Work suitability assessment device

Abstract

To provide a work propriety determination device capable of reducing time and effort of machine learning including labeling when work propriety determination is performed using image recognition by artificial intelligence.SOLUTION: For example, when gesture pointing a mating connector 3 (a target body) is defined to be a predetermined index for making artificial intelligence learn the mating connector 3 in a proper completion state after harness connection work, the pointing gesture is made to learn by artificial intelligence as a pre-stage of learning. When the pointing gesture exists in an image to be provided, the image is an image where the mating connector 3 exists in a state where harness connection work is properly completed. Therefore, labeling in an image extraction step of machine learning is performed by artificial intelligence through making artificial intelligence to extract images in each of which the pointing gesture that is an index exists.SELECTED DRAWING: Figure 4

Description

[Technical field]

[0001] The present invention relates to a work suitability determination device, and in particular to a work suitability determination device that trains artificial intelligence to learn images of an object (object) in a properly completed state where work has been performed properly, and determines that work has been performed properly when an image of an object that approximately matches the learned image of the object in the properly completed state is detected in an image taken after actual work has been performed. [Background technology]

[0002] For example, in the manufacturing industry, image analysis (image recognition) technology is being used in devices and methods that judge whether a specific task has been performed properly. For example, the invention described in Patent Document 1 below is a system in which, when it is determined through image analysis that a worker is performing irregular work (inspection in this case), the content of the work is captured using a wearable camera worn by the worker, in order to later verify the cause of the occurrence of the irregular work that differs from the standard work (inspection in this case) and the content of the work. This document also describes that image judgment is performed to determine whether the work captured by the wearable camera is an item that falls under inspection, which is standard work. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] International Publication No. WO2018 / 062238A1 Summary of the Invention [Problem to be solved by the invention]

[0004] The inspection-eligible product determination described in Patent Document 1 is said to be performed by comparing a stored image of the inspection product with an image of the work captured by a wearable camera, but the image of the work (object, target) captured by a wearable camera or the like is not uniform. For example, when the position, orientation, and appearance of the work in the captured image change, it becomes necessary to learn these extensively and to be able to detect (estimate) the work in the captured image from the learned content. Artificial intelligence (hereinafter referred to as AI) with a learning function can detect objects from images (videos) and detect their position, shape, and other conditions by machine learning. Attempts are being made to use this AI function to determine whether or not work has been performed properly.

[0005] In a predefined task in the manufacturing industry, for example, a determination of whether the task has been performed properly can be made by using, for example, an image of the target object (object) in a completed state. For example, in the case of a task of mating two connectors to wire a harness (wiring material), if the state (shape, orientation, position, etc.) of the mated connectors is proper, it can be determined that the wiring task has been performed properly. Therefore, this properly mated connector can be trained in advance by an AI, and the mated connectors after the wiring task are captured by, for example, a wearable camera (or a fixed camera) worn by a worker. If a properly mated connector is detected in the captured image, it can be determined that the wiring task has been performed properly.

[0006] However, for example, the object captured by the wearable camera, that is, the mating connector, is different in reality for each image in terms of the position of the image of the mating connector, the orientation of the image of the mating connector, and the way the mating connector is captured. When the way the mating connector is captured changes, the shape of the mating connector in the image changes. Therefore, in order to improve the accuracy of the work suitability judgment, specifically the detection accuracy of the object, it is necessary to extract the object from each of a large number of images (still images) cut out from the video in order to cover the changes in the state of these objects, and to have the AI ​​learn how they are captured in various ways. This learning process is called so-called supervised learning labeling, but this labeling requires, for example, a human teacher to extract images in which the object is captured and identify the area in the image in which the object is captured, which is very time-consuming. In other words, in the case of the mating connector, it is necessary for an AI to learn images in which the mating connector is captured, and this is done by a human being, from a huge number of images.

