Body hair detection system and machine learning model creation method for detecting body hair

Abstract

To provide a body hair detection system capable of detecting body hair adhering to clothes with high probability.SOLUTION: The body hair detection system includes: illumination devices 31 to 34 which are provided so as to illuminate clothes; imaging devices 41 and 42 which are provided so as to image the clothes; a computer 50 which performs body hair detection processing on the basis of imaging data obtained from the imaging devices; and a display device 60 which displays a body hair detection result detected by the body hair detection processing. The computer 50 includes: a machine learning model recording part 51 on which a machine learning model for detecting the body hair is recorded; an image data division part 53 which divides captured image data; a body hair detection processing part 54 which detects the body hair adhering to the clothes by performing AI inference to image data of each divided area by referring to the machine learning model; and a display control part 55 which performs display control for displaying the body hair detection result on a display screen of the display device 60.SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] The present invention relates to a hair detection system and a method for creating a machine learning model for hair detection. [Background technology]

[0002] In various manufacturing industries, there is a demand for highly accurate detection of foreign objects to prevent them from contaminating products. In particular, in the food manufacturing industry, there is a strong demand for detecting body hair attached to the clothing (including hats, gloves, etc.) worn by workers (cooks). This is because it is necessary to prevent body hair attached to work clothing from falling onto cooked food during cooking, causing the cooked food to be served without the worker noticing that the body hair has fallen.

[0003] To prevent such problems, when workers enter the kitchen, they generally undergo a process of removing body hair from their clothes using an air shower or adhesive roller, but even after undergoing such a process, it is not always possible to completely remove body hair from clothes. Therefore, in addition to these processes, it is desirable to add a process that can detect body hair on clothes with a high probability.

[0004] Various techniques for detecting body hair, for example, human hair, have been proposed in the past (see, for example, Patent Document 1). The hair detection device described in Patent Document 1 is a hair detection device that detects hair using a reflection spectrum obtained by irradiating hair, which is an object to be inspected, with light of multiple wavelengths selected from a predetermined wavelength range (1940 nm to 2400 nm). The hair detection device described in Patent Document 1 is intended to detect hair mixed in with an object to be inspected that is being transported by a belt conveyer. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Japanese Patent Application Laid-Open No. 2012-189390 Summary of the Invention [Problem to be solved by the invention]

[0006] As described above, the hair detection device described in Patent Document 1 detects hair using the reflection spectrum obtained by irradiating the object to be inspected with light of multiple wavelengths. However, even if such a hair detection device is used, it is not necessarily possible to detect body hair attached to clothing.

[0007] That is, when detecting body hair attached to the clothes worn by a worker, it is necessary to detect a wide range of the clothes. Also, even if the clothes are standardized uniforms, the colors of the uniforms vary depending on the manufacturer, and some uniforms have patterns. Furthermore, since wrinkles occur on the clothes, it is necessary to distinguish body hair from wrinkles when detecting them. Therefore, even if the body hair detection device described in Patent Document 1 is applied, it is considered difficult to detect body hair attached to the clothes worn by a worker.

[0008] Therefore, the present invention has been made to solve the above-mentioned problems, and aims to provide a body hair detection system and a method for creating a machine learning model for body hair detection that can detect body hair attached to clothing with a high probability. [Means for solving the problem]

[0009] [1] The hair detection system of the present invention is a system for detecting hair attached to clothing, and includes at least one lighting device that is arranged to be able to irradiate illumination light onto at least the clothing, at least one imaging device that is arranged to be able to image at least the clothing with the illumination light irradiated onto at least the clothing, a computer that performs a hair detection process based on captured image data obtained by the imaging device capturing at least the clothing, and a display device that displays an image corresponding to the captured image data and a hair detection result obtained by the hair detection process, and the computer includes a machine learning model recording unit that records a machine learning model for detecting hair, an image data dividing unit that divides the captured image data into a plurality of regions, and a display unit that divides the captured image data into a plurality of regions. The device comprises a hair detection processing unit that detects hair attached to the clothing by performing AI inference on the image data of each divided region obtained by dividing the image data into multiple regions and referring to the machine learning model, and a display control unit that performs display control to display an image corresponding to the captured image data and the hair detection results by the hair detection processing unit on the display screen of the display device, wherein the machine learning model for hair detection is obtained by machine learning a collection of learning data for each divided region that has been annotated and tagged for the hair samples present in each divided region obtained by dividing the sample captured image data obtained by capturing an image of at least one clothing sample to which at least one type of hair sample is attached into multiple regions.

[0010] [2] In the hair detection system of the present invention, when hair is detected as a result of the hair detection processing by the hair detection processing unit, it is preferable that the display control unit adds a mark indicating the presence of hair to the captured image data and displays it on the display screen.

[0011] [3] In the hair detection system of the present invention, if the hair detection processing by the hair detection processing unit results in no hair being detected, it is preferable that the display control unit displays information indicating that no hair has been detected on the display screen.

[0012] [4] In the body hair detection system of the present invention, it is preferable that the computer further has an image data cut-out unit that cuts out from the captured image data an area surrounded by the smallest rectangular frame that contains at least the image area corresponding to the clothing as image data to be processed, and that the image data division unit divides the image data to be processed cut out by the image data cut-out unit into multiple areas.

[0013] [5] In the body hair detection system of the present invention, it is preferable that the illumination device has an illumination direction set so that the optical axis of the light emitted by the illumination device does not coincide with the optical axis of the imaging device.

[0014] [6] In the body hair detection system of the present invention, it is preferable that the captured image data includes information on the shadow of body hair that appears on the clothing, and the machine learning model includes information on the shadow of the body hair sample that appears on the clothing sample.

[0015] [7] In the body hair detection system of the present invention, it is preferable that the clothing to be detected for the adhesion of body hair is clothing worn by a worker, and that the imaging device is configured to be able to capture an image of the worker wearing the clothing.

[0016] [8] In the body hair detection system of the present invention, it is preferable that the imaging device has at least one imaging device installed so as to be able to image the upper body side of the worker wearing the clothing, and at least one imaging device installed so as to be able to image the lower body side of the worker wearing the clothing.

[0017] [9] In the body hair detection system of the present invention, it is preferable that the lighting device has at least one lighting device installed so as to be able to illuminate the upper body side of the worker wearing the clothing, and at least one lighting device installed so as to be able to illuminate the lower body side of the worker wearing the clothing.

[0018]

[10] In the body hair detection system of the present invention, the lighting device is preferably provided with a polarizing plate that polarizes the illumination light from the lighting device.

[0019]

[11] In the body hair detection system of the present invention, the imaging device is provided with a polarizing plate that polarizes light incident on the lens of the imaging device, and it is preferable that the polarizing plate provided in the imaging device and the polarizing plate provided in the lighting device are set so that the vibration direction of light is the same.

[0020]

[12] In the body hair detection system of the present invention, it is preferable that the body hair includes at least one of hair, beard, eyelashes, eyebrows, leg hair, and nose hair, and that the color of the body hair includes at least one of black, brown, white, and golden.

