Hole Drilling Position Detection Method

Abstract

To allow for improving position detection efficiency of a plurality of holes formed in a wall surface of a concrete structure.SOLUTION: A plurality of original images including partial holes Hs from a captured image of a drilled wall surface, and a plurality of teacher images obtained by extracting outer peripheral edges of the partial holes Hs from the plurality of original images are generated, and an image recognition AI model for specifying hole outer peripheral edges is created by learning using the plurality of teacher images, and an image of a wall surface in which a hole H to be detected is formed is inputted to the AI model, and an outer peripheral edge of the hole H to be detected is specified by the AI model to detect a position of the hole H to be detected.SELECTED DRAWING: Figure 4

Description

Technical Field

[0001] The present invention relates to a method for detecting a drilling position, and more particularly to a technique for detecting the position of a hole formed in a concrete structure, for example.

Background Art

[0002] As reinforcement work for concrete structures, there are construction methods such as thickening methods for walls and post-construction shear methods. In these construction methods, post-construction anchor bars and post-construction shear bars are used to enhance the integrity with existing concrete structures, and insertion holes are formed in the concrete structures to insert them. However, since the number of these insertion holes is extremely large, exceeding several thousand, it is required to improve the efficiency of detecting the drilled shape (drilling position).

[0003] For example, Patent Document 1 discloses a technique for photographing a drilled wall surface and obtaining the hole position from the brightness difference between the hole and the wall surface. Further, Patent Document 2 discloses a technique for obtaining the hole position from the surface shape obtained by scanning the drilled wall surface with laser light.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Patent Document 2

Summary of the Invention

Problems to be Solved by the Invention

[0005] By the way, in order to improve the efficiency of detecting the drilled hole shape, while working on the development of a method for detecting the drilled hole position from an image of a wall surface (hereinafter referred to as a drilled hole wall surface) of a concrete structure in which a plurality of holes are formed, by performing SfM (Structure from Motion) processing on a plurality of images obtained by photographing the drilled hole wall surface, 3D point cloud data of the drilled hole wall surface is acquired, points in a range deeper by a predetermined depth than the drilled hole wall surface are extracted, and a method is developed to identify the location by regarding the hole as a circle for the extracted point cloud.

[0006] However, in the above method, it takes time to create point cloud data from a plurality of wall surface images, and since it is necessary to process a large amount of data, it takes time for data processing, so there is a problem that the detection efficiency of the drilled hole shape (drilled hole position) is significantly reduced.

[0007] The present invention has been made based on the above technical background, and an object thereof is to provide a technology capable of improving the detection efficiency of the positions of a plurality of holes formed in the wall surface of a concrete structure.

Means for Solving the Problems

[0008] In order to solve the above problems, the drilled hole position detection method of the present invention according to claim 1 cuts out a plurality of original images including some holes from an image of a wall surface in which a plurality of holes are formed, creates a plurality of teacher images in which the outer peripheral edges of the some holes are extracted from the plurality of original images, creates an image recognition AI model for identifying the outer peripheral edge of the hole by learning using the plurality of teacher images, inputs an image of the wall surface in which the hole to be detected is formed into the image recognition AI model for identifying the outer peripheral edge of the hole, and the outer peripheral edge of the hole to be detected is identified by the image recognition AI model for identifying the outer peripheral edge of the hole, and the position of the hole is detected.

[0009] The drilled hole position detection method of the present invention according to claim 2 is characterized in that, in the invention according to claim 1, the number of pixels of the teacher image is less than the number of pixels of the original image.

[0010] The method for detecting the drilling position of the present invention according to claim 3 is characterized in that, in the invention according to claim 1 or 2, the number of the partial holes is one or two.

Advantages of the Invention

[0011] According to the present invention, it becomes possible to improve the detection efficiency of the positions of a plurality of holes formed in the wall surface of a concrete structure.

Brief Description of the Drawings

[0012]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Embodiments for Carrying Out the Invention

[0013] Hereinafter, embodiments as an example of the present invention will be described in detail with reference to the drawings. In the drawings for explaining the embodiments, the same components are basically denoted by the same reference numerals, and the repeated explanations thereof are omitted.