[0007] The present invention has been made in consideration of the above-mentioned problems, and its purpose is to provide a work suitability judgment device that can reduce the effort of machine learning, including labeling, when judging work suitability using image recognition by artificial intelligence. [Means for solving the problem]

[0008] In order to achieve the above object, the work suitability assessment device of the present invention comprises: A work suitability determination device that trains an artificial intelligence on an image of an object in a properly completed state when a work is properly completed, and determines that the work has been performed properly when an image of an object that substantially matches the learned image of the object in the properly completed state is detected in an image after the work is completed, The system is characterized by comprising a first learning means for having the artificial intelligence learn image information of an indicator indicating the presence of an image of the object in the properly completed state; a proper image providing means for providing an image of the object in the properly completed state and an image which may include an image of the indicator; a second learning means for having the artificial intelligence detect an image in which image information of the indicator is present from among the images provided by the proper image providing means after the first learning means has learned the indicator, and for having the artificial intelligence learn image information of the object in the image in which the indicator is present; an imaging means capable of imaging at least an area including the object and the indicator after the work; and a work properness determination means for determining that the work has been performed properly when the artificial intelligence detects image information of the object that approximately matches the image information of the object in the properly completed state that has been learned in the image captured by the imaging means.

[0009] In addition, the indicator indicating the presence of an image of a target object in this invention is a mark indicating that image information of the target object is contained in the image, and broadly refers to, for example, a human movement, a marker displayed in the image, a cursor on a monitor screen, etc.

[0010] According to this configuration, when learning an object in a proper work completion state, detecting an object that substantially matches the learned object in an image after the actual work and judging whether the work is proper or not, for example, when the action of pointing at an object in a proper work completion state is used as an indicator, the pointing action (image information) is learned by the artificial intelligence as a pre-stage of learning. Next, when learning an object in a proper work completion state (image information), if a pointing action is present in the image provided, the image is an image in which an object in a proper work completion state exists, so that the artificial intelligence detects, i.e., extracts an image in which the pointing action, i.e., an indicator, exists, and labeling in the image extraction stage of machine learning is performed by the artificial intelligence. Therefore, before learning an object in a proper work completion state, an indicator such as a pointing action that indicates the presence of an object in a proper work completion state is learned, and the artificial intelligence detects and extracts an image in which this indicator exists, thereby reducing manual labeling, and thereby reducing the effort of machine learning including labeling. In addition, since it is possible to apply an index of a fixed shape that does not change form to the above-mentioned index, by learning this index of a fixed shape, the learning of the object in the proper work completion state can be, for example, rule-based learning, and in this way, the effort required to learn the object in the proper work completion state can be further reduced.

[0011] In another configuration of the present invention, the first learning means causes the artificial intelligence to learn image information of the indicator from the image provided by the appropriate image providing means.

[0012] According to this configuration, as an indicator of the presence of the object in the above-mentioned properly completed state, for example, the action of pointing at the object is learned by the artificial intelligence from an image that may include the pointing action and the object in the properly completed state of the work. Therefore, for example, if an image of an object in which the work has been properly completed and the action of pointing at it are provided, the artificial intelligence detects the image in which the pointing action is detected, and the image information of the object is learned by the artificial intelligence from the image.

[0013] A further configuration of the present invention is characterized in that the first learning means makes the artificial intelligence learn a predetermined area in which image information of the indicator is located in the image provided by the appropriate image providing means or a predetermined area in the vicinity of the image information of the indicator, and the second learning means makes the artificial intelligence learn image information of an object within the predetermined area of ​​the image in which the indicator is located.

[0014] According to this configuration, by having the AI ​​learn the indicator (image information) such as the pointing gesture as well as the predetermined area related to the indicator, and by having the AI ​​learn the target object (image information) within this predetermined area, it is possible to reduce the image learning capacity (number of pixels, etc.), which makes it possible for even a low-level AI to learn the target object with high accuracy, and as a result, high-accuracy work suitability determination can be expected through high-accuracy target object detection. In addition, since the AI ​​performs labeling of the area (area of ​​interest) identification stage that identifies the area where the target object exists for the image containing the indicator such as the pointing gesture, manual labeling can be further reduced.

[0015] A further configuration of the present invention is characterized in that the work suitability determination means detects, by the artificial intelligence, an image in which image information of the index is captured from the images captured by the imaging means, and determines that the work has been performed properly when the artificial intelligence detects image information of an object that approximately matches the object in the properly completed state learned within the specified area in the image in which the index is captured.

