JPH03138775A - Crack measuring system - Google Patents

Abstract

PURPOSE:To improve the measurement precision of the crack width by using the crack position and the crack width, which are calculated on a compressed picture of high compression ratio, as a guide map to stepwise perform binarization, noise elimination, conversion to centerlines, and measurement of the crack width up to an original picture. CONSTITUTION:The picked-up image of an object surface is stepwise compressed, and this compressed picture is subjected to extraction processing using the line extraction circular filtering, binarization, conversion to centerlines, vectoring, noise separation processing of crack candidate parts to extract a crack part, and the crack width is approximately calculated in accordance with this extracted crack part by the linear operator method, and approximate crack position and width are used as the first guide map. Hereafter, in the compression order opposite to that of the stepwise compression processing, crack position and width calculated on the compressed picture of high compression ratio are used as the guide map and the crack part of the picture whose compression ratio is lower than this compressed picture by one step is repeatedly stepwise subjected to binarization, noise elimination, conversion to centerlines, and measurement of the crack width up to the original picture. Thus, cracks are measured without an influence of dirt or the like on the surface.

Description

[Detailed Description of the Invention] [Object of the Invention] (Industrial Application Field) The present invention relates to a crack measurement system that analyzes and measures cracks generated in an object by image processing. Image processing can detect cracks with a width of 0.11Im to 3.0+a+s that occur not only in structures such as roads, water storage dams, and waterways, but also in concrete and metal that make up objects such as cars and airplanes. This invention relates to a crack measurement system that analyzes cracks and measures their width, position, shape, distribution, etc.

(Conventional technology) Buildings, bridges, roads, water storage dams, waterways, automobiles,
Cracks occur in reinforced concrete, metal, and the like that make up structures and objects such as airplanes due to aging and various other factors. If such cracks are left untreated,
The durability and earthquake resistance of structures and objects decrease.

Therefore, it is necessary to accurately measure the width etc. of this type of crack, but in general, in order to accurately measure a crack with a width of 0.1 to 3° Oas through image processing, the pixels constituting the image The density requires at least 4 pixels per 0.1 m+s (that is, a resolution of about 0.025 to 0.05 mm per pixel).

When an image of a crack is constructed with this pixel density, the crack has the general characteristic that it has a "linear structure" (0,
In addition to ``2m-width below'', it also has the characteristic of ``a long and large band-like area'' (with a width of 0°2mII or more).

Also, to determine the illumination of cracks through image processing is to process an image of the cracks captured in a photograph, etc., but this image contains shadows caused by the cracks. Extracting cracks in processing involves extracting areas where the shadow that appears in the image is darker than the other surface of the object. Because of the presence of shadows caused by adhesion, surface discoloration, surface irregularities, etc., it is not possible to simply binarize and extract cracks based on the brightness and darkness of the object's surface.

To detect cracks in an image of the surface of an object, it is necessary to examine the entire image of the entire surface of the object. An image processing algorithm is needed to detect crack patterns.

(Problems to be Solved by the Invention) As mentioned above, cracks are classified into two parts: a linear structure part and a long, band-like part. Image processing algorithms for identifying such crack parts are
Depending on the width of the crack, there are two options: choose an algorithm that analyzes linear structures, or an algorithm that analyzes long and large areas as cracks.
This requires the use of two algorithms, and there is a problem in that it cannot be uniquely determined regardless of the width of the crack.

Furthermore, in addition to cracks, there are brightness and darkness on the surface of objects due to the adhesion of dirt, etc., and there is a problem in that cracks cannot be simply extracted by binarizing them.

Furthermore, to detect cracks from an image of the surface of an object,
Although it is necessary to perform image processing on the entire surface image,
If you set the image resolution to the pixel density described above,
For example, when capturing an image of the surface of a 50 cm x 50 cm object, the image data becomes enormous, requiring a memory capacity of 400 MB, which increases the burden on image processing and reduces efficiency.

