Apparatus for evaluating concrete construction joints, evaluation method, and evaluation program

Abstract

This invention provides a concrete construction joint evaluation device, evaluation method, and evaluation program that can automatically determine whether additional green cuts are required on the construction joint surface. [Solution] The evaluation device includes an image acquisition unit that acquires images of concrete joint surfaces that have been treated with green cut; a joint surface information acquisition unit that inputs the images into a machine learning-based learning model and acquires joint surface information from the learning model that indicates the state of the joint surface in the images; and a green cut necessity determination unit that determines whether further green cuts are necessary on the joint surface based on the joint surface information.

Description

Technical Field

[0001] The present disclosure relates to an evaluation device, an evaluation method, and an evaluation program for determining whether an additional green cut is required on a concrete joint surface.

Background Art

[0002] Conventionally, when constructing concrete structures such as dam embankments, concrete joints are made. Jointing is a construction method in which the previously placed concrete is hardened and a process of placing new concrete on the joint surface is repeated.

[0003] In order to strengthen the joint part between two concretes placed with a time gap, it is necessary to remove the uncoagulated concrete formed on the placement surface of the previously placed concrete before placing the new concrete, and this removal work is called green cut. In Patent Document 1, green cut is executed by a green cut machine that automatically travels on the joint surface.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] The better the finish of the green cut on the joint surface, the stronger the joint part is joined, so the strength of the overlaid concrete is improved. Generally, it is said that green cut is preferably carried out to the extent that the coarse aggregate in the concrete is exposed on the joint surface.

[0006] The determination of whether the finish achieved through green cutting is satisfactory is made by construction site supervisors based on their experience. However, this determination relies on the supervisor's experience and intuition, which can lead to inconsistent results. If a supervisor judges a finish to be unsatisfactory when it is actually satisfactory, further green cutting may be required, increasing the workload at the construction site.

[0007] The purpose of this disclosure is to address the above-mentioned problems and to provide an evaluation device, evaluation method, and evaluation program for concrete construction joints that can automatically determine whether additional green cuts are necessary for the construction joint surface. [Means for solving the problem]

[0008] 1) An evaluation device for concrete construction joints according to at least one embodiment of the present disclosure is: An image acquisition unit that acquires images of the concrete joint surface with green cut applied, A joint surface information acquisition unit obtains joint surface information indicating the state of the joint surface in the captured image from the learning model by inputting the captured image into the learning model, A green cut necessity determination unit determines whether further green cuts are necessary on the joint surface based on the aforementioned joint surface information. It is equipped with.

[0009] According to the configuration described in 1) above, the green cut necessity determination unit determines whether further green cutting is necessary based on the construction joint information obtained by inputting the captured images into the learning model. Thus, a concrete construction joint surface evaluation device is realized that can automatically determine whether additional green cutting is necessary on the construction joint surface.

[0010] 2) In some embodiments, the concrete joint surface evaluation device described in 1) above, The learning model outputs the joint surface information for each of the multiple divided images obtained by dividing the captured image into multiple vertical and horizontal sections.

[0011] According to the configuration described in 2) above, the necessity of further green cutting is determined based on the joint surface information for each of the multiple segmented images. This allows for a more accurate determination of whether further green cutting is necessary.

[0012] 3) In some embodiments, the concrete joint surface evaluation device described in 2) above, The aforementioned joint surface information is, Multiple first divided joint surface information indicating the state of the joint surface, each of which is associated with a plurality of first divided images obtained by dividing the aforementioned captured image vertically and horizontally according to a first pattern, Multiple second-divided joint surface information indicating the state of the joint surface, each associated with a plurality of second-divided images obtained by dividing the captured image vertically and horizontally according to a second pattern different from the first pattern, Includes, The green cut necessity determination unit determines, based on the plurality of first divided joint surface information and the plurality of second divided joint surface information, whether it is necessary to perform the green cut on the joint surface shown in the captured image.

[0013] Construction joints that appear near the boundary between two adjacent first-segment images are separated into the two first-segment images and processed together with construction joints that appear in other areas of the first-segment images. Therefore, it can be difficult to accurately determine the state of construction joints that appear near the boundary of the first-segment images based on the first-segment images themselves. In this respect, with the configuration described in 3) above, the construction joints that appear near the boundary are contained within one of the second-segment images, so their state can be accurately evaluated. Since the state of construction joints that appear in the captured image can be evaluated more accurately, it is possible to more accurately determine whether further green cutting is necessary.

[0014] 4) In some embodiments, the concrete joint surface evaluation device described in 3) above, The second pattern is configured such that the intersections of four adjacent first division images do not overlap with the boundary lines of two adjacent second division images, either vertically or horizontally.

[0015] According to the configuration of 4) above, the joint surface near the intersection of four first divided images adjacent vertically and horizontally fits within one of the second divided images, so its state can be accurately evaluated. Therefore, it is possible to more accurately determine whether further green cutting is necessary.

[0016] 5) In some embodiments, it is an evaluation device for a concrete joint surface described in 3) or 4) above, The plurality of second divided images divide a part of the region of the captured image to be divided by the plurality of first divided images among the captured images.

[0017] According to the configuration of 5) above, the region in the captured image that is not the object to be divided by the first divided images is not included in any of the plurality of second divided images. Therefore, it is possible to suppress new noise that did not exist in the first divided images from entering the second divided images, which may be an obstacle when evaluating the state of the joint surface near the boundary of the plurality of first divided images. Thereby, the state of the joint surface near the boundary of the first divided images can be accurately evaluated based on the second divided images.