[0021]

[13] In the body hair detection system of the present invention, it is also preferable that the sample image data obtained by imaging at least one clothing sample having at least one type of body hair sample attached thereto using the imaging device is divided into multiple regions by the image data division unit, and that machine learning is performed on a collection of learning data for each divided region that has been annotated and tagged with the body hair samples present in each divided region obtained by the division.

[0022]

[14] The method for creating a machine learning model for body hair detection of the present invention is a method for creating a machine learning model for body hair detection used in a body hair detection system that detects body hair attached to clothing, and is characterized by having a computer execute the following steps: a division step of dividing sample image data obtained by imaging at least one clothing sample to which at least one type of body hair sample is attached into multiple regions; and a machine learning step of machine learning a collection of learning data for each divided region that has been annotated and tagged for the body hair samples present in each divided region obtained by the division step.

[0023]

[15] In the method for creating a machine learning model for body hair detection of the present invention, an image data cutting step is performed before the division step, in which an area surrounded by the smallest rectangular frame that at least contains an image area corresponding to the clothing sample is cut out from the sample captured image data as image data to be processed, and it is preferable that the division step divides the image data to be detected, cut out in the image data cutting step, into multiple areas. [Effects of the Invention]

[0024] The hair detection system of the present invention detects hair attached to clothing by performing AI inference on captured image data obtained by capturing images of clothing with an imaging device, referencing a machine learning model for hair detection. This allows for high-probability detection of hair attached to clothing. In particular, the machine learning model for hair detection included in the hair detection system of the present invention was created by machine learning a collection of training data for each divided region, which was annotated and tagged for the hair samples present in each divided region, after capturing image data of at least one clothing sample with at least one type of hair attached thereto. This allows for high-probability detection of hair attached to clothing. It also makes it possible to distinguish between wrinkles in clothing and hair.

[0025] By installing such a body hair detection system of the present invention, for example, at the entrance to a kitchen of a food manufacturer, if body hair is attached to the clothing worn by a worker, the attached body hair can be detected with a high probability.

[0026] The method for creating a machine learning model for detecting body hair of the present invention includes a division step of dividing the sample image data obtained by imaging at least one clothing sample having at least one type of body hair sample attached thereto into multiple regions, and a machine learning step of performing machine learning on a collection of training data annotated and tagged for the body hair samples present in each divided region obtained by the division. By using the machine learning model of the present invention created in this way in a body hair detection system, body hair attached to clothing can be detected with a high probability. It is also possible to distinguish between wrinkles in clothing and body hair. [Brief explanation of the drawings]

[0027] [Figure 1] 1 is a diagram showing the external configuration of a body hair detection system 1 according to a first embodiment. [Figure 2] 2 is a diagram showing a state in which a worker wearing clothes is standing at the body hair detection position of the body hair detection system 1 shown in FIG. 1. FIG. [Figure 3] 1 is a block diagram illustrating the functions of each component required for the hair detection process performed by the hair detection system 1 according to the first embodiment. [Figure 4] 1 is a flowchart illustrating a method for creating a machine learning model for body hair detection. [Figure 5] FIG. 10 is a diagram showing an example of sample captured image data cut out into the smallest rectangular frame A that contains at least an image area corresponding to a clothing sample. [Figure 6] FIG. 10 is a photograph showing an example image of one divided area obtained by dividing image data for one frame of sample captured image data obtained by capturing an image of the upper body side of a clothing sample. [Figure 7] FIG. 10 is a photograph showing an example image of one divided area obtained by dividing image data for one frame of sample captured image data obtained by capturing an image of the lower body side of a clothing sample. [Figure 8] 10 is a diagram showing an example in which a mark indicating the presence of body hair attached to the clothes 20 worn by the worker 10 is displayed on the display screen 61. FIG. [Figure 9] FIG. 10 is a block diagram illustrating the functions of the components necessary for the hair detection process performed by the hair detection system 2 according to the second embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0028] The body hair detection system of the present invention will be described below with reference to the drawings. The body hair detection system of the present invention can be applied to various fields, but in the following embodiment, a body hair detection system that detects body hair attached to clothing (cooking clothing) worn by a worker such as a chef will be described as an example. In the following embodiment, hair will be used as an example of body hair, but the system can also be applied to detecting not only hair but also beard, eyebrows, eyelashes, leg hair, and the like. Body hair also includes animal hair. As mentioned above, in the following embodiment, clothing is exemplified as cooking clothing, but in the following description, it will simply be referred to as "clothing." Furthermore, if the worker is wearing a hat and gloves, the clothing to be detected for hair attachment will also include the hat and gloves.

[0029] [Embodiment 1] Fig. 1 is a diagram showing the external configuration of body hair detection system 1 according to embodiment 1. Fig. 2 is a diagram showing a state in which worker 10 wearing clothes (cooking clothes) is standing at the body hair detection position of body hair detection system 1 according to embodiment 1 shown in Fig. 1. The external configuration of body hair detection system 1 according to embodiment 1 will be described below.

[0030] The hair detection system 1 of embodiment 1 includes at least one (four in embodiment 1) lighting device 31-34 that is configured to be able to irradiate illumination light onto at least clothing, at least one (two in embodiment 1) imaging device 41, 42 that is configured to be able to capture at least clothing while the illumination light emitted by the lighting devices 31-34 is irradiated onto at least the clothing, a computer 50 (see Figure 3) that performs hair detection processing based on captured image data obtained by imaging at least the clothing with the imaging devices 41, 42 capturing images of at least the clothing, and a display device 60 that displays an image corresponding to the captured image data and also displays the hair detection results detected by the hair detection processing performed by the computer 50.

[0031] The above-mentioned "at least the clothes" means not only the clothes but also the worker wearing the clothes. Therefore, for example, "the illumination light can be irradiated onto at least the clothes" also includes "the illumination light can be irradiated onto the worker wearing the clothes."

[0032] When describing the lighting devices 31 to 34 individually, they may be referred to as the first lighting device 31, the second lighting device 32, the third lighting device 33, and the fourth lighting device 34. Similarly, when describing the imaging devices 41 and 42 individually, they may be referred to as the first imaging device 41 and the second imaging device 42. Furthermore, although the computer 50 is housed in the controller box 70 shown in FIGS. 1 and 2, it may be located elsewhere. Furthermore, the computer 50 may be a single-board computer having a function specialized for body hair detection processing.

[0033] Furthermore, a plate 80 on which the worker 10 stands when detecting body hair is provided in front of the controller box 70. A position mark 81 indicating the position where the worker 10 should stand is drawn on this plate 80. An example of this position mark 81 is a mark representing the outline of a foot, indicating the position where the foot should be placed.