The present invention has been made in view of the above, and its purpose is to repair cracks ranging from very thin cracks with a width of about 0.1 mm to thick cracks with a width of about 2 to 31 mm, without being affected by surface dirt, etc. Accurately and efficiently recognize tlPJ without
Our objective is to develop a crack measurement system that will

[Structure of the invention]

(Means for Solving the Problems) In order to achieve the above object, the crack measurement system of the present invention is a crack measurement system that measures cracks generated on the surface of an object by image processing, and which captures an image of the surface of the object. hierarchical compression means for hierarchically compressing the image to a predetermined compression ratio while applying line-preserving smoothing filtering to the image;
The image compressed to a predetermined compression ratio by the hierarchical compression means is subjected to crack candidate extraction processing using line extraction circular filtering, binarization, skeletonization, vectorization, and noise separation processing to extract cracks. an extraction means for extracting the
An approximate crack width calculation means for calculating a crack width from the crack extracted by the extraction means by a straight line operator method, and the crack position and width calculated by the approximate crack line extraction means and the approximate crack width calculation means as an initial guide map. From then on, using the crack position and width calculated in the compressed image with the higher compression ratio in the reverse order of the order compressed by the hierarchical compression means as a guide map, the crack position and width of the image with the compression ratio one step lower than the compressed image are used as a guide map. Step by step 2
The main feature is that the detailed crack extraction needle 7111 performs value conversion, noise removal, skeletonization, and crack width measurement until the original image is reached.

(Function) In the crack measurement system of the present invention, a captured image of an object surface is compressed hierarchically, and crack candidate parts are extracted using line extraction circular filtering on this compressed image.
The crack width is roughly calculated using the straight line operator method from the crack portion extracted by value conversion, core line conversion, vectorization, and noise separation processing, and this rough crack position and width is used as the first guide map. Using the crack positions and widths calculated from compressed images with a higher compression ratio in the reverse order of the compression process as a guide map, the cracks in images with a compression ratio one step lower than this compressed image are binarized step by step. , noise removal, core line formation, and crack width measurement are repeated until the original image is reached.

(Example) Hereinafter, the present invention will be explained in detail using the drawings.

FIG. 1 shows a crack A? according to an embodiment of the present invention. FIG. 1 is an overall configuration diagram of the fixed system. As shown in the figure, the crack 'AP1 constant system of the present invention, for example, 0.1 mm formed on the surface of a structure made of concrete or the like,
It uses image processing to accurately measure cracks in the range from about 3mm width to 3II 1m wide.
The photoelectric image reading device 3 reads an image of the photographic film 1 of the surface of the structure taken with a camera, and the image processing device t5 analyzes the read image to detect cracks formed on the surface of the structure. Width, length, position, etc. are measured, and finally characteristic data including these various data is output from an output device 7 consisting of a personal computer or the like.

It outputs distribution, characteristic diagrams, crack diagrams, etc. The image reading device 3 includes, for example, a CCD camera and a microscope to read the image on the photographic film 1. In this embodiment, the image reading device 3 reads an image of the surface of the structure on the photographic film 1 taken with a camera, and the image processing device 5 analyzes the image. Instead of reading the image on the film 1, the surface of the structure may be directly imaged with the image reading device 3, and this captured image may be used directly.

The image reading device 3 reads the photographic film 1 by dividing a crack width of 0.1 mm, which is the detection target, into four parts.
．． Using a film reader that can read cracks at a reading density of 1m+s/4 pixels,
The entire frame of the film 1 is subjected to shading correction and sensor noise correction at the same time, and the read image information is sequentially sent to an image processing device 5 consisting of a computer, converted into numerical image data, and subjected to image processing.

The image data read from the film 1 by the image reading device 3 and supplied to the image processing device 5 is
The cracks are extracted and measured by sequentially undergoing the following general crack extraction and measurement process (A) and detailed crack extraction and measurement process (B).

(A) General crack extraction and measurement processing (1) Image compression processing and hierarchical structuring (2) Crack extraction in 1/32 compressed images (3)
11! Rough Crack Position and Width Measurement (B) Detailed Crack Extraction and Measurement Processing Figure 2 is a diagram showing the general flow of crack extraction measurement image processing in the crack measurement system shown in Figure 1, along with explanatory diagrams. Next, a description will be given with reference to FIG.

As mentioned above, when the image on the photographic film 1 is read by the image reading device 3 (step 11 in FIG.
0), this read original image data 9 (second
) is subjected to compression and hierarchical structuring processing (step 1
20) Hierarchically compressed as 1/4 → 1/8 → 1/16 → 1/32 (steps 121, 125, 126.1
27).

That is, the original image data 9 is first reduced to 1/4 and then sequentially compressed to 1/2' to become 1/2' compressed image data 1o to 1/32 compressed image data 11.