[0014] FIG. 1 is a longitudinal sectional view of a seismic reinforcement part formed in a concrete structure, and FIG. 2 is a longitudinal sectional view of a main part of a hole formed in a concrete structure to form the seismic reinforcement part.

[0015] As shown in FIGS. 1 and 2, the post-construction shear reinforcement method is one of the reinforcement (seismic reinforcement) methods for an existing concrete structure S. For example, for an existing concrete structure S in contact with the ground G, or an existing concrete structure S constructed on the ground close to structures such as railways and roads, after inserting a post-construction shear reinforcement bar (hereinafter referred to as "shear reinforcement bar") R into the inside of a hole H formed in the structure, a fixing material (mortar) M is filled and injected and cured to integrate the shear reinforcement bar R with the structure S to improve the shear resistance of the structure S. As the shear reinforcement bar R, for example, a bar generally used with one side threaded, diagonally cut, and a hexagonal nut R2 attached as a fixing body at the tip is used.

[0016] By the way, in the post-construction shear reinforcement method, in order to confirm whether a plurality of drilled holes H are drilled at an appropriate position and with an appropriate size, it is necessary to detect the drilled shape (hole position and radius) of the plurality of holes H.

[0017] As a method thereof, there is a drilled hole position detection method in which the drilled hole wall surface is photographed and the hole position is obtained from the brightness difference between the hole and the wall surface. In this method, however, when the brightness difference between the hole and the wall surface becomes unclear during photographing, it becomes difficult to specify the drilled hole position.

[0018] Also, there is a drilled hole position detection method in which the drilled hole wall surface is scanned with a laser beam and the hole position is obtained from the surface shape obtained. In this method, however, since the tip of the drilling device shakes at the start of drilling or the concrete at the peripheral part of the hole H is damaged (see FIG. 2), the detected hole diameter may be larger than the actual hole diameter (inner hole diameter).

[0019] Taking these problems into consideration, the inventors obtained 3D point cloud data of the drilled hole wall surface by performing SfM processing on a plurality of images obtained by photographing the drilled hole wall surface, extracted points in a range deeper than the wall surface by a predetermined depth, and developed a method for identifying the location by regarding the hole as a circle for the extracted point cloud. However, in that case, it takes time to create point cloud data from a plurality of wall surface images, and since it is necessary to process a large amount of data, it takes time for data processing, resulting in a significant decrease in the detection efficiency of the drilled hole shape (hole position and radius).

[0020] Therefore, in the present embodiment, the drilled hole shape (hole position and radius) is specified based on the contour of the hole extracted by an image recognition AI (Artificial Intelligence) model for specifying the outer periphery of the hole (hereinafter simply referred to as the AI model) learned by the teacher image. That is, a plurality of original images including some holes are cut out from the photographed image of the drilled hole wall surface, a plurality of teacher images are created by extracting the outer peripheries of some holes from the plurality of original images, the AI model is learned using the plurality of teacher images, an image of the wall surface on which the hole H to be detected is formed is input to the AI model, the outer periphery of the hole H to be detected is specified by the AI model, and the position of the hole H is detected. Hereinafter, an example of the drilled hole position detection method of the present embodiment will be described with reference to FIGS. 3 to 7.

[0021] FIG. 3 is a partial plan view of an example of an image of the drilled hole wall surface of a test structure photographed by a digital camera or the like. The test structure Ss is made of, for example, concrete, and a plurality of holes Hs are drilled in its wall surface. The diameter of the holes Hs is, for example, about 20 to 40 mm, and the depth of the holes Hs is, for example, about 500 mm. However, in the present invention, the diameter and depth of the holes Hs are not limited to the above values and are set to necessary values according to the state of the structure to be detected for the drilled hole position.

[0022] First, prior to detecting the drilling hole position, an AI model is trained using an image of the reference hole Hs. As input images for training, from among the images obtained by photographing the drilling hole wall surface of the test structure Ss, for example, a plurality of original images with a size of 500×500 pixels are cut out so that one or two holes Hs are included, and further, a plurality of teacher images obtained by resizing (reducing) the original images to a size of, for example, 125×125 pixels are used. However, in the present invention, the various image sizes are not limited to the values described above and can be changed variously.