[0018] 6) In some embodiments, it is an evaluation device for a concrete joint surface described in any of 3) to 5) above, Define the vertical dimension and the horizontal dimension of the first divided image as X1 and Y1 respectively, Define the vertical dimension and the horizontal dimension of the second divided image as X2 and Y2 respectively, 0.95 ≦ X1 / X2 ≦ 1.05 ··· Formula (1) 0.95 ≦ Y1 / Y2 ≦ 1.05 ··· Formula (2) Both hold.

[0019] According to the configuration of 6) above, the sizes of the first divided image and the second divided image become substantially the same. Since the states of the joint surfaces shown in both divided images are equally treated and the captured image is input into the learning model, the joint surface information can more accurately represent the state of the joint surface shown in the captured image. Therefore, it is possible to more accurately determine whether further green cutting is necessary. s

[0020] 7) In some embodiments, there is an evaluation apparatus for a concrete joint surface according to any one of 2) to 6) above, where the joint surface information includes a plurality of divided joint surface information respectively associated with the plurality of divided images, the plurality of divided joint surface information includes error information indicating that the joint surface in the divided image is covered with water or foreign matter and is in a state where it is impossible to determine the quality of the green cut, the green cut necessity determination unit determines whether it is necessary to perform a further green cut based on the remaining divided images obtained by excluding the divided images associated with the error information from the plurality of divided images.

[0021] According to the configuration of 7) above, even when the joint surface is covered with water or foreign matter in a part of the captured image, it is possible to determine whether a further green cut is necessary based on the state of the joint surface shown in the other part of the captured image.

[0022] 8) In some embodiments, there is an evaluation apparatus for a concrete joint surface according to 7) above, the plurality of divided joint surface information includes determinable information indicating that the joint surface in the divided image is visible and is in a state where it is possible to determine the quality of the green cut, the plurality of divided images include determinable divided images associated with the determinable information, error divided images associated with the error information, and when an error image occupancy rate correlated with a value obtained by subtracting the number of determinable divided images from the number of error divided images exceeds a reference value, the evaluation apparatus further includes an error determination unit that determines that the joint surface shown in the captured image is covered with the water or the foreign matter, and when it is determined that the joint surface shown in the captured image is covered with the water or the foreign matter, the green cut necessity determination unit does not determine whether a further green cut is necessary.

[0023] According to the configuration described in 8) above, for captured images where the error image occupancy rate exceeds a standard value, the system does not determine whether further green cutting is necessary for the joint surface. This avoids inaccurate determination of whether green cutting is necessary and allows for automatic determination of whether the joint surface in the captured image is covered with water or foreign matter.

[0024] 9) In some embodiments, the concrete joint surface evaluation device described in 8) above, The error image occupancy rate is obtained by subtracting from the number of error division images the number of divisible division images multiplied by a coefficient greater than 1.

[0025] After applying a green cut to the concrete joint surface, cleaning with water is performed, making it easy for water or foreign matter to remain on the joint surface. In this regard, according to the configuration of 9) above, the number of judging segmented images is substantially increased when calculating the error image occupancy rate, making it more difficult for the error judgment unit to issue an error judgment. As a result, even if some water or foreign matter is captured in the image, the necessity of further green cut can be determined based on the state of the joint surface captured in the judging segmented images. Because it is less likely to issue an error judgment and the necessity of green cut can be determined, the practicality of evaluation devices used in actual field work can be improved.

[0026] 10) In some embodiments, the concrete joint surface evaluation device described in 1) above, The aforementioned captured image is An error image region in which the joint surface is covered with water or foreign matter and is in a state where it is not visible, A determinable image region in which the aforementioned joint surface is exposed and visible, Includes, The joint surface information is either judgment error information indicating that the joint surface in the captured image is in a state where it is impossible to determine whether the green cut is good or bad, or quality information indicating whether the green cut is good or bad for the joint surface in the captured image. The learning model outputs either the judgment error information or the pass / fail information according to the ratio of the error image region to the judgeable image region in the captured image. The green cut necessity determination unit determines, based on the quality information, whether further green cuts are necessary.

[0027] According to the configuration described in 10) above, if the error image region occupies a relatively large proportion of the captured image, the learning model outputs error information. On the other hand, if the judgment image region occupies a relatively large proportion of the captured image, the learning model outputs pass / fail information. The pass / fail information corresponds to the quality of the green cut of the joint surface visible in the judgment image region. Therefore, even if water or foreign matter is visible in the captured image, it is possible to determine whether further green cut is necessary on the joint surface in the captured image.

[0028] 11) In some embodiments, the concrete joint surface evaluation device described in any of 1) to 10) above, The aforementioned learning model, Machine learning is performed using training data that links training images showing concrete joint surfaces with correct data on the condition of the joint surfaces in the training images.

[0029] According to the configuration described in 11) above, by applying supervised learning to the learning model, the learning model can output construction joint information that more accurately reflects the state of the construction joint as seen in the captured image.

[0030] 12) A method for evaluating concrete construction joints according to at least one embodiment of the present disclosure is: Image acquisition step to obtain a photographic image of the concrete joint surface with green cut applied, A joint surface information acquisition step involves inputting the captured image into a machine learning model, thereby obtaining joint surface information from the model that indicates the state of the joint surface in the captured image, and A green cut necessity determination step is to determine whether further green cuts are necessary on the joint surface based on the aforementioned joint surface information. It is equipped with.