Note that this compression processing is performed using "line-preserving filtering" that leaves the line structure, and "reduction thinning using the maximum value selection method (because the image data value of cracked areas is larger than that of areas such as concrete)". By applying such hierarchical compression, as you move up the hierarchy,
In other words, as it is compressed, the wide cracks that are considered to be band-like regions gradually transform into narrow line structures.
In addition, the line structure characteristics of narrow cracks are preserved as they are in the upper image. By compressing the image in this way, the cracks have only linear characteristics regardless of their width, and by applying a line extraction algorithm, it is possible to image analyze the macroscopic line structure of the cracks. It becomes like this.

Next, using a line structure extraction algorithm as described later on the image compressed to 1/32 in this way, cracks are extracted by focusing on the line structure, which is a feature of the image, and the extracted cracks are Perform skeletonization, vectorization, and
Processing such as noise separation is performed to extract a rough crack pattern (step 130). In addition, in the binarization of cracks performed in the line structure extraction algorithm, the compressed (reduced) image is further divided into smaller blocks in order to prevent the negative effects of local dirt and smudges on the surface of the object as much as possible. The image is divided into blocks, and a binarization threshold is calculated for each divided block to perform binarization.

Then, for the crack pattern extracted in this way, the crack width is roughly calculated using the direct operator method (step 140).

Next, based on the rough crack pattern and width extracted from the 1/32 compressed image as described above, cracks are searched for in the lower layer image step by step in the reverse order of compression, and this is used when reading the film. Detailed crack extraction and measurement processing is performed repeatedly until the original image is reached, thereby improving the measurement accuracy of the crack center position and crack width (step 150).

In other words, in this detailed crack extraction and total ΔIII processing, the process of searching for cracks on the lower layer image data using the crack data measured on the upper layer image data as a guide map is compressed to 1/32 of that in Figure 2. The crack measurement 71IIJ is performed step by step from the entire image 13 to the lower layer images such as 1/16-1/8-1/4-original image 15 to refine the crack measurement (steps 151 to 154). ).

To explain more specifically, it is not possible to measure the exact position and width of the crack using the macroscopic crack pattern extracted from the 1/32 compressed image, so the approximate crack width is used to measure the crack width in the underlying image. Exploring the cracks,
By repeating this process up to the film reading original image, the accuracy of the crack position and crack width is improved by 71p1 in total.

The search range is a band-shaped range around the crack pattern extracted from the upper image.

In other words, a band-shaped search area 2 to 3 times the crack width (calculated at the same time as the crack pattern) is set up on both sides of the crack pattern (line) found in the upper image, and the local search range is sequentially searched. Perform binarization to distinguish cracks. After that, the center position of the crack is determined in the differentiated crack part, and the crack width (thickness of the binarized part) is also calculated at the same time. This also improves the accuracy of crack width measurement.

The crack data consisting of the crack position and width searched as above is supplied from the image processing device 5 to the output device 7, and is stored in an external storage device such as a magnetic disk, and the crack data is analyzed. , a crack diagram, a crack characteristic diagram, etc. are created.

FIG. 3 is a flowchart showing a more detailed total 71p1 process of the crack measurement system described above. The detailed crack measurement process shown in the same figure consists of the general crack extraction and measurement process and the detailed crack extraction and measurement process, which follow the crack photographic film reading process described above, and the detailed crack extraction and measurement process described below. In addition, in the flow of FIG. 3, the same steps as those in the flow of FIG. 2 are given the same numbers.

(A) Rough crack extraction and measurement processing (1) Image compression processing and hierarchical structuring (a) Creation of 1/4 reduced image (b) Line-preserving smoothing filtering (c) 1/4 reduction from 1/4 reduced image 8.1/16.1/3
Hierarchical compression to 2 (2) Extraction of cracks in 1/32 compressed images (a) Extraction of line structures Calculation of feature quantities by circular filtering and binarization (C) Core lineization (d) Vectorization of cracks (e) Noise Separation (3) Rough measurement of crack position and width (a) Manual specification of crack measurement area (b) Calculation of conversion coefficient from photo coordinates to actual structure coordinates (C)! Confirmation of rough crack extraction results (d) Calculation of rough crack width (e) Registration of roughly extracted crack vector and width CB) Detailed crack extraction and measurement processing (1) Selection of subdivided images where cracks exist (2) Crack position Search and binarization (3) Noise removal (4) Core line formation (5) Crack width measurement (6) Crack meter i1? Scale conversion of J result Hereinafter, detailed processing will be sequentially explained in accordance with the flowchart shown in FIG. 3 and with reference to FIGS. 4 and subsequent figures.