[0034] In addition, the example shown is one in which the display device 60 is installed at a position beside the worker 10 when the worker 10 stands on the position mark 81 of the plate 80 as shown in Fig. 2, but the installation position of the display device 60 is not limited to the position shown in Fig. 1 and Fig. 2. For example, the display device 60 may be installed at a position where the worker 10 can see the display screen 61 without turning to the side when the worker 10 stands on the position mark 81 of the plate 80 so as to face the imaging devices 41 and 42.

[0035] Furthermore, the lighting devices 31 to 34 and the imaging devices 41 and 42 are attached to support columns 91 to 93 that are erected in the vertical direction along the z-axis as shown in the figure. Specifically, the lighting devices 31 to 34 are attached to support columns 92 and 93 located on both the left and right sides diagonally behind the central support column 91 as seen from the perspective of an operator 10 (see FIG. 2) standing on the plate 80. Of the lighting devices 31 to 34, the first lighting device 31 and the second lighting device 32 are attached to the upper ends of the support columns 92 and 93 via flexible arms 31 a and 32 a, and the third lighting device 33 and the fourth lighting device 34 are attached to the lower ends of the support columns 92 and 93 via flexible arms 33 a to 34 a.

[0036] The lighting devices 31 to 34 are configured to illuminate the worker 10 from each direction. For example, the first lighting device 31 can illuminate the worker 10 from diagonally above on the left side, and the second lighting device 32 can illuminate the worker 10 from diagonally above on the right side. Furthermore, the third lighting device 33 can illuminate the worker 10 from diagonally below on the left side, and the fourth lighting device 34 can illuminate the worker 10 from diagonally below on the right side.

[0037] The light sources used in the lighting devices 31 to 34 are not particularly limited, but are preferably LEDs, and a plurality of LEDs (not shown) are arranged in a plane in each of the lighting devices 31 to 34. The lighting of the lighting devices 31 to 34 can be controlled individually. Therefore, all of the lighting devices 31 to 34 can be turned on, or only any one of the lighting devices can be turned on.

[0038] The illumination direction of the illumination devices 31 to 34 can be adjusted by flexible arms 31a, 32a, 33a, and 34a. The illumination direction of the illumination devices 31 to 34 may be manually adjustable, or may be automatically adjusted by an illumination direction adjustment unit (not shown). Each of the illumination devices 31 to 34 is dimmable, and the brightness (e.g., illuminance) can be adjusted within a predetermined range.

[0039] The imaging devices 41 and 42 are attached at a predetermined interval in the vertical direction along the z-axis to the central support 91, as seen from the worker 10 standing on the plate 80, among the three support columns 91 to 93. Specifically, the first imaging device 41 is provided so as to be able to capture an image of the upper body of the worker 10 when the worker 10 stands substantially upright facing the imaging devices 41 and 42 as shown in FIG. 2. Here, the upper body includes the head and hands of the worker. Therefore, when the worker 10 is wearing a hat and gloves, "capable of capturing an image of the upper body of the worker 10" means that the hat and gloves can also be captured.

[0040] 2, the second imaging device 42 is provided so as to be able to capture an image of the lower body side of the worker 10, including the footwear, when the worker 10 stands substantially upright facing the imaging devices 41, 42. However, the imaging areas of the first imaging device 41 and the second imaging device 42 may be set so as to partially overlap.

[0041] Furthermore, it is preferable that the imaging devices 41 and 42 are set so that their optical axes are approximately perpendicular to the worker 10. On the other hand, as described above, the lighting devices 31 to 34 are set so that the illumination light is irradiated obliquely onto the worker 10. That is, the lighting devices 31 to 34 are set so that the optical axis of each illumination light from the lighting devices 31 to 34 does not coincide with the optical axis (optical axis of the lens) of the imaging devices 41 and 42. In other words, the irradiation direction of each illumination light (optical axis of the illumination light) of the lighting devices 31 to 34 is set so as to form a predetermined angle with the optical axis of each lens of the imaging devices 41 and 42.

[0042] In this way, the lighting devices 31 to 34 are set so that the optical axis of each illumination light from the lighting devices 31 to 34 does not coincide with the optical axis (optical axis of the lens) of the imaging devices 41 and 42. As a result, if body hair is attached to the clothing 20 worn by the worker 10, the shadow of the body hair can be cast on the clothing. Therefore, information about the shadow of body hair is incorporated into the captured image data obtained from the imaging devices 41 and 42. By incorporating information about the shadow of body hair into the captured image data, it is possible to detect white (such as white hair) or golden body hair with a particularly high probability.

[0043] In addition, it is preferable that the support columns 91-93 are extendable in the vertical direction, which allows the vertical positions of the lighting devices 31-34 and the imaging devices 41, 42 to be set to suit the height of the worker 10. Note that the lighting devices 31-34 and the imaging devices 41, 42 may be movable in the vertical direction.

[0044] Incidentally, the lighting devices 31 to 34 may be provided with a polarizing plate (not shown) on the light-emitting surface side of each lighting device. This polarizing plate is expected to have the effect of suppressing scattering of illumination light from the lighting devices 31 to 34 on the clothing 20, which is the illuminated surface. It is preferable that the polarization plates provided on the lighting devices 31 to 34 are configured so that the vibration direction of light passing through each polarization plate is the same, either vertically or horizontally. Polarizing plates may be provided not only on the lighting device 31 to 34 side but also on the incident side of the imaging devices 41 and 42. In this case, the polarization plates provided on the lighting devices 31 to 34 and the polarization plates on the imaging devices 41 and 42 side have the same vibration direction of light passing through them.

[0045] By providing such polarizing plates to at least the lighting devices 31 to 34, it is possible to expect the elimination of glare caused by light scattering and the improvement of contrast, thereby making it possible to make the color and contrast of the captured image clearer and to obtain a captured image in which details are more clearly reproduced.

[0046] The display screen 61 of the display device 60 is a vertically long screen, which makes it easy to display an image of the whole body of the worker 10 standing at the position mark 81. The display screen 61 may display the whole body of the worker 10, or may display a zoomed-in image of part of the worker 10, such as the upper or lower body.

[0047] Next, the hair detection process performed by the hair detection system 1 according to the first embodiment will be described. Fig. 3 is a block diagram showing the components necessary for the hair detection process performed by the hair detection system 1 according to embodiment 1. As shown in Fig. 3, the components necessary for the hair detection process include lighting devices 31 to 34, imaging devices 41 and 42, a display device 60, and a computer 50 that performs the hair detection process.

[0048] The computer 50 includes a machine learning model recording unit 51 in which a machine learning model for detecting body hair is recorded; an image data cutting unit 52 that cuts out image data to be processed (described later) from the captured image data obtained by the imaging devices 41 and 42; an image data division unit 53 that divides the image data to be processed cut out by the image data cutting unit 52 into multiple regions; a hair detection processing unit 54 that detects body hair attached to clothing by performing AI (artificial intelligence) inference on image data (referred to as divided image data) corresponding to each divided region divided into multiple regions by the image data division unit 53 by referring to the machine learning model recorded in the machine learning model recording unit 51; and a display control unit 55 that performs display control to display the captured images obtained by the imaging devices 41 and 42 and the body hair detection results by the hair detection processing unit 54 on the display screen 61 of the display device 60.