[0031] According to the configuration in 12) above, the same technical advantages as in 1) above can be obtained.

[0032] 13) The concrete joint evaluation program according to at least one embodiment of the present disclosure is On the computer, Image acquisition step to obtain a photographic image of the concrete joint surface with green cut applied, A joint surface information acquisition step involves inputting the captured image into a machine learning model, thereby obtaining joint surface information from the model that indicates the state of the joint surface in the captured image, and A green cut necessity determination step is to determine whether further green cuts are necessary on the joint surface based on the aforementioned joint surface information. Make it run.

[0033] According to the configuration in 13) above, the same technical advantages as in 1) above can be obtained. [Effects of the Invention]

[0034] This disclosure provides an evaluation device, evaluation method, and evaluation program for concrete construction joints that can automatically determine whether additional green cuts are required on the construction joint surface. [Brief explanation of the drawing]

[0035] [Figure 1] This is a schematic diagram of a concrete structure. [Figure 2] This is a schematic diagram of the evaluation device according to the first embodiment. [Figure 3] This is a schematic diagram showing the data that is input and output in the learning model. [Figure 4] This figure shows an actual photograph of the joint surface. [Figure 5]This is a schematic diagram showing the details of the data input and output in the learning model. [Figure 6] This is a schematic diagram showing the segmentation pattern of the captured image. [Figure 7] This is a schematic diagram of an evaluation device with additional components. [Figure 8] This is a schematic diagram of the training data used in machine learning. [Figure 9] This diagram shows a classification of the condition of construction joints. [Figure 10] This flowchart shows the evaluation method for concrete construction joints. [Figure 11] This is a schematic diagram of the evaluation device according to the second embodiment. [Figure 12] This is a schematic diagram of the captured image. [Figure 13] This is a schematic diagram showing a comparative example of the division target regions of the first and second divided images. [Modes for carrying out the invention]

[0036] Hereinafter, several embodiments of this disclosure will be described with reference to the attached drawings. However, the dimensions, materials, shapes, relative arrangements, etc., of the components described or shown in the drawings as embodiments are not intended to limit the scope of this disclosure, but are merely illustrative examples. For example, expressions describing relative or absolute arrangements such as "in a certain direction," "along a certain direction," "parallel," "orthogonal," "center," "concentric," or "coaxial" should not only strictly describe such arrangements, but also describe states of relative displacement with tolerances or angles or distances that allow for the same function to be achieved. For example, expressions such as "identical," "equal," and "homogeneous" that describe things being in an equal state not only describe a state of being strictly equal, but also describe a state in which there is a tolerance or a difference that is sufficient to achieve the same function. For example, expressions describing shapes such as squares or cylinders shall not only represent geometrically precise shapes such as squares or cylinders, but also shapes that include protrusions, chamfers, etc., to the extent that the same effect can be achieved. On the other hand, expressions such as "to possess," "to include," or "to have" a single component are not exclusive expressions that exclude the existence of other components. Note that similar configurations may be denoted by the same reference numerals and their explanations may be omitted.

[0037] <Overview> Figure 1 is a schematic diagram showing a construction site for a concrete structure 1, such as a dam embankment. The concrete 3 that makes up the concrete structure 1 includes a horizontal construction joint surface 4 where green cuts have been performed. A supervisor 9 at the construction site determines whether further green cuts are necessary on the construction joint surface 4. If the finished state of the green cuts on the construction joint surface 4 is determined to meet the prescribed quality standards, further green cuts are deemed unnecessary. Conversely, if the finished state is determined not to meet the quality standards, further green cuts are performed on the construction joint surface 4.

[0038] Previously, there was a problem in that the judgments of the judges 9, who had to rely on their own experience and intuition, were inconsistent. The inventors of this invention believed that if this inconsistency could be suppressed, it would be possible to eliminate the additional green cuts that were being performed unnecessarily, thereby reducing the workload at construction sites.

[0039] Based on the above concept, the inventors of this invention have devised a concrete joint surface evaluation device 8 (hereinafter simply referred to as "evaluation device 8"). The evaluation device 8 is configured to determine whether additional green cuts are necessary for the joint surface 4. The evaluator 9 makes the final decision while referring to the evaluation result of the evaluation device 8, so it is possible to determine whether additional green cuts are necessary more accurately than before, thereby improving work efficiency at construction sites.

[0040] The evaluation device 8 may be a portable computer device such as a smartphone, tablet, or laptop computer, or it may be a stationary computer device such as a desktop computer. Below, the evaluation device 8A according to the first embodiment (Figures 2 to 10) and the evaluation device 8B according to the second embodiment (Figures 11 and 12) will be described in order.

[0041] <Evaluation device 8A(8) according to the first embodiment> As shown in Figure 2, the evaluation apparatus 8A according to the first embodiment includes a camera 7, a control unit 2, and a monitor 6. The camera 7 is configured to generate an image Im of the joint surface 4 as data and output it to the control unit 2. The control unit 2 consists of a processor and memory, etc. The monitor 6 is configured to display various information output from the control unit 2.

[0042] The control unit 2 functions as an image acquisition unit 11, a joint surface information acquisition unit 13, and a green cut necessity determination unit 15A (15) according to the first embodiment.

[0043] The joint surface information acquisition unit 13 is configured to acquire captured images Im from the camera 7. The joint surface information acquisition unit 13 is also configured to communicate with a server 50 located in a remote location away from the construction site. The server 50 stores a learning model 55A (55) according to the first embodiment, and the joint surface information acquisition unit 13 is configured to transmit the acquired captured images Im to the learning model 55A.