[0016] According to this configuration, by detecting an indicator (image information) such as the pointing motion from the captured image and using the image in which the indicator is captured to perform the task suitability judgment, the timing of the task suitability judgment is specified, so that the processing power (processing speed, etc.) of the image judgment can be reduced compared to the case where the task suitability judgment is performed for all images in a video, for example. In addition, by performing the task suitability judgment based on the detection of an object within a predetermined area related to the indicator, it is possible to reduce the capacity of the image judgment (number of pixels, etc.). Therefore, even with a low level of artificial intelligence, a highly accurate task suitability judgment can be performed by highly accurate object detection.

[0017] A further configuration of the present invention is characterized in that the task suitability determination means includes a notification means for notifying the user when the task has been performed improperly.

[0018] According to this configuration, since an indication is given that work has been performed improperly, it is possible to prevent defects from occurring in mass-produced automobile manufacturing industries, for example. Effect of the Invention

[0019] As described above, according to the present invention, when determining the appropriateness of a task using image recognition of an object after the task by artificial intelligence, indices equivalent to labels are learned before learning the object, and the artificial intelligence detects and extracts images in which the indices are present, thereby reducing the effort of machine learning, including labeling. [Brief description of the drawings]

[0020] [Figure 1] 1 is a schematic perspective view showing an embodiment of a mass-produced automobile manufacturing process to which a work suitability determination device of the present invention is applied; [Diagram 2] 2 is an explanatory diagram showing an example of an image captured in the manufacturing process of FIG. 1. [Diagram 3] FIG. 13 is an explanatory diagram of a pointing gesture made by a worker as an indication of the presence of a target object after a task. [Figure 4]4 is a flowchart showing a learning procedure in the task suitability determination device of FIG. 1. [Diagram 5] 3 is a flowchart showing a procedure for judgment in the task suitability judgment device of FIG. 1. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0021] An embodiment of the work suitability determination device of the present invention will be described in detail below with reference to the drawings. FIG. 1 is a schematic perspective view of a mass-produced automobile manufacturing process to which the work suitability determination device of this embodiment is applied. This process includes a task of fitting a connector 3 on the passenger compartment side of a rear gate 1 to connect a harness (wiring material) 4 in an outfitting line for vehicles such as station wagons. The connector 3 is located, for example, slightly below the rear window 2 in the vertical direction of the vehicle. The harness connection task by fitting the connector 3 is performed from below the rear gate 1 with the rear gate 1 open, as shown in the figure. The work suitability determination device of this embodiment determines that the task has been performed properly if a properly fitted connector 3 can be detected in an image after the harness connection task by image recognition.

[0022] The work suitability assessment device of this embodiment is constructed by installing the required application software (hereinafter, application) on a personal computer (hereinafter, PC) 5. The main application is an AI-equipped application capable of image recognition. Needless to say, the PC 5 is a computer system with advanced arithmetic processing capabilities, and is configured with a storage device for storing programs and data, and an input / output device for inputting signals from external devices such as cameras and sensors, and outputting signals to external devices such as monitors and speakers. The application used in this embodiment is a machine learning application based on "supervised learning," but it is also possible to use an application based on unsupervised learning or deep learning. Note that "detection" in image recognition is also called "estimation (inference)."

[0023] The work suitability determination device of this embodiment is configured to include a wearable camera (imaging means) 6 that can be worn by a worker in order to obtain a video (image) of the connector 3 as a target object and the worker's hand. As a condition of this wearable camera 6, it is desirable that it can image an area approximately equal to or larger than the field of vision of the worker. In this embodiment, after the harness wiring work by mating the connector 3, the worker is required to make a gesture (motion) of pointing at the mated connector 3 while looking at it. Therefore, when making a gesture of pointing at the mated connector 3, it is considered that the worker can see the "mating connector" and "the worker's own hand", so the wearable camera 6 may be placed so as to image an area equal to or larger than the field of vision of the worker. In this embodiment, in order to image an area equal to or larger than the field of vision of the worker, the wearable camera 6 is attached to the upper surface of the visor provided in front of the worker's hat. The image (video) signal of this wearable camera 6 is transmitted to the PC 5 by a well-known wireless transmission and reception means (not shown). In addition, the mounting position of the wearable camera 6 is not limited to the above. Moreover, the image signal may be transmitted by wire to the PC 5. Moreover, it is preferable that the pointing gesture of the fitting connector 3 is defined, for example, in a work standard manual.