[0023] FIG. 4(a) is a plan view of an example of an original image obtained by cutting out some holes from an image of the drilling hole wall surface of a test structure, and FIG. 4(b) is a plan view of an example of a teacher image obtained by reducing the size of the original image in FIG. 4(a).

[0024] As shown in FIG. 4(a), the original image contains, for example, two holes Hs. By setting the size of the original image to an extent that one or two holes Hs are included in this way, the data volume can be reduced, so the time required for data processing can be shortened, and the efficiency of data processing can be improved.

[0025] Also, as shown in FIG. 4(b), as the teacher image, an image in which a circle (the ring-shaped white line in FIG. 4(b)) is drawn according to the contour (outer peripheral edge) of the hole Hs in FIG. 4(a) is created using image editing software. As the teacher image, for example, an image created using an image obtained by rotating the original image or inverting it vertically, horizontally, or diagonally may be used. Thereby, the labor of photographing can be reduced, and the reliability of the AI model can be improved.

[0026] Here, when creating the teacher image from the original image, the image size is reduced within a range that does not affect the discrimination of the hole Hs in consideration of the reliability of the AI model. Thereby, when training the AI model, the data volume can be reduced, so the time required for data processing can be shortened, and the efficiency of data processing can be improved.

[0027] Also, here, in order to ensure the reliability of the AI model, for example, 400 teacher images were used. However, the number of teacher images is not limited to 400. For example, if the number is increased beyond 400, the reliability of the AI model can be improved.

[0028] Figure 5 is an explanatory diagram showing the schematic structure of the AI model.

[0029] The FCN (Fully Convolutional Network), which is often used in semantic segmentation (estimating which class for each pixel), was used for the AI model. In semantic segmentation, models such as Vgg16 and U-Net have been developed. However, in this study, on the premise of dividing the target image and reducing the image size for processing, first, as shown in Figure 5, a simple model was tried. Considering the size of the input image used for learning, it is a model composed of about 40 convolutional layers L1 to L40 and having no fully connected layer. Batch Normalization was introduced into the convolutional layers L1 to L40.

[0030] Using the AI model learned as described above, as an inference image (drilled hole wall image), similar to the teacher image, for example, a captured image of 500×500 pixels was cut out and an image resized to 125×125 pixels was used for inference. As a result, holes could be detected well.

[0031] Note that the inference image was captured with, for example, a digital camera. The learning image and the inference image may be captured using the same digital camera or different digital cameras.

[0032] In addition, for the inference after the learning of the AI model, for example, 50 images different from the images used for the learning of the AI model were used. However, here, for the sake of simplicity of explanation, the number of images used for inference was set to 50, but it is not limited to this, and for example, the number can be determined from thousands of images required at each site. Also, the drilled hole wall surface image of the position detection target may be used as the input image for learning.

[0033] FIG. 6 is a plan view of an example of an image of a hole showing the inference result.

[0034] The left column of FIG. 6 shows the input images, and images of holes H with different numbers and arrangements are illustrated. The image of the hole H in the second row illustrates a case where the distance between the photographing position and the drilled hole wall surface is different from others.

[0035] The central column of FIG. 6 shows the detection images (detection results) by the AI model, and the right column of FIG. 6 shows the superimposed images of the input images and the detection images. It can be seen that the outer periphery of the hole H in the input image coincides with the circular contour of the detection image. For images other than this example, there was no case where the hole H could not be detected at all, and the hole H could be detected fairly well.

[0036] The coincidence rate was calculated as an evaluation of the detection result. The coincidence rate obtained here means that for the images used for inference, in the same way as the creation of the teacher image, an image in which the contour of the hole was detected as a circle in advance with image editing software was created, and for the pixels constituting the circle of this image, if the value of the pixel at the same position in the detection image was other than "0 (zero)", it was regarded as "correct". Then, the total number of pixels regarded as "correct" was divided by the total number of pixels constituting the circle in the image in which the hole was detected as a circle to obtain the coincidence rate. The coincidence rate for the 50 images used for inference was 0.53 to 0.97, and the average value was as high as 0.82, indicating high accuracy.