[0044] Figure 3 shows an overview of the data input and output in the learning model 55A (further details will be described later using Figure 5). The learning model 55A is configured to output construction joint information 20 indicating the state of the construction joint surface 4 in the captured image Im when it is input. As a more specific example, the learning model 55A is configured to output construction joint information 20 indicating the state of the construction joint surface 4 for each of the multiple divided images D obtained by dividing the captured image Im vertically and horizontally. The divided images D exemplified in Figure 3 consist of a total of 16 images arranged in 4 columns vertically and 4 columns horizontally.

[0045] The joint surface information 20 includes multiple divided joint surface information 25. Each of the multiple divided joint surface information 25 is associated with a multiple divided image D, and the state of the joint surface 4 in each divided image D is indicated by each divided joint surface information 25. The multiple divided joint surface information 25 may also be a heat map. In other words, the multiple divided joint surface information 25 may be information indicated by colors superimposed on the captured image Im.

[0046] The segmented joint surface information 25 is classified into "excessive" (where the joint surface 4 shown in segmented image D is excessively green-cut), "good" (where the joint surface 4 is well green-cut), "insufficient" (where the green-cut is insufficient), "puddle" (where the joint surface 4 is covered with water W (see Figure 4)), and "foreign object" (where the joint surface 4 is covered with foreign object F (see Figure 4)). Foreign object F is a watering hose used when cleaning the joint surface 4, or reinforcing bars protruding from the concrete 3.

[0047] Of the segmented joint surface information 25, those corresponding to "excessive," "good," and "insufficient" are treated as determinable information 27. Determinable information 27 indicates that the joint surface 4 in segmented image D is visible and that the quality of the green cut on this joint surface 4 can be determined. In contrast, of the segmented joint surface information 25, those corresponding to "puddle" and "foreign matter" are treated as error information 26. Error information 26 indicates that the joint surface 4 in segmented image D is not visible and that the quality of the green cut cannot be determined.

[0048] Figure 4 shows the actual images T1 to T5 of the captured image Im schematically shown in Figure 3. Image T1 shows a joint surface 4 with excessive green cut, image T2 shows a joint surface 4 with good green cut finish, image T3 shows a joint surface 4 with insufficient green cut, image T4 shows a joint surface 4 partially covered by a puddle W, and image T5 shows a joint surface 4 partially covered by foreign matter F such as a hose. Typically, a single captured image Im shown in Figure 3 contains a mixture of these joint surface 4 states corresponding to images T1 to T5. Therefore, it is preferable to determine the state of the joint surface 4 for each of the multiple divided images D, and then determine the overall state of the joint surface 4 captured in the captured image Im.

[0049] Hereinafter, among the multiple segmented images D (see Figure 3), those associated with the judging information 27 may be referred to as "judgable segmented image Dc," and those associated with the error segmented image De may be referred to as "error segmented image De."

[0050] Returning to Figure 2, the joint surface information acquisition unit 13 of the control unit 2 is configured to acquire joint surface information 20 output from the learning model 55A. Furthermore, the green cut necessity determination unit 15A(15) according to the first embodiment is configured to determine whether further green cutting is necessary on the joint surface 4 based on the joint surface information 20.

[0051] More specifically, the green cut necessity determination unit 15A may exclude the error division image De from the multiple division images D and determine whether further green cut is necessary based on the remaining determinable division image Dc. Several methods can be applied to this determination method.

[0052] As a first example of the method, the proportion of the segmented images Dc that constitute the captured image Im that are judgeable as "good" may be identified. The green cut necessity determination unit 15A may determine whether additional green cut is necessary by comparing the above proportion with a first threshold. More specifically, if the proportion of judgeable segmented images Dc that are judged as "good" exceeds the first threshold, it may be determined that additional green cut is "unnecessary".

[0053] As a method relating to the second example, the proportion of the determinable segmented images Dc that fall under the category of "insufficient" may be identified. The green cut necessity determination unit 15A may determine whether additional green cut is necessary by comparing the above proportion with a second threshold. More specifically, if the proportion of the determinable segmented images Dc that fall under the category of "insufficient" exceeds the second threshold, it may be determined that additional green cut is "necessary".

[0054] According to the above configuration, the green cut necessity determination unit 15A determines whether further green cutting is necessary based on the joint surface information 20 obtained by inputting the captured image Im into the learning model 55A. Thus, an evaluation device 8A is realized that can automatically determine whether additional green cutting is necessary for the joint surface 4. Furthermore, since the joint surface information 20 (more specifically, the information of multiple divided joint surfaces 25) for each of the multiple divided images D is input to the green cut necessity determination unit 15A, the green cut necessity determination unit 15A can make a more accurate determination.

[0055] Furthermore, the green cut necessity determination unit 15A excludes the error division image De from the multiple division images D and determines the necessity of green cut based on the remaining determinable division image Dc. With this configuration, even if the joint surface 4 is covered with water W or foreign matter F in a part of the captured image Im, the necessity of further green cut can be determined based on the state of the joint surface 4 as seen in other parts of the captured image Im.

