(b) is an output monitor image of the hole-filling processing unit 61, and FIG. 10(C) is a monitor image after the line has been thinned by the line thinning processing unit 62. FIG. 10(d) is a monitor image after fine noise has been removed from the thin line image and then combined processing is performed in the combining processing unit 64, and FIG. This is the later monitor image, and this is the determination unit 66
This is the crack feature extraction image input to . [Effects of the Invention] As described above, according to the present invention, image information obtained by optically imaging the surface of an object to be inspected is combined with an image processing device optimized for crack detection and feature extraction/ Since the processing is performed using a judgment device, cracks on the surface can be detected efficiently and automatically using fixed judgment criteria without any restrictions on the material structure or measurement conditions of the target object, and can be used to detect surface cracks on surfaces such as paved roads and slabs or boards made of various materials. It is possible to automatically detect and judge cracks on the surface of a specimen by extracting characteristics of the crack shape using image processing and pattern recognition technology, and performing judgment processing based on certain criteria, and output the data along with the crack position information. .

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing the configuration of an embodiment of an apparatus in which the present invention is applied to crack detection on a paved road surface, FIG. FIG. 4 is a diagram showing an example of a density histogram of a human-powered image, and FIGS. 5(a) and 5(b) are diagrams for explaining density threshold determination and half-threshold processing. Density histogram diagram and half-threshold processing human output characteristic diagram, Figures 6(a) to (d)
) is an explanatory diagram showing an example of a direction-based weighted matrix operator for edge extraction processing, Fig. 7 is a flowchart showing the flow of the crack extraction/judgment algorithm in the judgment extraction device, and Fig. 8 is a diagram showing target pixels for hole-filling processing. An explanatory diagram showing the area, FIG. 9 is an explanatory diagram showing an example of a crack image, and FIGS. 10 (a) to (
e) is an explanatory diagram showing an example of a monitor image of crack image extraction processing performed by the determination extraction device. (Explanation of symbols of main parts) 1: Foot, 2: Illuminator, 32 Imaging camera, 4 Image processing device, 5 Monitor device, 6: Judgment extraction device, 7: Monitor device, 8: Output device, 9: Road surface to be inspected, 41: Smoothing processing section, 42: Density threshold determining section, 43: Half threshold processing section, 44:. ．． T-tsuji extraction processing section, 45: Binarization processing section, 46. Isolated point removal processing unit, 6l: Hole filling processing unit, 6
2: Thinning processing unit, 63: Small noise removal processing unit, 64
: combination processing unit, 65: short line noise removal processing unit, 66: continuous line figure determination unit.

Claims (1)

[Scope of Claims] 1. Image input means for capturing an image of the surface of a subject to obtain two-dimensional image information of a digital grayscale image, and scanning processing of the obtained two-dimensional image information to detect pixels surrounding a pixel of interest. Using a plurality of direction-specific weighted matrix operators that multiply the density of the target pixel by a predetermined weighting coefficient and add the resultant values, the direction-specific gray scale difference between each pixel of interest and its surrounding pixels is computed, and The maximum value of the other calculation results is saved for each pixel of interest to generate a series of multi-value data corresponding to the degree of density change of the contour edge of the grayscale image, and this multi-value data is an image processing means for converting the two-dimensional image information into binarized edge image data by binarizing it using a threshold; and extracting and determining characteristics of crack shapes by subjecting the binarized edge image data to pattern recognition processing. 1. A crack detection device comprising: an extraction determination means for detecting cracks on a surface of a subject by detecting cracks on the surface of a subject. 2. The image processing means, prior to calculating the gray scale difference for each direction, processes the two-dimensional image information at a density threshold that gives a predetermined area ratio on the density histogram of the image pixel. 2. The crack detection apparatus according to claim 1, further comprising means for performing half-threshold processing for uniformly converting and fixing pixel densities on the high-brightness side corresponding to background information to the same value as the density threshold. 3. Hole-filling processing in which the extraction determination means converts the pixels of the interval between the edge images in the binarized edge image data into the same binary level as the edge image for locations where the interval between the edge images is less than or equal to a predetermined number of pixels. means, a thinning processing means for converting the edge image data after the hole filling processing into a line figure;
a combination processing means for converting the pixels between the end points into the same binary level as the edge image for locations where the interval between the end points of the edge image of the edge image data after the thinning process is less than or equal to a predetermined number of pixels; 2. The crack detection apparatus according to claim 1, further comprising means for detecting a continuous line figure exceeding a predetermined two-dimensional dimension from among the edge images obtained by the above process as a crack image.