[0049] 3, components necessary for the hair detection process performed by the hair detection system 1 according to embodiment 1 include an illumination device control unit (not shown) that controls the illumination devices 31 to 34 and an imaging device control unit (not shown) that controls the imaging devices 41 and 42. These may be provided as functions of the computer 50, or may be provided as functions separate from the computer 50.

[0050] Furthermore, the machine learning model for body hair detection recorded in the machine learning model recording unit 51 may be created by a machine learning model creation system for body hair detection prepared separately from the body hair detection system 1 shown in Fig. 1, or may be created by the body hair detection system 1 shown in Fig. 1. In the body hair detection system 1 according to the first embodiment, the machine learning model for body hair detection will be described as being created by a machine learning model creation system for body hair detection (not shown) prepared separately from the body hair detection system shown in Fig. 1.

[0051] Here, the process of creating a machine learning model for detecting body hair will be described. When creating a machine learning model for detecting body hair, it is preferable to set various lighting conditions and image capturing conditions to find conditions that make it easy to detect body hair when actually performing body hair detection using the body hair detection system 1 according to the first embodiment (when performing body hair detection on the clothing 20 worn by the worker 10).

[0052] 4 is a flowchart illustrating a method for creating a machine learning model for detecting body hair. The steps for creating a machine learning model for detecting body hair will be outlined with reference to the flowchart shown in FIG.

[0053] A clothing sample with a body hair sample attached thereto is imaged to obtain sample captured image data (step S1). From the sample captured image data obtained in step S1, image data to be processed is extracted (step S2). A specific example of the process of extracting image data to be processed from the sample captured image data in step S2 will be described later.

[0054] The image data to be processed cut out in step S2 is then divided into a plurality of regions (step S3). The body hair samples present in each divided region obtained by the division are annotated and tagged to create training data (step S4). Then, a collection of the annotated and tagged training data for each divided region (training data set) is subjected to machine learning (step S5). Of steps S1 to S5 shown in FIG. 4, at least the processes of steps S2, S3, and S5 are executed by a computer.

[0055] Among the processes to be executed by the computer, the process of machine learning the collection of annotated and tagged learning data (learning data set) for each divided region (processing of step S5) is preferably executed by a workstation or a cloud-based computer with high processing power. In this case, the collection of learning data (learning data set) is provided to the workstation or the cloud-based computer, machine learning is performed on the workstation or the cloud-based computer, and the machine learning model for body hair detection created thereby is downloaded and recorded in the machine learning model recording unit 51.

[0056] The processing of each step shown in FIG. 4 will be specifically described. The body hair samples prepared here are black, brown, white (such as white hair), and blond body hair samples. In addition, in this embodiment, since cooking clothes are used as an example of clothing samples, light-colored clothing such as white, light blue, and light brown are prepared as clothing samples. The various prepared body hair samples are then attached to various clothing samples to prepare various clothing samples with the various body hair samples attached. In the following description, "various clothing samples with various body hair samples attached" may also be simply referred to as "clothing samples."

[0057] Here, clothing samples will be prepared for the upper body and for the lower body, but unless the upper body clothing samples and the lower body clothing samples are described separately, both will be collectively referred to as "clothing samples."

[0058] The processing of step S1 is performed on the clothing samples thus prepared (various clothing samples with various body hair samples attached). That is, while illuminating light from a lighting device (not shown) prepared for creating a machine learning model for body hair detection is irradiated onto each clothing sample, each clothing sample is imaged using an imaging device (not shown) also prepared for creating the machine learning model, to obtain sample captured image data. At this time, the imaging device captures color video at, for example, 5 frames per second.

[0059] When capturing images of clothing samples with an imaging device, it is preferable to capture images of the clothing samples worn by a person or on a mannequin, but it is also possible to capture images of the clothing samples placed on the floor, a table, etc. When capturing images of clothing samples, it is preferable to place a monochrome screen or the like behind the clothing sample so that it can be clearly distinguished from the clothing. This makes it possible to capture image data of the clothing captured with an imaging device in monochrome, with the background clearly distinguishable from the clothing, making it easier to perform various image processing.

[0060] Furthermore, when capturing an image of the clothing sample, it is preferable to set the optical axis of the illumination light from the lighting device so that it does not coincide with the optical axis of the image capturing device. This setting allows the shadow of body hair attached to the clothing sample to be cast on the clothing sample. Therefore, the sample captured image data obtained by the image capturing device includes information about the shadow of the body hair sample.

[0061] In this way, information about the shadow of the hair sample is captured in the sample captured image data, which is particularly effective when detecting whitish (such as white hair) or golden hair. When capturing images of whitish hair such as white hair or golden hair, it is preferable to capture the image under illumination settings that do not cause so-called "blown-out highlights" in the captured image.

[0062] As described above, clothing samples for the upper body and clothing samples for the lower body are prepared as clothing samples, and therefore, the upper body clothing samples with the respective body hair samples attached thereto are imaged, and the lower body clothing samples with the respective body hair samples attached thereto are imaged to obtain respective sample image data.

[0063] Once a large number of clothing samples (clothing samples with body hair samples attached) have been imaged and the sample captured image data for each sample has been obtained in this manner, the process of step S2 in Fig. 4 is carried out. Specifically, step S2 is a process of extracting image data to be processed from the sample captured image data obtained in step S1.

[0064] The process of extracting the image data to be processed from the sample image data will be specifically described below. The process of extracting the image data to be processed is performed for each frame of the sample image data acquired by acquiring the clothing sample image data frame by frame.

[0065] The process of extracting image data to be processed from the sample captured image data is a process of extracting, as image data to be processed, an area enclosed by the smallest rectangular frame that can fit at least an image area corresponding to the clothing sample. Here, "at least an image area corresponding to the clothing sample" refers to, for example, "an image area corresponding to the mannequin wearing the clothing sample" when an image of a mannequin wearing the clothing sample is captured by an imaging device.

[0066] Therefore, "the smallest rectangular frame that can fit inside at least the image area corresponding to the clothing sample" means "the smallest rectangular frame that can fit inside the image area corresponding to a mannequin wearing the clothing sample." On the other hand, if the clothing sample alone is imaged by an imaging device without the clothing sample being worn by a mannequin, "the smallest rectangular frame that can fit inside at least the image area corresponding to the clothing sample" means "the smallest rectangular frame that can fit inside the image area corresponding only to the clothing sample."

[0067] Specifically, the process of setting the "smallest rectangular frame that can fit at least the image area corresponding to the clothing sample" can be performed as follows. Here, an example is shown in which an image of a mannequin wearing a clothing sample is captured. In this case, the contour of the mannequin is recognized from the captured image data obtained by capturing an image of the mannequin wearing the clothing sample, and the smallest rectangular frame that can fit the contour of the mannequin is set based on the coordinate values ​​of the contour of the mannequin obtained from the recognition result.