[0056] Figure 5 is a schematic diagram showing the details of the data input and output in the learning model 55A. In the example shown in the figure, the first captured image Ima and the second captured image Imb are generated from the captured image Im acquired by the image acquisition unit 11 (see Figure 2). The first captured image Ima is the same data image as the captured image Im, and is composed of multiple first segmented images D1 obtained by dividing the captured image Im vertically and horizontally according to a first pattern. The second captured image Imb is a data image obtained by trimming a part of the captured image Im, and is composed of multiple second segmented images D2 obtained by dividing the trimmed region of the captured image Im according to a second pattern different from the first pattern.

[0057] The learning model 55A, illustrated in Figure 5, receives a first captured image Ima and a second captured image Imb as input, and outputs multiple segmented joint surface information 25 corresponding to each. More specifically, the multiple segmented joint surface information 25 includes first segmented joint surface information 21 indicating the state of the joint surface 4, each linked to a plurality of first segmented images D1, and second segmented joint surface information 22 indicating the state of the joint surface 4, each linked to a plurality of second segmented images D2.

[0058] The first split joint surface information 21 may be a heat map displayed overlaid on the first captured image Ima. Similarly, the second captured image Imb may be a heat map displayed overlaid on the second captured image Imb. The green cut necessity determination unit 15A determines whether further green cutting is necessary based on the multiple first split joint surface information 21 and the multiple second split joint surface information 22 (details will be described later).

[0059] Referring to Figure 6, the relationship between the first and second patterns for dividing the captured image Im will be explained. The multiple first captured images Ima that make up the first captured image Ima have multiple intersection points P of four vertically and horizontally adjacent first divided images D1 (in other words, four first divided images D1 forming two vertical and two horizontal rows). Coordinates corresponding to multiple intersection points P also exist in the second captured image Imb (for the sake of explanation, these coordinates are also illustrated as intersection points P in the second captured image Imb).

[0060] The second pattern for dividing the second captured image Imb is configured such that the boundary line L of two vertically or horizontally adjacent second divided images D2 does not overlap with multiple intersection points P. More specifically, the boundary line L includes the horizontal boundary line Lh of two vertically adjacent second divided images D2 and the vertical boundary line Lv of two horizontally adjacent second divided images D2, and both the horizontal boundary line Lh and the vertical boundary line Lv avoid multiple intersection points P.

[0061] The multiple first-part images D1 that make up the first captured image Ima are all the same size. Similarly, the multiple second-part images D2 that make up the second captured image Imb are all the same size. If we define the vertical and horizontal dimensions of the first-part image D1 as X1 and Y1, respectively, and the vertical and horizontal dimensions of the second-part image D2 as X2 and Y2, respectively, then the following equations (1) and (2) hold true. 0.95≦X1 / X2≦1.05...Equation (1) 0.95≦Y1 / Y2≦1.05...Equation (2) Furthermore, in more detail, X1 and X2 have the same value, and Y1 and Y2 have the same value. The screen coordinates of the diagonal center of the second divided image D2 coincide with the screen coordinates of the intersection point P.

[0062] The technical advantages of outputting multiple first-division joint surface information 21 and multiple second-division joint surface information 22 from the learning model 55A are explained. As shown in Figure 6, joint surfaces 4 that appear near the boundary between two adjacent first-division images D1 (for example, joint surfaces 4 in the region indicated by the dashed line M) are divided into two first-division images D1, and the divided joint surfaces 4 are processed together with joint surfaces 4 that appear in other regions of each first-division image D1. Therefore, it can be difficult to accurately determine the state of joint surfaces 4 that appear near the boundary of the first-division image D1 based on the first-division image D1. In this respect, with the above configuration, the joint surfaces 4 that appear near the boundary are contained in one of the second-division images D2, so their state can be accurately evaluated. Since the state of joint surfaces 4 that appear in the captured image Im can be evaluated more accurately, it can be determined more accurately whether further green cut is necessary.

[0063] Furthermore, in the second pattern for dividing the second divided image D2, if the boundary line L of the second divided image D2 does not overlap with multiple intersection points P, then the construction joint surface 4 near the intersection points P will be contained within one of the second divided images D2, allowing for accurate evaluation of its state.

[0064] As previously mentioned, the first captured image Ima, which is the target of division in the first divided image D1, coincides with the captured image Im, and the second captured image Imb, which is the target of division in the second divided image D2, is a part of the captured image Im. In other words, the second divided image D2 divides a portion of the region of the captured image Im that is the target of division in multiple first divided images D1, both vertically and horizontally. That is, regions of the captured image Im that were not targeted for division by the first divided image D1 are not included in the region targeted for division in the second divided image D2.

[0065] For example, as shown in Figure 13 as a comparative example, if the region to be divided in the second divided image D2 includes multiple regions R that were not divided in the first divided image D1, then the construction joint surface 4 that was not visible in any of the first divided images D1 will be visible in the second divided image D2 along with the construction joint surface 4 near the intersection P mentioned above. This newly visible construction joint surface 4 may become noise that hinders the accurate evaluation of the properties of the construction joint surface 4 near the intersection P.

[0066] In this embodiment, regions in the captured image Im that were not divided by the first divided image D1 are not included in any of the multiple second divided images D2. Therefore, it is possible to suppress the intrusion of new noise into the second divided images D2 that was not present in the first divided image D1, which could be an obstacle when evaluating the state of the joint surface 4 near the boundaries of the multiple first divided images D1. As a result, the state of the joint surface 4 near the boundaries of the first divided images D1 can be accurately evaluated based on the second divided images D2.