First, the cracked photographic film reading process (step 110) is performed on the entire cracked photographic film (effective screen 50).
mmX50+w) is read by a film reader. To read the film, 1.28mm x 1.2mm (
This is sequentially divided into minute rectangular areas (corresponding to the reading area of the CCD area sensor, which is the sensor unit of the Film League, of 512 x 480 pixels), and one entire frame of film is read. Therefore, the cracked image is composed of approximately 2,000 sub-images as shown in FIG. 4, including the effective screen of the photographic film and its surrounding areas.

In addition, when reading images, it is assumed that the crack is photographed at a 1/10 scale, and a crack of 0.1 dragon width is divided into 4 pixels (one pixel is 2.5 μ on the film, and 1 pixel is 2.5 μ on the structure). The reading density is adjusted to 0.1 m+s/4 pixels (0.025+n+s).

Generally, when an image is formed through a lens, the radial illuminance differs between the center of the lens and its periphery, and the image around the lens becomes darker than the center (this phenomenon is called peripheral dimming of the lens). Furthermore, the sensitivity of each of the light receiving elements (512×480 elements in total) making up the CCD area sensor differs slightly.

Therefore, even if a film with uniform density is read, the read image data will not result in an image with negative numerical values due to peripheral dimming of the lens and non-uniformity inherent to the CCD area sensor element.

When reading film in crack image processing, such image non-uniformity is corrected each time a subdivided image is read (shading correction and sensor noise correction).
(Step 111 and Step 112). As shown in Figure 5, the correction method involves importing data (considered to be the brightest uniform image) from the film reader without setting the film in advance, and dividing the sequentially read cracked subdivided images by this data. A method is used to normalize the loaded image data.

That is, the shading and sensor noise correction shown in FIG. 5 is performed by setting the film 1 as shown in FIG. 5(a) and reading the subdivided image data 0(i, j) into FIG. 5(b). As shown, shading correction and sensor noise correction image data P (i, j) is obtained by dividing by the image data S (i, j) read without setting the film 1 and multiplying this quotient by a constant. In addition, i-
1 to 512, and J-1 to 480.

After reading the image data as described above, in the rough crack extraction and measurement process, rough crack pattern extraction and total a11 processing are performed from the image read as a subdivided image of the entire surface of the cracked photographic film. Creation of 1/4 reduced image (step 1) in data compression and hierarchical structuring (step 120)
21).

In creating this 1/4 reduced image, in order to efficiently perform line-preserving smoothing filtering (step 123), the original image is Perform the reduction.

Specifically, since the original image is 0.025 m/1 pixel, it is reduced to 1/4. This reduction method is a 4 pixel x 4 pixel 1
An average value is calculated from six pixels, and this value is used as one new pixel.

Then, as shown in FIG. 6(a), the 1/4 reduced image is divided into points 1 to 1 for each point (i, j) of the reduced image.
Calculate the average value AVE n (n −1 to 8) in the eighth direction, search for the maximum value among the average values AVEn, and find the point (
Line-preserving smoothing filtering is performed using the values of t, j) (step 123).

This is because the image data value of the cracked part is higher than that of the surrounding area. This is done to prevent isolated shadows other than cracks from being mixed in as noise in the subsequent compression process.The values of cracks and other linear shapes do not change, but isolated points are smoothed. , so that the value becomes smaller.

Note that the length of the filter was determined to be approximately 1 + n + e on the structure, taking into account the linearity of cracks.

Next, the 1/4 reduced image obtained as described above is shown in Figure 6 (
As shown in c), it is compressed hierarchically to a power of 1/8.1/16.1/32 (steps 125 to 125).
127). Image compression is performed by 2 as shown in Figure 6(b).
This is done by thinning out using the maximum value selection method, which selects the maximum value from the four pixels of ×2 as the representative value.Thin cracks are left as thin lines by this process, and thick cracks gradually become thin line structures. converted.

Incidentally, for example, for a minimum detection crack width of 0.1, it is possible to adjust the minimum film reading density to 0.025 wus/1 pixel (corresponding scale on the surface of the object to be measured such as concrete). Therefore, when performing line-preserving smoothing filtering, the image is reduced to 0.1 mm/1 pixel (by simple 4-pixel x 4-pixel average calculation and thinning), and this image is Processing is in progress.

As a result, the size of one pixel is 0.025 mm (71
1/32 of 1 pixel 0.8 mm from the original image on N constant object)
A compressed image is generated (see FIGS. 6(a) and (b)).