[0024] FIG. 2 shows an example of an image taken by the wearable camera 6 and read by the PC 5 before the harness wiring work. This image is a schematic diagram of the opened rear gate 1 photographed from below, in which a slightly large square in the center of the figure indicates the rear window 2, two small elongated squares slightly above the image indicate the connector 3 to be mated, and the line segment connected to the left and right of the image of the connector 3 indicates the harness 4. FIG. 3 shows an example of an image taken by the wearable camera 6 and read by the PC 5 after the harness wiring work. As described above, the connector 3 is mated with the harness wiring work, and the mated connector 3 is pointed at by the hand of the worker. In this embodiment, before learning the mated connector 3, which is the target object, the AI-equipped application is made to learn a pointing gesture (image information, pixel information) (first learning means). Specifically, for example, the shape of the worker's hand photographed with the pointing gesture is detected and learned. Such detection can be done by object detection image recognition AI, for example, convolutional neural networks. This pointing gesture learning is done to extract images containing the pointing gesture from the training data images. In other words, it is the labeling of the image extraction stage that allows the AI-equipped application to extract images containing the pointing gesture.

[0025] In this embodiment, the AI-equipped application learns an area as shown by the dashed line in FIG. 3 along with learning the pointing gesture. This area is the location pointed to by the pointing gesture and where the mating connector 3 is present. Since the pointing gesture is an index indicating the location where the target mating connector 3 is present, the area where the worker's finger is pointing is set as a square area shown by the dashed line in the figure along with the pointing gesture, and this area is learned as the location where the mating connector 3 should be detected. That is, the AI ​​is made to perform labeling in the identification stage of the area (also called the area of ​​interest) that identifies the area where the target mating connector 3 is present. Therefore, before learning the target object, the mating connector 3 itself, it is possible to learn the pointing gesture indicating the presence of the target object as an indicator, and further learn the specific area related to the pointing gesture as the location where the target object is present. This makes it possible to detect (extract) an image in which the pointing gesture is present and detect (identify) the area in that image in which the target object is present. By combining these, it is possible to allow the AI-equipped application to perform advanced labeling for target learning.

[0026] In this way, when an image of the mating connector 3 and the pointing gesture after the work is detected and the location (area) of the mating connector 3 in the image is labeled, the AI-equipped application learns the form (shape, position, orientation, etc.) of the mating connector 3 (image information, pixel information) present at that location (second learning means). That is, the AI-equipped application extracts an image of the mating connector 3 as the target from the image after the harness wiring work, and further identifies the imaging location of the mating connector 3 and learns the mating connector 3 that is imaged. When learning the pointing gesture (index), location (area), and mating connector (target) 3, learning can be performed using videos (images) actually captured by the wearable camera, as well as videos and images captured by other cameras, or artificially created videos and images (appropriate image providing means).

[0027] Furthermore, in this embodiment, the appropriateness of the work is determined when a pointing gesture after the work is detected in images captured by the wearable camera 6 during and after the harness wiring work. If a pointing gesture after the work is detected, the location (area) in the image where the mating connector 3 is imaged is identified (extracted) in the same manner as described above, and if a properly mated mating connector 3 (image information, pixel information) is detected in that location, it is determined that the harness wiring work was performed properly. Furthermore, if it is determined that the harness wiring work was performed improperly, a notification to that effect is issued.

[0028] FIG. 4 is a flowchart showing the flow of learning performed by the AI-equipped application installed on the PC 5. Note that this flowchart shows supervised learning in which learning is taught a required number of times, but as described above, unsupervised learning or deep learning may also be used. In this learning flow, first, in step S1, learning is performed on the location (area) where the pointing gesture (indicator) and the mating connector 3 are present from the image (video) captured by the wearable camera 6 and read by the PC 5. Note that, as for the image (video) used for learning, in addition to the image (video) captured by the wearable camera 6 as described above, it is also possible to use, for example, an image (video) captured by another fixed camera capturing an image of a similar area or an artificially created image (video).