[0068] Fig. 5 is a diagram illustrating an example of setting the smallest rectangular frame A that can fit at least an image area corresponding to a clothing sample from sample captured image data. Fig. 5 shows an example of setting an area surrounded by the smallest rectangular frame A that can fit a mannequin wearing a clothing sample (in this case, the upper body including part of the lower body of the mannequin). The area surrounded by the rectangular frame A shown in Fig. 5 becomes the image data to be processed. However, because Fig. 5 is a diagram illustrating an example of processing to set the rectangular frame A, the image data to be processed shown in Fig. 5 itself will not necessarily be used in subsequent processing (for example, processing to divide the image data).

[0069] Incidentally, although not shown, sample captured image data obtained by an imaging device capturing an image of a mannequin wearing a clothing sample actually has a wide range of monochromatic background outside the rectangular frame A shown in Fig. 5. By cutting out the range surrounded by the rectangular frame A from the sample captured image data containing a wide range of monochromatic background, an image is obtained from which the excess background outside the rectangular frame A has been removed. This makes it possible to significantly reduce the amount of data processing required for processing from step S3 onwards in Fig. 4.

[0070] Next, the process of step S3 in Fig. 4 is performed. The process of step S3 is a process of dividing the image data to be processed cut out in step S2 into regions of a predetermined size, and obtaining divided image data for each divided region.

[0071] Here, the number of divisions is not particularly limited, but for example, in the case of sample captured image data corresponding to a clothing sample of the upper body side (image data to be processed cut out by rectangular frame A), the sample captured image data of the upper body side (image data to be processed cut out by rectangular frame A) acquired on a frame-by-frame basis is divided into, for example, 70. On the other hand, in the case of sample captured image data corresponding to clothing of the lower body side (image data to be processed cut out by rectangular frame A), the sample captured image data of the lower body side (image data to be processed cut out by rectangular frame A) acquired on a frame-by-frame basis is divided into, for example, 50. Note that each divided area is square, and has the same number of pixels in the vertical direction and the horizontal direction, but the number of pixels is not limited to this, and the number of pixels in the vertical direction and the horizontal direction is not limited to being the same.

[0072] Next, the process of step S4 in Fig. 4 is performed. In the process of step S4, the body hair samples present in each divided region divided in step S3 are annotated and tagged to create learning data for machine learning.

[0073] In addition, the processing of step S4 involves annotating and tagging the divided image data of each divided area obtained by dividing the sample captured image data (image data to be processed) into each frame unit, and therefore, by performing the processing of step S4, a huge amount of learning data is created.

[0074] Specifically, for example, if image data for one upper body clothing sample (a clothing sample with a body hair sample attached) is output for 10 seconds at 5 frames per second and each frame is divided into 70 parts, then 70 x 5 x 10 = 3,500 pieces of training data will be obtained. Furthermore, by performing similar processing with different types of body hair samples or different types of clothing samples, a huge amount of training data can be obtained.

[0075] Next, the process of step S5 is performed. In the process of step S5, machine learning is performed using a collection of a huge amount of annotated and tagged learning data (learning dataset) to create a machine learning model.

[0076] Figure 6 is a photograph showing an example of an image of one divided area obtained by dividing one frame of image data of sample captured image data (image data to be processed) obtained by capturing an image of a clothing sample on the upper body side. Note that Figure 6(a) is a diagram showing a state before annotation and tagging of body hair samples present in the image of one divided area, and Figure 6(b) is a photograph showing an example of annotation and tagging of body hair samples present in the image of one divided area shown in Figure 6(a).

[0077] Figure 7 is a photograph showing an example of an image of one divided region obtained by dividing one frame of sample captured image data (image data to be processed) obtained by capturing an image of a clothing sample on the lower body. Note that Figure 7(a) is a diagram showing the state before annotation and tagging of the body hair sample present in Figure 7(a), and Figure 7(b) is a photograph showing an example of annotation and tagging of the body hair sample present in Figure 7(a).

[0078] Here, because the hair is very fine, the hair sample attached to the clothing sample is difficult to see in Figures 6 and 7. Note that in Figures 6 and 7, one hair sample is denoted by the reference numeral 101, but many hair samples are scattered within one divided area.

[0079] 6(b), which shows an example of annotation and tagging of an image of a divided region, shows the state in which some representative hair samples among the many hair samples shown in FIG. 6(a) have been annotated and tagged. In FIG. 6(b), the annotated and tagged hair samples are shown surrounded by rectangular frames. However, in reality, the other hair samples present in FIG. 6(a) are also annotated and tagged.

[0080] Similarly, Fig. 7(b) shows an example of annotating and tagging an image of a divided region, showing the annotated and tagged state of several representative hair samples among the many hair samples shown in Fig. 7(a). In Fig. 7(b), the annotated and tagged hair samples are shown surrounded by rectangular frames. However, in reality, the other hair samples present in Fig. 7(a) are also annotated and tagged.

[0081] In the computer 50, a machine learning model for body hair detection is created by performing machine learning processing using a collection of a huge amount of annotated and tagged learning data (learning data set) shown in Figures 6(b) and 7(b). The machine learning model for body hair detection created in this manner is recorded in the machine learning model recording unit 51 of the computer 50 (see Figure 3) installed in the body hair detection system 1 according to the first embodiment shown in Figure 1.

[0082] Next, a hair detection process for detecting whether or not body hair is attached to the clothing 20 worn by the worker 10 using the hair detection system 1 according to the first embodiment will be described. In the first embodiment, the worker is assumed to be a cook, and therefore to be wearing cooking clothing 20 (upper-body clothing 20a and lower-body clothing 20b). When the worker 10 wearing the clothing 20 enters a kitchen, the hair detection system 1 shown in FIG. 1 detects whether or not body hair is attached to the clothing 20. If body hair is detected, the worker 10 removes the hair. Needless to say, when entering a kitchen, several steps, such as hand washing and sterilization, as well as hair detection, are performed, and hair detection is one of these steps.

[0083] First, as shown in Fig. 2, worker 10 wearing cooking clothes 20 stands in a substantially upright position at the position of position mark 81 drawn on plate 80, facing image capture devices 41 and 42, and slowly turns around clockwise or counterclockwise from that position. It is preferable to install a monochrome screen or the like behind worker 10 facing image capture devices 41 and 42, so that the color of the screen can be clearly distinguished from the color of the clothing 20. In this way, the images obtained by image capture devices 41 and 42 capturing the worker have a monochrome background that can be clearly distinguished from the clothing, allowing various image processing processes to be performed with high precision.