Next, a crack pattern is extracted from the top layer image compressed to 1/32 as described above (step 130).
. In this process, as shown in Fig. 7, line structure extraction circular filtering is used to calculate the line structure (crack) feature mS at each pixel of the image, and the line structure (crack) feature mS is calculated based on the average feature amount S of the image. It is determined whether the interview is a crack candidate area (binary) based on the binarization threshold (step 132).

When a line structure extraction circular filter is applied to an image, large features ff1S are calculated where pixels are located on line structures. For the parameters of the circular filter necessary for this calculation, values obtained from the results of original research (target window size ws-i, processing radius R-5) are used.

In calculating the binarization threshold, the entire image is
As shown in the figure, each block is divided into 36 small blocks, and a binarization threshold (
The feature value S calculated for each pixel is averaged within the target project, and the value S is used as the binarization threshold, and the values above the average value S are considered line element candidate points, and the values below the average value are the background). do.

Then, in the binarization, the values with feature values larger than the binarization standard in each block are set as "1" as line element candidate points, and the other parts are set as "0" as the background, thereby identifying crack candidate areas. Extract and do. By dividing the image into small blocks and calculating the binarization criteria, it is possible to suppress erroneous determinations caused by fluctuations in the binarization criteria due to local dirt in the image.

Next, the crack candidate portion obtained by the line structure extraction circular filtering is converted into a core line using the Hi Iditch line thinning method, that is, the process of determining the center line of the crack candidate portion is performed (step 134).

After core line formation, the position coordinates on the image of the sequence of points constituting the center line are arranged in the order of joints, and the position coordinate value sequence of the cracked part, that is, the crack vector (see FIG. 18) is obtained (step 136). ).

In order to remove parts other than cracks from the crack candidate point sequence obtained in this way, removal of isolated points and isolated areas (removal of small areas), continuity check (removal of short crack candidate point sequence), whisker-like noise Removal (removal of noise generated during skeletonization (see FIG. 19)) and filling of loop regions are performed (step 138).

Note that the parameters used in programs such as noise removal processing are set to the values shown in FIG.

Next, the approximate crack width is further calculated using the straight line operator method for the approximate crack pattern extracted in the 1/32 compressed image as described above (step 140).
), in this process, the operator first specifies a crack measurement area (an arbitrary polygonal area) (step 141).
).

Then, the conversion coefficients from the photo coordinates to the actual structure coordinates, that is, the two-dimensional projection transformation coefficients (detailed crack extraction in the next step 150 and the crack total n1 data in a total of 7111 stages) are converted from the coordinate system on the image to the actual structure. The coefficients of eight elements for scale conversion to the physical coordinate system are calculated (step 143). Note that the reason for calculating the coefficients to be used in the next step 150 at this stage is that if the conversion coefficients are also determined when displaying the entire screen for checking extracted cracks, it will not be necessary to display them again in subsequent processing. This is because it was determined that the following process could proceed automatically without operator intervention. The projective transformation coefficient is calculated by solving simultaneous equations by correlating the coordinates of the four reference points (points whose coordinate values on the object surface are known) imprinted on the crack photograph with the coordinates on the image data. (See Figure 8).

Next, the rough crack extraction results are checked, and additional corrections to cracks that were not extracted and deletion of erroneously recognized cracks are manually performed (step 145). Then, the operator calculates the crack width using the linear operator method (see Figure 9) for each constituent point of the rough crack vector finally checked and confirmed by the operator (step 146), and calculates the roughly extracted crack position and The width is registered (step 148). Note that the approximate crack width and crack position are referred to as initial values in the next step of detailed crack extraction and measurement.

Next, detailed crack extraction and measurement processing (step 150)
Implement. First, the rough crack vector (center line position and width of the crack) obtained from the image at the top of the hierarchy compressed to 1/32 in the rough crack extraction and rough crack width measuring 71PJ processing described above is used as a guide map to calculate the rough crack width one step lower. The crack in the compressed image, that is, the 1/16 compressed image, is converted to 21iff, the center position is determined, and the width is increased to 8-1 in total (see FIG. 10). The crack search range is a band-shaped range around the crack pattern extracted from the upper image. In other words, the crack pattern (line) found in the upper image
A band-shaped search area of 2 to 3 times the crack width (calculated at the same time as the crack pattern) is set up on both sides of the area, and the local search area is sequentially binarized (local search area is an improved version of Otsu's binarization method). Repeated discrimination binarization method) is performed to distinguish cracks.