[0029] Next, the process proceeds to step S2, where it is determined whether the pointing gesture (indicator) and the location (area) have been learned the required number of times, and if the learning has been completed the required number of times, the process proceeds to step S3, and if not, the process proceeds to step S1. The required number of learning times may be set based on experience, for example, or may be determined by whether the detection (learning) accuracy of the indicator and the area is equal to or higher than a predetermined value.

[0030] In step S3, an image is captured from the work video including the post-work state captured by the wearable camera 5.

[0031] Next, the process proceeds to step S4, where it is determined whether or not a pointing gesture (index) is captured in the image captured in step S3. If a pointing gesture (index) is captured, the process proceeds to step S5; if not, the process proceeds to step S3.

[0032] In step S5 above, the location (area) where the mating connector 3 is imaged is labeled in the image where the pointing gesture (index) is imaged.

[0033] Next, the process proceeds to step S6, where learning of the object in the location (area) labeled in step S5, ie, the mating connector (object) 3, is performed.

[0034] Next, the process proceeds to step S7, where it is determined whether the required number of learnings of the mating connector 3 (target object, object) has been completed, and if the required number of learnings has been completed, the process returns, and if not, the process proceeds to step S3. The required number of learnings may be set, for example, based on experience, or may be determined by whether the detection (learning) accuracy of the target object (object) is equal to or greater than a predetermined value.

[0035] 5 is a flowchart showing the flow of work suitability determination performed by the AI-equipped application installed on the PC 5. In this work suitability determination flow, first, in step S11, it is determined whether or not the pointing gesture (index) is detected in the image (video) of the harness wiring work captured by the wearable camera 6 and read by the PC 5. If the pointing gesture (index) is detected, the process proceeds to step S12, and if not, the process returns.

[0036] In step S12, the location (area) where the mating connector 3 is located in the image where the pointing gesture (indicator) is detected is identified (extracted), and the target object within that location (area), i.e., the mating connector (object) 3, is detected.

[0037] Next, the process proceeds to step S13, where it is determined whether or not an object in the properly completed state of the work, i.e., the mating connector (object) 3, was detected in step S12. If the mating connector 3 was properly detected, the process proceeds to step S14; if not, the process proceeds to step S15.

[0038] In step S14, the process returns after judging that the predetermined work on the target object has been properly performed. In a vehicle outfitting line, for example, the judgment of the proper work may be accompanied by an individual process, such as automatically transporting the vehicle that has been completed to the next process.

[0039] In step S15, the process returns after determining that the predetermined work on the target object was not performed properly. In response to this determination that the work was not performed properly, for example, a notice that the work was not performed properly may be sent via a display, a signal lamp, a buzzer, etc. In addition, for example, in a vehicle outfitting line, individual processing may be performed such that the vehicle is not automatically transported to the next process even when the work is completed.

[0040] According to these calculation processes, when a pointing gesture (indicator) is detected in an image (moving image) of the harness wiring work, the location (area) where the mating connector 3 exists in the image is specified (extracted), and if an object in a properly completed state, i.e., the mating connector (object) 3, is detected in the location (area), it is determined that the harness wiring work was properly performed, and if not, it is determined that the harness wiring work was not properly performed. In learning the mating connector 3, a pointing gesture (indicator) indicating the presence of the mating connector 3 is learned in the image of the learning data (teacher data), and thereafter, images in which the pointing gesture (indicator) exists can be detected (extracted) from among the images of the learning data. In addition, by specifying the location (area) related to the pointing gesture (indicator) where the target mating connector (object) 3 is thought to exist, the area in which the mating connector 3 exists can be limited, and labeling for subsequent learning is performed based on these, so that the labor required for manually labeling the mating connector 3 for learning can be reduced in addition to the detection (extraction) of the image in which the pointing gesture (indicator) exists.