[0084] In this way, when worker 10 stands at position mark 81, imaging devices 41 and 42 are activated to capture images of worker 10 moving around at, for example, 5 frames per second, and imaging devices 41 and 42 output color captured image data. At this time, first imaging device 41 outputs captured image data for one rotation of worker 10's upper body, and second imaging device 42 outputs image data for one rotation of worker 10's lower body. When capturing images with imaging devices 41 and 42, it is preferable to set the illumination conditions, such as the illumination angle of lighting devices 31-34 with respect to worker 10 and the illuminance on the surface of clothing 20, to conditions optimal for body hair detection.

[0085] The captured image data of one circumference of the worker's upper body obtained from the first imaging device 41 and the captured image data of one circumference of the worker's lower body obtained from the second imaging device 42 are provided to an image data clipping unit 52 shown in Fig. 3. The image data clipping unit 52 acquires the captured image data (captured image data of the upper body side and captured image data of the lower body side) obtained by the imaging devices 41, 42 on a frame-by-frame basis, and clips the captured image data of the upper body side and the captured image data of the lower body side on an acquired frame-by-frame basis into the smallest rectangular frame that can accommodate at least an image area corresponding to the clothing 20 (in this case, an image area corresponding to the worker 10 wearing the clothing 20), thereby obtaining image data to be processed.

[0086] The process of cutting out the rectangular frame can be performed in substantially the same manner as the process described in step S2 when creating a machine learning model. However, in this case, the processing target image data is not cut out from the sample captured image data, but is cut out from captured image data obtained by capturing an image of the clothes worn by the worker. In this way, by the image data cutout unit 52 cutting out the processing target image data, processing target image data from which unnecessary background and the like have been removed can be obtained. As a result, when performing each process after the image data division unit 53, unnecessary background and the like can be excluded from the processing target.

[0087] The processing target image data cut out by the image data cutout unit 52 is then divided into a plurality of regions by the image data division unit 53. Here, the processing target image data for the upper body side is divided into, for example, 70 parts. On the other hand, the processing target image data for the lower body side is divided into, for example, 50 parts. In this case, the divided image data obtained by division have the same number of pixels in the vertical direction and the same number of pixels in the horizontal direction.

[0088] Each divided image data obtained from the image data dividing unit 53 is given to the hair detection processing unit 54, where hair detection processing is performed. The hair detection processing performed by the hair detection processing unit 54 is performed by performing AI inference on each divided image data divided in units of frames, by referring to the machine learning model recorded in the machine learning model recording unit 51.

[0089] When hair detection processing unit 54 detects hair in certain divided image data as a result of performing AI inference, hair detection processing unit 54 provides the detection result to display control unit 55. For example, if hair detection processing unit 54 detects that hair is attached to the left chest area of ​​upper body clothing 20a worn by worker 10, hair detection information is output to display control unit 55. Upon receiving the hair detection information from hair detection processing unit 54, display control unit 55 performs display control on display device 60 to display a mark indicating the presence of hair at the position where hair was detected.

[0090] Fig. 8 is a diagram showing an example in which a mark indicating the presence of body hair adhering to clothing 20 worn by worker 10 is displayed on display screen 61. Note that Fig. 8 shows an example in which a mannequin is used as an experiment, but the same experiment can be performed on a human worker. Fig. 8 also shows an example in which a rectangular frame surrounding detected body hair is displayed on display screen 61 as mark M indicating the presence of body hair.

[0091] Because FIG. 8 is a monochrome image, it is not possible to display different colors, but as an example, a rectangular frame (mark M) surrounding the body hair is displayed in red, for example. Note that while the example shown here illustrates a case where a rectangular frame is displayed as mark M, the mark M is not limited to a rectangular frame and may be a circle, or the detected body hair may be indicated by an arrow or the like. Also, while FIG. 8 illustrates an example where the upper half of the worker's body is displayed, the entire body may also be displayed.

[0092] The worker can see at a glance which parts of the clothing 20 he is wearing have body hair attached by looking at his own image displayed on the display screen 61. When the worker 10 visually identifies the part with body hair attached, he removes the attached body hair with an adhesive roller or the like while looking at the display screen 61.

[0093] It should be noted that the location where body hair is attached is not limited to one location, and there may be, for example, multiple locations. In this case, a mark M indicating the presence of each of the multiple attached body hairs is displayed. Furthermore, if body hair is detected in a location that the worker cannot see (for example, on the back), the display screen 61 displays the worker's back. In this way, the worker 10 can reliably recognize that body hair is attached, even if body hair is attached in multiple locations or in a location that the worker cannot see.

[0094] On the other hand, if no hair is detected as a result of the hair detection process by the hair detection processing unit 54, it is preferable to display information indicating that no hair is detected on the display screen 61. In this case, the display control unit 55 may display, for example, information such as "not detected" or a "○" mark indicating no detection on the display screen 61, although this is not shown in the drawings. This allows the worker 10 to see at a glance that no hair is attached to the garment 20.

[0095] Incidentally, the clothes worn by chefs (clothes for cooking) are generally light-colored clothing such as white or light blue. Because these light-colored clothing and black and brown body hair are easily distinguishable, information (feature values) about these body hairs can be easily extracted from image data obtained by capturing an image of clothing with black and brown body hairs attached.

[0096] On the other hand, if the body hair is white, such as gray hair, or golden, such as blonde, even if the white or golden hair is attached to white clothing, it is difficult to extract information (features) about the white or golden hair from image data obtained by capturing an image of the clothing with hair of these colors. In particular, if the illumination on the surface of the clothing is too high, the "whiteout" phenomenon may occur. In this case, it becomes even more difficult to extract information (features) about the white or golden hair from the image data.

[0097] To prevent this, when detecting white hair such as gray hair or golden hair such as blonde hair, it is preferable to capture an image after setting the illuminance so that information about white hair such as gray hair or golden hair such as blonde hair can be easily extracted from the captured image data. In this case, it is preferable to set the illuminance suitable for detecting white or golden hair. Furthermore, it is preferable to capture an image under lighting conditions set by the lighting device so that shadows of white or golden hair are easily cast on clothing. For example, lighting conditions set by the lighting device so that shadows of hair are easily cast on clothing include selectively operating the lighting devices 31 to 34.

[0098] As described above, in the body hair detection system 1 according to the first embodiment, body hair attached to the clothing worn by the worker 10 is detected by performing AI inference with reference to a machine learning model for body hair detection on image data obtained by imaging the worker 10 wearing clothing 20 such as a cook's suit using the imaging devices 41 and 42. As a result, if body hair is attached to the clothing 20 worn by the worker 10, the attached body hair can be detected with a high probability.

[0099] In particular, the machine learning model is created by machine learning a huge collection of training data (training dataset) that is annotated and tagged for each divided image data of each divided region obtained by dividing a large number of sample image data obtained by capturing images of various clothing samples with various body hair samples attached thereto. As a result, even when various types of body hair are attached to various types of clothing, it is possible to detect attached body hair with a high probability. It is also possible to distinguish between wrinkles in clothing and body hair.