Thereafter, the center line position of the crack is determined in the differentiated crack portion, and the crack width (thickness of the binarized portion) is calculated.

Then, this process is performed from 1/32 compressed image to 1/16.1
/8. Detailed crack extraction and it i
Perform l#I. In this case, the total i1 at each stage? The determined crack position and width are used as a guide map in the image planning of the next layer, thereby making it possible to accurately measure the crack position and width.

Next, the smaller the size of the image to be processed, the shorter the time it takes to complete image processing, so in order to perform detailed crack measurement, it is important to make the processing unit as small as possible. Therefore, the images in each layer are managed by being divided into a number of small blocks as shown in FIG. As shown in FIG. 11, a small block to be processed is selected, that is, a subdivided image in which a crack exists is selected (step 16).
1). As a result, image processing of unnecessary portions is omitted, and the entire processing can be shortened.

In addition, image processing is performed only on local areas where cracks exist within the selected small blocks, thereby achieving even more efficient processing.

Next, the crack position is searched and binarized using a local and sequential iterative binarization method (step 163). This is done by selecting only a small local area near the crack in a small block selected based on the approximate crack vector measured in the upper image and binarizing it to obtain a crack candidate area (see Figure 11). . The small area near the crack is actually an elliptical small area along the crack, as shown in FIG.

When converting cracks into binary data, one frame of image data (5
In addition to the problem of not being able to determine a single binarization threshold for the entire area (0 cm x 50 cm to 100 cm x 100 marks), there is a problem inherent to Otsu's discriminant binarization method, namely the area to be binarized (cracks). When the area ratio of the surface and the background (concrete, etc.) is extremely biased,
Since a problem has been pointed out that the valuation threshold is not calculated, in this crack measurement system, the binarization area is narrowed down to a local elliptical small area centered on each component point of the rough crack vector. local value 2
In addition to selecting the digitized small area, the size of the small area is determined by referring to the roughly calculated crack width so that the area ratio between the binarized part and the background is not biased and is appropriate (
(See Figure 10).

In addition, when binarizing each small area, in order to reduce the influence of noise, especially when binarizing narrow cracks, [
A repeated binarization process is adopted in which the cycle of "binarization → expansion - noise removal - determination of next binarization area" is repeated to determine cracked areas while removing noise (13th
(see figure). Note that the method for determining the end of repetition is to compare the previous binarization area with the binarization area of that time, and add 90% to the binarization area.
This is when a change of % or more is no longer observed.

Among the above-mentioned cycles, the appropriate number of times of expansion treatment also differs depending on the crack width. In other words, it is necessary to increase the number of expansions for thick cracks, although the treatment time will be longer, and to reduce the number of expansions for thin cracks, in order to avoid bias in the area ratio between the cracked part and the background (see Figure 13). Furthermore, it has been confirmed that when the width becomes thicker than a certain level, a good binary image can be obtained without repeatedly performing the binarization processing described above.

As described above, by limiting the number of repetitions and the number of expansions using the total crack width determined in the upper image, cracks can be binarized with high accuracy using the total Δ-1 method that matches the crack width. can. Further, FIG. 21 shows the correspondence between each parameter used during binarization and the crack width measured in the upper image.

Next, isolated noise is removed (step 165) and skeletonized (step 167) for the binarized crack candidate region.

First, among the holes created inside the crack candidate part, holes with a small area are filled as shown in FIG. 14. This is to prevent the crack width from being measured too small when measuring the crack width.

Then, by moving the approximate crack vector to the center of the crack candidate part, core line processing is performed to gradually bring the center line position of the crack closer to the true center position as shown in FIG. During core line conversion, isolated areas (noise) that remain even after binarization are removed at the same time. This method focuses on the connectivity of cracks and determines the destination of the vector based on the principle of majority voting when there are multiple candidates for the destination of the crack vector due to isolated area noise. This is a method to prevent this from happening (see Figure 16).

To explain this skeletonization and isolated noise removal in more detail, first, the binarized crack candidate portions are numbered for each region based on the crack vector extracted from the image at the upper level of the hierarchy.

Next, at each constituent point of the approximate crack vector, a point where a straight line perpendicular to the crack direction intersects with the boundary of the crack candidate area is determined. The midpoint of this intersection is set as a new rough crack vector (see Figure 15). By doing so, - the center line position of the crack can be found in the image at the lower level, and - a rough crack vector with improved step accuracy can be obtained. .