[0037] As a method for detecting the drilling position, the contour image of the hole detected by the deep AI model is approximated by a circle by the least squares method, and the position of the hole is obtained by calculating the center position of the approximated circle.

[0038] Here, when looking at the detected image, there are cases where, in the input image, parts with unclear contrast between the boundary of the hole and the surface of the concrete structure result in a detected image where part of the circle is unclear or data is missing.

[0039] Fig. 7(a) is a diagram showing a hole image with a part missing, and Fig. 7(b) is a diagram showing the result of calculating the center and radius of the circle by the least squares method from the hole image with a part missing in Fig. 7(a).

[0040] Using such a hole image with a part missing, as shown in the following formula (1), the center and radius of the circle were calculated by the least squares method. As a result, as shown in Fig. 7(a), even if it could not be detected as a complete circle, as shown in Fig. 7(b), the center position and radius of the circle (i.e., the hole) could be calculated by the least squares method.

[0041] (xi - a) 2 +(yi - b) 2 = r 2 ····Formula (1)

[0042] Here, xi and yi are the values of the pixel positions on the X-axis and Y-axis of any pixel constituting the detected hole, a and b are the values of the pixel positions on the X-axis and Y-axis of the center of the circle, and r is the radius of the circle.

[0043] Thus, according to this embodiment, by detecting the drilled shape (hole position and radius) of the drilled wall surface of the object to be position-detected by the AI model, the work at the time of position detection can be simplified, so that the position detection efficiency of a plurality of holes on the drilled wall surface of the object to be position-detected can be improved.

[0044] In addition, by making the size of the original image used for learning the AI model such that it contains one or two holes, the data capacity can be reduced during the learning of the AI model, so that the time required for data processing can be shortened and the efficiency of data processing can be improved. For this reason, the position detection efficiency of a plurality of holes on the drilled wall surface of the object to be position-detected can be improved.

[0045] Furthermore, by making the size of the teacher image used for training the AI model smaller than the size of the original image, the data capacity can be reduced during the training of the AI model, so the time required for data processing can be shortened, and the efficiency of data processing can be further improved. Therefore, the position detection efficiency of a plurality of holes on the perforated wall surface of the object to be position-detected can be further improved.

[0046] The invention made by the present inventor has been specifically described based on the embodiments. However, the embodiments disclosed in this specification are illustrative in all respects and are not limited to the disclosed technology. That is, the technical scope of the present invention should not be construed restrictively based on the description in the above embodiments, but should be construed according to the description in the claims. All changes that are equivalent to the technology described in the claims and do not deviate from the gist of the claims are included.

[0047] For example, when detecting the positions of the holes on the perforated wall surface, the image of the perforated wall surface of the object to be perforated position-detected may be segmented, or physical distance (the distance between the camera for shooting and the perforated wall surface) information may be given to the image.

Industrial Applicability

[0048] In the above description, the case where the present invention is applied to the position detection of holes formed for inserting shear reinforcement bars into a concrete structure has been shown. However, the present invention is not limited to the holes for inserting shear reinforcement bars, and can be widely applied to the position detection of holes formed in concrete structures for various purposes.

Explanation of Reference Numerals

[0049] S Structure Ss Specimen Structure H Hole Hs Hole R Post-construction Shear Reinforcement Bar R1 Reinforcement Bar R2 Hexagonal Nut M Fastening Material Ground G

Claims

1. Cut out a plurality of original images including some holes from an image of a wall surface with a plurality of holes formed therein, create a plurality of teacher images by extracting the outer peripheral edges of the some holes from the plurality of original images, create an image recognition AI model for identifying the outer peripheral edges of holes by training using the plurality of teacher images, input an image of a wall surface with a hole to be detected into the image recognition AI model for identifying the outer peripheral edges of holes, identify the outer peripheral edge of the hole to be detected by the image recognition AI model for identifying the outer peripheral edges of holes, and detect the position of the hole. A hole drilling position detection method characterized by the above.

2. The number of pixels of the teacher image is less than the number of pixels of the original image. The hole drilling position detection method according to claim 1, characterized by the above.

3. The number of the some holes is one or two. The hole drilling position detection method according to claim 1 or 2, characterized by the above.

Images