[0067] Note that the first captured image Ima, before being divided by the first divided image D1, is the same image data as the captured image Im in this embodiment, as described above. However, in Figure 13, for the sake of clarity, the first captured image Ima is shown as a part of the captured image Im. Since the present invention includes both embodiments in which the first captured image Ima is the same image as the captured image Im, and embodiments in which the first captured image Ima is a part of the captured image Im (i.e., the embodiment in Figure 13), there is no problem in illustrating it as in Figure 13.

[0068] Furthermore, in a configuration where equations (1) and (2) hold, the sizes of the first divided image D1 and the second divided image D2 are substantially the same. Since the state of the joint surface 4 shown in both divided images is treated equally and the captured image Im is input to the learning model 55A, the multiple divided joint surface information 25 can more accurately represent the state of the joint surface 4 shown in the captured image Im. Therefore, it is possible to more accurately determine whether further green cutting is necessary.

[0069] Figure 7 is a schematic diagram of the evaluation device 8A with additional components. The control unit 2 of the evaluation device 8A may further include an error determination unit 17. The error determination unit 17 is used to determine whether the joint surface 4 in the captured image Im is covered with water W or foreign matter F to such an extent that the quality of the green cut cannot be determined. If a large portion of the joint surface 4 in the captured image Im is covered with water W or foreign matter F, the result of the determination based on the visible joint surface 4 may not accurately reflect the actual condition, so the error determination unit 17 is used.

[0070] The error determination unit 17 is configured to determine that the joint surface 4 is covered with water W or foreign matter F if the error image occupancy rate in the captured image Im (first captured image Ima or second captured image Imb) exceeds a standard value. Here, the error image occupancy rate is an index that indicates the extent to which an error segmented image De occupies multiple segmented images D.

[0071] As an example, if we take the total number of first-partitioned images D1 and multiple second-partitioned images D2, and let Me be the number of error-filled divisional images De and Mc be the number of identifiable divisional images Dc, then A, shown in equation (3) below, is the error image occupancy rate. Note that the sum of Me and Mc is equal to the total number of first-partitioned images D1 and second-partitioned images D2. A = Me - Mc ···(3)

[0072] For example, if the error image occupancy rate (A) exceeds a standard value (e.g., 0), the error determination unit 17 determines that the joint surface 4 in the captured image Im is covered with water W or foreign matter F. In this embodiment, if it is determined that the joint surface 4 in the captured image Im is covered with water W or foreign matter F, the green cut necessity determination unit 15A does not determine whether further green cut is necessary. With the above configuration, it is possible to avoid inaccurate determination of the necessity of green cut, and it is possible to automatically determine whether the joint surface 4 in the captured image Im is covered with water W or foreign matter F. Therefore, the judge 9 at the construction site can immediately decide to start work to remove the water W or foreign matter F from the joint surface 4.

[0073] The method for calculating the error image occupancy rate is not limited to equation (3). For example, the value obtained by dividing Me by Mc may be treated as the error image occupancy rate (A). Alternatively, equation (3A) may be applied instead of equation (3). A = Me - Mc × N ... Equation (3A) In equation (3A), N is a value greater than 1. For example, if Me=16, Dc=9, and N=2, then A=-2, which is less than or equal to the reference value (e.g., 0). In this case, the error determination unit 17 determines that the joint surface 4 captured in the image Im is not covered by water W or foreign matter F as a whole. At this time, the green cut necessity determination unit 15A may determine the state of the joint surface 4 by identifying the type with the highest occupancy among the determinable segmented images Dc. For example, if all of the determinable segmented images Dc (9 images) are determinable segmented images Dc associated with "insufficient", then it may be determined that the joint surface 4 is in a state of insufficient green cut.

[0074] The technical advantages of applying formula (3A) are explained below. After applying a green cut to the concrete 3 joint surface 4, cleaning with water is performed, making it easy for water W or foreign matter F to remain on the joint surface 4. In this regard, the above configuration substantially increases the number of judgeable segmented images Dc when calculating the error image occupancy rate (A), making it less likely for the error judgment unit 17 to issue an error judgment. As a result, even if some water W or foreign matter F is captured in the captured image Im, it is possible to determine whether further green cuts are necessary based on the state of the joint surface 4 captured in the judgeable segmented images Dc. Because it is less likely for errors to be detected and it is possible to determine whether green cuts are necessary, the practicality of the evaluation device 8A used in actual field work can be improved.

[0075] Referring to Figures 8 and 9, the machine learning of the learning model 55A will be explained. The machine learning applied to the learning model 55A may be unsupervised learning such as clustering, but it is preferable to employ supervised learning using training data Dm. The training data Dm comprises multiple sets of data Dt, which associate a training image Ig showing the construction joint surface 44 of the concrete 33 with ground truth data Dj regarding the state of the construction joint surface 44. It is preferable that the concrete 33 shown in the training image Ig is made of the same material as the concrete 3 (see Figure 1) at the construction site. In this example, the concrete 33 is a test specimen of concrete 3.

[0076] The ground truth data Dj for the construction joint surface state represents the state of the construction joint surface 4 corresponding to each of the multiple training segmented images Dg obtained by dividing the training image Ig into multiple vertical and horizontal sections. The nine items shown in "Classification 2" in the table in Figure 9 represent the specific state of the construction joint surface 4. In other words, when the ground truth data Dj for the construction joint surface state is represented as a heatmap, it is represented by nine colors.

[0077] Figure 10 shows a flowchart of the evaluation method for the concrete joint surface 4 performed using the evaluation device 8A. This flowchart is executed by the processor and other components of the control unit 2. Specifically, this flowchart is executed when the processor reads a program (concrete joint surface evaluation program) stored in the memory of the control unit 2. In the following explanation, "step" may be abbreviated as "S".