If there are multiple destination candidates for a rough crack vector due to isolated noise, etc., focus on the connectivity of the cracks, aggregate the destination numbers of the constituent points of a series of crack vectors, and select the most frequent destination number by majority vote. The area where the crack vector is moved is determined as the destination of the crack vector (see FIG. 16). Then, for the constituent points of the crack vector that have not moved to the area determined by majority vote, the area with that number is searched again and the center is moved to prevent the center line from moving to the isolated area (16th (See figure) Next, at the crack center position, add a total of 1TP1 to the crack width (
Step 169).

As shown in FIG. 17, the crack width is measured by setting a circle inscribed in the crack part centered on the center line of the crack and calculating the diameter of the inscribed circle. At this time, as shown in Figure 1.4, in order to avoid estimating the crack width too small due to the shape of the crack candidate, the center position of the inscribed circle is moved in a direction perpendicular to the crack direction and the circle is gradually expanded. Find the inscribed circle with the largest value, and use the diameter of that circle as the crack width.

Next, the crack measurement data is scale-converted (step 171).

Once the original image at the bottom of the hierarchy has been measured, the crack 1llllJ constant value (the position and width have been expressed in the coordinate system on the image up to this stage) is expressed in the coordinate system on the surface of the object such as concrete ( Scale conversion is performed to represent the actual size. The scale conversion is performed by the two-dimensional projective conversion shown in FIG. 8, and the projective conversion coefficients obtained at the final stage of rough crack extraction are used here to convert to actual dimensions.

When the detailed crack extraction and measurement processing is completed as described above, the measured crack data is stored from the image processing device 5 to the external storage device (magnetic disk) of the output device 7, and the crack data is analyzed. A crack diagram and a crack characteristic diagram are created. The output format of the crack data is shown in Figure 22. In this table, the crack position data is represented by three-dimensional coordinate values.

〔Effect of the invention〕

As described above, according to the present invention, a captured image of the surface of an object is compressed hierarchically, and crack candidate extraction processing using line extraction circular filtering on the compressed image;
The crack position and width are roughly calculated using the straight line operator method from the crack extracted after binarization, core line conversion, vectorization, and noise separation processing, and this rough crack position and width is used as the first guide map. The crack positions and widths calculated using compressed images with a higher compression ratio in the reverse order of the order of compression in the hierarchical compression process are used as a guide map, and the crack positions and widths of images with a compression ratio one step smaller than those of the compressed images using this guide map are used as a guide map. Since the process of binarization, noise removal, core line generation, and crack width measurement are repeated step by step until the original image is reached, the image compression allows the cracks to have only linear characteristics regardless of the width, resulting in a line-extracted image. In addition to being able to apply processing algorithms, it is possible to reduce the risk of misrecognizing local linear noise as cracks by extracting crack patterns based on the macro field of view. In addition, since the extracted rough crack pattern is developed step by step on images lower in the hierarchy and the total center line position and width of the crack is calculated, the crack position can be adjusted while maintaining the connectivity of the crack. The total n1 accuracy of accuracy and crack width can be improved, and since the crack patterns extracted in the upper layer are used as guide maps to sequentially analyze images in the lower layers, micro images without cracks can be removed from the processing target.
Processing efficiency can be improved, and the measured values can be gradually refined by the stepwise needle n1, allowing highly accurate measurements to be performed.

[Brief explanation of the drawing]

FIG. 1 is a schematic configuration diagram of a crack measurement system according to an embodiment of the present invention, FIG. 2 is a diagram showing a schematic flow of crack extraction measurement image processing in the crack measurement system shown in FIG. Figure 3 is a flow chart showing more detailed measurement processing of the crack measurement system in Figure 1, Figure 4 is an explanatory diagram showing how an image reading device reads a crack photograph, and Figure 5 shows shading correction and FIG. 6(a) is an explanatory diagram showing a method of sensor noise correction. (b) and (C) are respectively an explanatory diagram of line-preserving smoothing filtering, an explanatory diagram of thinning by maximum value selection method, and an explanatory diagram of image layering and compression method. Figure 7 is a binary value in line extraction circular filtering. Fig. 8 is an explanatory diagram of coordinate conversion from photo image coordinates to coordinates on the actual structure, Fig. 9 is an explanatory diagram of the straight line operator method, and Fig. 10 is detailed crack extraction. Figure 11 is an explanatory diagram showing the relationship between crack data extracted at the top of the hierarchy and lower layer image data selected based on this data. Figure 12 is local iterative sequential determination 2 using approximate crack vector constituent points. An explanatory diagram of the digitization method, Fig. 13 is an explanatory diagram of repeated binarization and expansion of a cracked part, and Fig. 14 is an explanation of moving the measurement position, inflating the inscribed circle, and removing the influence of noise by filling in the holes. Figure 15 is an explanatory diagram of detailed crack core line formation by moving rough crack vector constituent points,
Fig. 16 is an explanatory diagram of core lineization of cracks using ignoring isolated points based on majority voting, Fig. 17 is an explanatory diagram showing the definition of crack width, Fig. 18 is an explanatory diagram of thinning and vectorization, and Fig. 19 is an explanatory diagram showing the occurrence of whisker-like noise due to line thinning, Fig. 20 is a diagram showing the image processing method and main parameters adopted for crack image processing, and Fig. 21 is a diagram showing the image processing method and main parameters adopted for crack image processing. FIG. 22 is a diagram showing the format of the crack data, and FIG. 23 is a diagram showing the size of the unit processed image at each stage of the stepwise crack measuring method. 1. Φ film, 3. Image reading device, 5. Image processing device, 7. Output device.