[0100] Furthermore, with the body hair detection system 1 according to the first embodiment, the worker can detect body hair by simply standing on the position mark 81 of the plate 80 and turning around once, for example, as shown in FIG. 2. Therefore, since one body hair detection can be performed easily and in a short time, body hair detection can be performed repeatedly while working. For example, if the worker is a cook, he or she can perform body hair detection and then temporarily stop cooking, perform body hair detection again using the body hair detection system 1 according to the first embodiment, and then return to cooking. In this way, body hair detection can be performed repeatedly at predetermined intervals. Even when body hair detection is performed repeatedly in this way, one body hair detection can be performed easily and in a short time, so work efficiency is not significantly affected.

[0101] [Embodiment 2] A hair detection system 2 according to embodiment 2 will be described below. The hair detection system 2 according to embodiment 2 has a function of creating a machine learning model for hair detection. The hair detection system 2 according to embodiment 2 will be described below. Note that the external configuration of the hair detection system 2 according to embodiment 2 is the same as that of the hair detection system 1 according to embodiment 1, so illustration of the external configuration and description of each component will be omitted.

[0102] 9 is a block diagram illustrating the functions of each component required for the hair detection process performed by the hair detection system 2 according to the second embodiment. In the hair detection system 2 according to the second embodiment, the lighting device that illuminates the clothing sample to which the hair sample is attached and the imaging device that images the clothing sample to which the hair sample is attached are performed using lighting devices 31 to 34 (see FIG. 1) and imaging devices 41 and 42 (see FIG. 1) provided in the hair detection system 2 according to the second embodiment. Then, a machine learning processing unit 56 performs machine learning on a collection of learning data (learning data for each annotated and tagged divided region) created based on the sample captured image data obtained by the imaging devices 41 and 42. Note that, among the components shown in FIG. 9, the same components as those shown in FIG. 1 are denoted by the same reference numerals.

[0103] The process of creating a machine learning model for body hair detection performed by the machine learning processing unit 56 can be performed through processing steps substantially similar to steps S1 to S5 in Fig. 4. In addition, in the body hair detection system 2 according to the second embodiment, the lighting device that illuminates the clothing sample to which the body hair sample is attached and the imaging device that images the clothing sample to which the body hair sample is attached can use the lighting devices 31 to 34 (see Fig. 1) and imaging devices 41 and 42 (see Fig. 1) provided in the body hair detection system 2 according to the second embodiment, as described above.

[0104] Then, the prepared clothing samples (various clothing samples with various body hair samples attached) are illuminated with illumination light from lighting devices 31 to 34, and each clothing sample is imaged by imaging devices 41 and 42 to obtain sample image data.

[0105] The sample captured image data obtained by the imaging devices 41 and 42 is provided to the image data cropping unit 52. The image data cropping unit 52 acquires the sample captured image data (sample captured image data of the upper body side and sample captured image data of the lower body side) obtained by the imaging devices 41 and 42 on a frame-by-frame basis, and crops the acquired frame-by-frame sample captured image data of the upper body side and sample captured image data of the lower body side into a minimum rectangular frame that fits at least an image area corresponding to the clothing sample inside, to obtain image data to be processed.

[0106] The image data to be processed obtained by the image data cropping unit 52 is then divided into multiple regions by the image data dividing unit 53. After that, the body hair samples present in each divided region are annotated and tagged. The machine learning processing unit 56 then performs machine learning using a collection of a huge amount of annotated and tagged learning data (learning dataset) to create a machine learning model. The machine learning model created by the machine learning processing unit 56 is recorded in the machine learning model recording unit 51.

[0107] As described above, in the hair detection system 2 according to the second embodiment, a machine learning model is created in the hair detection system 2. In this case, the lighting devices and imaging devices used when creating the machine learning model can be the same as those used when actually detecting hair. This allows the lighting conditions and imaging conditions used when creating the machine learning model to be set closer to the lighting conditions and imaging conditions used when actually detecting hair. This allows for a higher probability of detecting hair attached to the clothing 20 worn by the worker 10. Note that in the hair detection system 2 according to the second embodiment, in order to create a machine learning model in the hair detection system 2, it is preferable to use a computer 50 such as a workstation with high processing power.

[0108] The hair detection process for detecting hair attached to clothing worn by a worker by the hair detection system 2 according to the second embodiment can be performed in the same manner as the hair detection system 1 according to the first embodiment. Therefore, a description of the hair detection process by the hair detection system 2 according to the second embodiment will be omitted in the second embodiment. Furthermore, the machine learning model created in the hair detection system 2 according to the second embodiment can be used as a machine learning model for hair detection not only in the hair detection system 2 according to the second embodiment, but also in hair detection systems that perform hair detection under similar conditions.

[0109] The present invention is not limited to the above-described embodiment, and various modifications can be made without departing from the spirit of the present invention. For example, modifications such as those shown below are also possible.

[0110] (1) In the above-described embodiment, hair has been used as an example of body hair, but body hair other than hair includes, for example, beard, eyebrows, eyelashes, and leg hair. Body hair also includes animal hair. By creating a machine learning model based on machine learning of these body hairs other than hair, it is possible to detect these body hairs with a high probability.

[0111] (2) The configuration of the body hair detection system shown in Fig. 1 is not limited to the configuration shown in Fig. 1 and can be modified as appropriate. For example, the configuration of the supports to which the lighting devices 31-34 and the imaging devices 41, 42 are attached can be modified in various ways. In addition, although the body hair detection system 1 shown in Fig. 1 uses two imaging devices, the number of imaging devices is not limited to two, and one imaging device or three or more imaging devices may be used as long as they can capture an image of the entire body of the worker with high accuracy. The same applies to the lighting devices. In the body hair detection system 1 according to the first embodiment shown in Fig. 1, four lighting devices are used as an example, but the number of lighting devices is not limited to four.

[0112] (3) In the above-described embodiments, a hair detection system that detects hairs attached to the clothing of a worker such as a chef when the worker enters a kitchen has been described as an example, but the hair detection system of the present invention can also be used in other industries, such as the semiconductor and other precision parts and precision equipment manufacturing industry, the pharmaceutical manufacturing industry, and the optical equipment manufacturing industry.

[0113] (4) In the above-described embodiments, the imaging device captures images at 5 frames per second, but this is not limited to 5 frames per second. Also, still images may be captured instead of moving images.

[0114] (5) In the above-described embodiments, when detecting whether or not body hair is attached to clothing, an example is given in which the worker stands at position mark 81 and moves around that position clockwise or counterclockwise. However, this is not limiting. For example, a mechanism may be provided in which the image capturing device (e.g., image capturing device 41, 42) moves around the worker while the worker remains stationary. Furthermore, multiple image capturing devices may be arranged at corresponding positions so as to capture images of areas necessary for body hair detection, such as the back (the back side of the worker) and both sides of the worker when the worker stands at position mark 81.