[0078] First, an image acquisition step (S1) is performed to acquire a captured image Im. The processor sends a capture command to the camera 7, and then acquires the captured image Im output from the camera 7. The processor that performs S1 is an example of an image acquisition unit 11.

[0079] Next, the processor performs a joint surface information acquisition step (S3) in which it inputs the captured image Im to the learning model 55A and acquires joint surface information 20 indicating the state of the joint surface 4 in the captured image Im from the learning model 55A. Specifically, the processor sends the captured image Im to the server 50, and the captured image Im is input to the learning model 55A by another processor constituting the server 50. After that, the joint surface information 20 output from the learning model 55A is sent back from the server 50 to the processor of the evaluation device 8A. The processor of the evaluation device 8A that performs S3 is an example of a joint surface information acquisition unit 13.

[0080] Next, the processor determines (S5) whether the joint surface 4 visible in the captured image Im is covered with water W or foreign matter F to such an extent that the quality of the green cut cannot be determined. The determination in S5 is performed based on the joint surface information 20 acquired in S3. If a positive determination is made in S5 (S5:YES), the processor controls the monitor 6 to display an error message (S7). The processor that executes S5:YES is an example of an error determination unit 17.

[0081] If a negative determination is made in S5 (S5: NO), the processor executes a green cut necessity determination step (S9) to determine whether further green cuts are necessary on the joint surface 4, based on the joint surface information 20 acquired in S3. The processor executing S9 is an example of a green cut necessity determination unit 15A.

[0082] If a negative result is obtained in S9 (S9: NO), the processor controls Monitor 6 to display information indicating that an additional green cut is "not needed" (S11). If a positive result is obtained in S9 (S9: YES), the processor controls Monitor 6 to display information indicating that an additional green cut is "needed".

[0083] <Evaluation device 8B(8) according to the second embodiment> The evaluation apparatus 8B(8) according to the second embodiment will be described with reference to Figures 11 and 12. In Figure 11, the same reference numerals are used for the same components as in the first embodiment, and their descriptions may be omitted or simplified below.

[0084] Prior to describing the evaluation device 8B, the captured image Im (see Figure 12) will be described. In the second embodiment, the captured image Im is not divided. The captured image Im includes an error image region Re in which the joint surface 4 is covered by water W and is not visible, and a judging image region Rc in which the joint surface 4 is exposed and visible.

[0085] The learning model 55B according to the second embodiment (see Figure 11) is configured to output construction joint surface information 20 when a captured image Im is input. When the ratio of error image region Re in the captured image Im is relatively high, the construction joint surface information 20 output from the learning model 55B becomes judgment error information. This judgment error information indicates that most of the construction joint surface 4 in the captured image Im is covered by water W, and that it is impossible to determine whether the green cut is good or bad. The green cut necessity determination unit 15B (15) according to the second embodiment does not determine whether further green cut is necessary if the construction joint surface information 20 output from the learning model 55B is judgment error information.

[0086] On the other hand, if the proportion of the determinable image region Rc in the captured image Im is relatively high, the joint surface information 20 output from the learning model 55B will be good / bad information. This good / bad information indicates the quality of the green cut for the joint surface 4 in the captured image Im. More specifically, the determinable image region Rc contains a mixture of states of the joint surface 4 corresponding to "excessive," "good," and "insufficient" as explained in Figure 3, and the state that occupies the largest proportion within the determinable image region Rc is indicated as good / bad information. Even if some water W is captured in the captured image Im, if the proportion of the determinable image region Rc is relatively high, good / bad information will be output from the learning model 55B. If the green cut necessity determination unit 15B(15) is good / bad information, it determines whether additional green cut is necessary based on this good / bad information.

[0087] Furthermore, whether the joint surface information 20 output from the learning model 55B is judgment error information or pass / fail information depends on the training data (not shown) used for machine learning of the learning model 55B. Also, the training data used to output the pass / fail information of the joint surface 4 based on the judgeable image region Rc is the same as in the first embodiment, so the details are omitted.

[0088] According to the above configuration, if the proportion of the determinable image region Rc in the captured image Im is relatively large, the learning model 55B outputs good / bad information. The good / bad information corresponds to the quality of the green cut of the joint surface 4 captured in the determinable image region Rc. Therefore, even if water W is captured in the captured image Im, it is possible to determine whether further green cut is necessary for the joint surface 4 in the captured image Im. Note that the error image region Re may be the region in which foreign matter F covering the joint surface 4 is captured.

[0089] <Other> The construction joint surface 4 according to the above embodiment is a horizontal construction joint surface, but it is also possible to apply the embodiments of this disclosure to a vertical construction joint surface. [Explanation of Symbols]

[0090] 1: Concrete structures 2: Control Unit 3,33: Concrete 4,44: Joint surface 6: Monitor 7: Camera 8A, 8B(8): Evaluation device 9: Judge 11: Image acquisition unit 13: Joint surface information acquisition unit 15A, 15B (15): Green cut necessity determination unit 17: Error detection unit 20: Information on joint surfaces 21: Information on the first split joint surface 22: Information on the second split joint surface 25: Information on split joint surfaces 26: Error Information 27: Determinable information 50: Server 55A, 55B (55): Learning Model D: Split image D1: First split image D2: Second split image Dc: Determinable segmented image De: Error split image Dg: Segmented images for learning DJ: Correct data on the condition of the joint surface Dm: Training data Dt: Data F: Foreign matter Ig: Image for learning Im: Photographed image Ima: First image Imb: Second image L: Boundary line Lh: Horizontal boundary line Lv: Vertical boundary line M: Two-dot chain line P: Intersection Rc: Image area that can be determined Re: Error image area T1~T5: Images W:Water

Claims

1. An image acquisition unit that acquires images of the concrete joint surface with green cut applied, A joint surface information acquisition unit obtains joint surface information indicating the state of the joint surface in the captured image from the learning model by inputting the captured image into a machine learning model, A green cut necessity determination unit determines whether further green cuts are necessary on the joint surface based on the aforementioned joint surface information. An evaluation device for concrete construction joints.