Claims (5)

[Claims] (1) A crack measurement system that measures cracks that occur on the surface of an object by image processing, and compresses images of the surface of the object hierarchically to a predetermined compression ratio while applying line-preserving smoothing filtering. a hierarchical compression means;
The image compressed to a predetermined compression ratio by the hierarchical compression means is subjected to crack candidate extraction processing using line extraction circular filtering, binarization, skeletonization, vectorization, and noise separation processing to extract cracks. an extraction means for extracting and calculating the position; an approximate crack width calculation means for calculating the crack width by a straight line operator method from the crack extracted by the extraction means; and calculation by the approximate crack extraction means and the approximate crack width calculation means. The crack positions and widths calculated using the compressed images are used as the first guide map, and thereafter, the crack positions and widths calculated using compressed images with larger compression ratios are used as guide maps in the reverse order of the order compressed by the hierarchical compression method, and the compression ratios are calculated from the compressed images. Detailed crack extraction measurement means that performs stepwise binarization, noise removal, core line formation, and crack width measurement for cracked parts of an image that is one step smaller until the original image is reached. system. (2) The hierarchical compression means hierarchically compresses the image taken of the surface of the object to a predetermined compression ratio, converts thick cracks into thin line structures, and saves thin cracks as they are as thin line structures. A claim characterized by comprising a compression means and a line-preserving smoothing filtering means for performing line-preserving smoothing filtering on each point constituting the image compressed by the compression means to reduce noise other than cracks. (1)
Described crack measurement system. (3) The binarization in the detailed crack extraction measurement means is local binarization in which the binarization area is narrowed down to local elliptical small areas centered on each component point of the rough crack vector and binarized. a small area selecting means; a small area determining means for determining the small area by referring to a crack width that is roughly calculated so as not to cause bias in the area ratio between the binarized portion and the background; Iterative binarization means repeats binarization, expansion, noise removal, and previously binarized area determination processing for narrow cracks to reduce the influence of noise, and the number of repetitions in the iterative binarization means; 2. The crack measurement system according to claim 1, further comprising a number-of-times limiting means for limiting the number of times of expansion. (4) Noise removal and skeletonization in the detailed crack extraction measurement means are performed by numbering each region of the binarized crack candidate portions based on the crack vector extracted from the upper layer image. candidate area numbering means; center position searching means for determining the point where a straight line perpendicular to the crack direction and the boundary of the crack candidate area intersect at each point of the rough crack vector; and totaling destination numbers of a series of crack points; A means for determining the movement result by majority vote to move to the area having the same number as the area where the most points moved by majority vote, and for crack points that have not moved to the area determined by majority vote, search again for the area having that number. The center of the crack is determined based on the rough crack vector, and the center position of the crack is determined based on the approximate crack vector. 2. The crack measurement system according to claim 1, wherein the crack measurement system identifies the portion based on a majority vote. (5) The measurement of the crack width by the detailed crack extraction measurement means includes an inscribed circle setting means for setting a circle inscribed in the crack part centered on the position of the core-lined crack center line; a maximum inscribed circle search means for searching for an inscribed circle with a maximum diameter by gradually expanding the circle while shaking the center position of the inscribed circle in a direction perpendicular to the crack direction so as to remove the influence; 2. The crack measuring system according to claim 1, further comprising a hole filling means for filling the missing portions left behind in an island shape before determining the inscribed circle.