[0115] (6) In the above-described second embodiment, the lighting devices that illuminate the clothing sample having the body hair sample attached thereto and the imaging devices that image the clothing sample having the body hair sample attached thereto use the lighting devices 31 to 34 (see FIG. 1) and imaging devices 41 and 42 (see FIG. 1) provided in the body hair detection system 2 according to the second embodiment. The example illustrates a case in which the machine learning processing unit 56 performs machine learning on a collection of learning data (learning data for each annotated and tagged divided region) created based on the sample captured image data obtained by the imaging devices 41 and 42. That is, in the second embodiment, the example illustrates a case in which machine learning is performed by the machine learning processing unit 56 included in the computer 50 of the body hair detection system 2 according to the second embodiment, but the machine learning may also be performed by an external computer.

[0116] In this case, the collection of learning data for each annotated and tagged divided area is provided to an external computer (e.g., a workstation or a computer on the cloud), which performs machine learning on the external computer, and the machine learning model for body hair detection created thereby is downloaded and recorded in the machine learning model recording unit 51.

[0117] (7) Although not mentioned in the above embodiments, the machine learning model for the body hair detection system can be updated at a predetermined timing. An example of a method for updating the machine learning model is fine tuning. For example, newly acquired data is added to an already trained machine learning model, and the model is trained to adjust parameters for body hair detection. By performing such fine tuning, the accuracy of body hair detection can be further improved. [Explanation of symbols]

[0118] 1,2...Body hair detection system, 10...Worker, 20...Clothing, 31-34...Lighting device, 41,42...Imaging device, 50...Computer, 51...Machine learning model recording unit, 52...Image data extraction unit, 53...Image data division unit, 54...Body hair detection processing unit, 55...Display control unit, 56...Machine learning processing unit, 60...Display device, 61...Display screen, 70...Control box, 80...Plate, 81...Position mark, 91,92,93...Frame, 101...Body hair (hair), A...Minimum rectangular frame that can fit at least an image area corresponding to a clothing sample inside, M...Mark indicating the presence of body hair

Claims

1. A hair detection system for detecting hair attached to clothing, comprising: at least one lighting device that is capable of irradiating illumination light onto at least the garment; at least one imaging device that is configured to be able to capture an image of at least the clothing in a state where the illumination light is irradiated onto at least the clothing; a computer that performs a body hair detection process based on captured image data obtained by the imaging device capturing an image of at least the clothing; a display device that displays an image corresponding to the captured image data and also displays a result of the hair detection process; Equipped with The computer a machine learning model recording unit in which a machine learning model for detecting body hair is recorded; an image data dividing unit that divides the captured image data into a plurality of regions; a hair detection processing unit that detects hair attached to the clothing by performing AI inference on image data of each divided region obtained by dividing the captured image data into a plurality of regions and referring to the machine learning model; a display control unit that performs display control to display an image corresponding to the captured image data and a result of the hair detection by the hair detection processing unit on a display screen of the display device; and The machine learning model for detecting body hair comprises: A body hair detection system characterized by being obtained by dividing sample image data obtained by imaging at least one clothing sample having at least one type of body hair sample attached into multiple regions, and performing machine learning on a collection of learning data for each divided region that has been annotated and tagged for the body hair samples present in each divided region.

2. 10. The hair detection system of claim 1, The hair detection system is characterized in that, when hair is detected as a result of the hair detection processing by the hair detection processing unit, the display control unit adds a mark indicating the presence of hair to the captured image data and displays it on the display screen.

3. 10. The hair detection system of claim 1, The display control unit is characterized in that, if the result of the hair detection process by the hair detection processing unit is that no hair is detected, information indicating that no hair is detected is displayed on the display screen.

4. 10. The hair detection system of claim 1, The computer an image data cropping unit that crops, from the captured image data, an area surrounded by a minimum rectangular frame that contains at least an image area corresponding to the clothing, as image data to be processed; The body hair detection system is characterized in that the image data dividing unit divides the processing target image data cut out by the image data cutting unit into a plurality of regions.

5. 10. The hair detection system of claim 1, The body hair detection system is characterized in that the illumination device is set so that the illumination direction of the light emitted by the illumination device does not coincide with the optical axis of the imaging device.

6. 6. The hair detection system according to claim 5, the captured image data includes information about a shadow of body hair that appears on the clothing, The hair detection system, wherein the machine learning model includes information about the shadow of the hair sample that appears on the clothing sample.

7. 10. The hair detection system of claim 1, A body hair detection system characterized in that the clothing to be detected for body hair adhesion is clothing worn by a worker, and the imaging device is configured to be able to capture an image of the worker wearing the clothing.

8. 8. The hair detection system according to claim 7, The body hair detection system is characterized in that the imaging device has at least one imaging device installed so as to be able to image the upper body of the worker wearing the clothing, and at least one imaging device installed so as to be able to image the lower body of the worker wearing the clothing.

9. 8. The hair detection system according to claim 7, The body hair detection system is characterized in that the lighting device has at least one lighting device installed so as to be able to illuminate the upper body side of the worker wearing the clothing, and at least one lighting device installed so as to be able to illuminate the lower body side of the worker wearing the clothing.

10. 10. The hair detection system of claim 1, The body hair detection system is characterized in that the lighting device is provided with a polarizing plate that polarizes illumination light from the lighting device.

11. 11. The hair detection system of claim 10, A body hair detection system characterized in that the imaging device is provided with a polarizing plate that polarizes light incident on the lens of the imaging device, and the polarizing plate provided in the imaging device and the polarizing plate provided in the lighting device are set so that the vibration direction of the light is the same.

12. 10. The hair detection system of claim 1, The body hair detection system is characterized in that the body hair includes at least one of hair, beard, eyelashes, eyebrows, leg hair, and nose hair, and the color of the body hair includes at least one of black, brown, white, and golden colors.

13. 10. The hair detection system of claim 1, A body hair detection system characterized by: capturing an image of at least one clothing sample having at least one type of body hair sample attached thereto using the imaging device; dividing the sample captured image data obtained by the imaging device into multiple regions using the image data division unit; and performing machine learning on a collection of learning data for each divided region that has been annotated and tagged for the body hair samples present in each divided region obtained by the division.

14. A method for creating a machine learning model for detecting body hair to be used in a body hair detection system for detecting body hair attached to clothing, comprising: a dividing step of dividing sample image data obtained by imaging at least one clothing sample having at least one type of body hair sample attached thereto into a plurality of regions; a machine learning step of performing machine learning on a collection of learning data for each divided region that has been annotated and tagged with respect to the body hair samples present in each divided region obtained by the division step; A method for creating a machine learning model for body hair detection, comprising causing a computer to execute the above steps.

15. 15. The method for creating a machine learning model for body hair detection according to claim 14, an image data cutting step is performed before the dividing step, in which an area surrounded by a minimum rectangular frame that contains at least an image area corresponding to the clothing sample is cut out from the sample captured image data as image data to be processed; A method for creating a machine learning model for body hair detection, characterized in that the division step divides the image data to be detected and cut out in the image data cut-out step into multiple regions.

Images