2. The learning model outputs the joint surface information for each of the multiple divided images obtained by dividing the captured image vertically and horizontally into multiple sections. The apparatus for evaluating concrete joint surfaces according to claim 1.

3. The aforementioned joint surface information is, Multiple first divided joint surface information indicating the state of the joint surface, each of which is associated with a plurality of first divided images obtained by dividing the aforementioned captured image vertically and horizontally according to a first pattern, Multiple second-divided joint surface information indicating the state of the joint surface, each associated with a plurality of second-divided images obtained by dividing the captured image vertically and horizontally according to a second pattern different from the first pattern, Includes, The green cut necessity determination unit determines, based on the plurality of first divided joint surface information and the plurality of second divided joint surface information, whether it is necessary to perform the green cut on the joint surface shown in the captured image. The apparatus for evaluating concrete joint surfaces according to claim 2.

4. The second pattern is configured such that the intersections of four adjacent first division images do not overlap with the boundary lines of two adjacent second division images, either vertically or horizontally. The concrete joint surface evaluation device according to claim 3.

5. The plurality of second division images divide a portion of the captured image that is subject to division by the plurality of first division images. The concrete joint surface evaluation device according to claim 3 or 4.

6. The vertical and horizontal dimensions of the first divided image are defined as X1 and Y1, respectively. The vertical and horizontal dimensions of the second divided image are defined as X2 and Y2, respectively. 0.95≦X1 / X2≦1.05...Formula (1) 0.95≦Y1 / Y2≦1.05...Formula (2) Both of these are true. The concrete joint surface evaluation device according to claim 3 or 4.

7. The aforementioned joint surface information includes a plurality of segmented joint surface information, each associated with the plurality of segmented images. The information of the multiple divided joint surfaces includes error information indicating that the joint surface in the divided image is covered with water or foreign matter, and that it is impossible to determine the quality of the green cut. The green cut necessity determination unit determines whether further green cuts are necessary based on the remaining divided images after excluding the divided images associated with the error information from the plurality of divided images. The concrete joint surface evaluation device according to claim 2 or 3.

8. The information of the plurality of divided joint surfaces includes judging information indicating that the joint surface in the divided image is visible and that the quality of the green cut can be determined. The aforementioned multiple segmented images are A determinable segmented image associated with the aforementioned determinable information, The error segmented image associated with the aforementioned error information, Includes, The system further includes an error determination unit that determines that the joint surface visible in the captured image is covered with water or foreign matter if the error image occupancy rate, which correlates with the value obtained by subtracting the number of determinable division images from the number of error division images, exceeds a standard value, If the Green Cut Necessity Determination Unit determines that the joint surface visible in the captured image is covered with water or foreign matter, it will not determine whether further green cutting is necessary. The apparatus for evaluating concrete joint surfaces according to claim 7.

9. The error image occupancy rate is obtained by subtracting from the number of error segmented images the number of determinable segmented images multiplied by a coefficient greater than 1. The apparatus for evaluating concrete joint surfaces according to claim 8.

10. The aforementioned captured image is An error image region in which the joint surface is covered with water or foreign matter and is in a state where it is not visible, A determinable image region in which the aforementioned joint surface is exposed and visible, Includes, The joint surface information is either judgment error information indicating that the joint surface in the captured image is in a state where it is impossible to determine whether the green cut is good or bad, or quality information indicating whether the green cut is good or bad for the joint surface in the captured image. The learning model outputs either the judgment error information or the pass / fail information according to the ratio of the error image region to the judgeable image region in the captured image. The green cut necessity determination unit determines, based on the good / bad information, whether further green cuts are necessary. The apparatus for evaluating concrete joint surfaces according to claim 1.

11. The aforementioned learning model, Machine learning is performed using training data that links training images showing concrete joint surfaces with ground truth data on the state of the joint surfaces in the training images. An evaluation device for concrete construction joints according to any one of claims 1 to 3.

12. Image acquisition step to obtain a photographic image of the concrete joint surface with green cut applied, A joint surface information acquisition step involves inputting the captured image into a machine learning model, thereby obtaining joint surface information from the model that indicates the state of the joint surface in the captured image, and A green cut necessity determination step is to determine whether further green cuts are necessary on the joint surface based on the aforementioned joint surface information. A method for evaluating concrete construction joints, comprising the following features.

13. On the computer, Image acquisition step to obtain a photographic image of the concrete joint surface with green cut applied, A joint surface information acquisition step involves inputting the captured image into a machine learning model, thereby obtaining joint surface information from the model that indicates the state of the joint surface in the captured image, and A green cut necessity determination step is to determine whether further green cuts are necessary on the joint surface based on the aforementioned joint surface information. A program for evaluating concrete joint surfaces that performs the following actions.

Images