[0041] In this way, in the work suitability determination device of this embodiment, the fitted harness 3 in the work proper completion state is learned, and the fitted harness 3 in the image after the actual work is detected to approximately match the learned fitted harness 3, and the AI ​​learns the pointing gesture in the image read into the PC5 as a pre-learning stage. If the pointing gesture is present in the provided image, the image is an image in which the fitted connector 3 after the harness wiring work is present, so that the AI ​​extracts the image in which the pointing gesture, i.e., the indicator, is present, and labeling in the image extraction stage of machine learning is performed by the AI. Therefore, before learning the fitted connector 3 in the proper completion state after the harness wiring work, an indicator such as a pointing gesture indicating the presence of the fitted connector 3 is learned, and the AI ​​detects and extracts the image in which the indicator is present, thereby reducing manual labeling, and thereby reducing the effort of machine learning including labeling.

[0042] Also, by having the AI ​​learn the specified area related to the indicator, such as the pointing gesture, and by having the AI ​​learn the mating connector 3 within this specified area, it is possible to reduce the image learning capacity (number of pixels, etc.), which makes it possible for even a low-level AI to learn the mating connector 3 with high accuracy, and as a result, highly accurate work suitability determination can be expected through highly accurate object detection. Also, since the labeling in the area identification stage, which identifies the area where the mating connector 3 exists, is performed by AI for images containing indicators such as the pointing gesture, manual labeling is further reduced when performing labeling in the area identification stage.

[0043] In addition, by detecting an indicator such as the pointing gesture from an image captured by the wearable camera 6 and using the image in which the indicator is captured to perform work suitability judgment, the timing of the work suitability judgment is specified, so that the processing power (processing speed, etc.) of the image judgment can be reduced compared to, for example, performing work suitability judgment for all images in a video. In addition, by performing work suitability judgment based on the detection of the mating connector 3 within a predetermined area related to the indicator, it is possible to reduce the capacity of the image judgment (number of pixels, etc.). Therefore, even with a low level of artificial intelligence, highly accurate work suitability judgment can be performed by highly accurate object detection.

[0044] In addition, by reporting any improperly performed work, it is possible to prevent defects from occurring in mass-produced automobile manufacturing industries, for example.

[0045] Although the work suitability determination device according to the embodiment has been described above, the present invention is not limited to the configuration described in the above embodiment, and various modifications are possible within the scope of the gist of the present invention. For example, in the above embodiment, the indicator is a gesture of pointing at the connector 3 mated after the harness wiring work, but this indicator is not limited to this. For example, if the target object is the same mated connector 3, it is possible to use a square frame surrounding the mated connector 3 as the indicator, or a marker provided near the mated connector 3 as the indicator.

[0046] Furthermore, for example, even if a single process has multiple specified tasks and, as a result, multiple objects, if there is sufficient learning capacity in the AI-equipped application and storage capacity in the hardware, the multiple objects can be learned and stored, allowing for a similar assessment of task suitability.

[0047] Furthermore, in the above embodiment, if the specified work on the target object has not been performed properly, a notification to that effect is provided; however, even if the improper work is not notified, it is possible to recognize that the specified work has not been performed properly, for example, if the work has been completed on a vehicle body outfitting line but the vehicle is not automatically transported to the next process.

[0048] In the above embodiment, only the work of fitting the connector 3 of the rear gate 1 of the vehicle and connecting the harness 5 in the body outfitting line is described in detail, but the work suitability determination device of the present invention can be applied to any work in any industry so long as it is generally a manufacturing industry. As an example, it is also possible to target instruments that show the operating state of machinery and equipment after maintenance and inspection work of the machinery and equipment, or the combustion state of combustion gas after repair of combustion equipment.

[0049] In the above embodiment, a gesture of pointing at an object is used as an indicator indicating the presence of an object in a proper work completion state, and this indicator is machine-learned. However, a mark such as a marker other than a human action can also be applied as the indicator, and an example of this is a two-dimensional code. A pointing gesture requires machine learning because it changes depending on the person or the direction of the hand, whereas a two-dimensional code does not require machine learning because it has a constant shape without change. In addition, by having an artificial intelligence learn such an indicator with a constant shape without any morphological change in the first learning stage (means), it is possible to make the second learning stage (means) for learning an object in a proper work completion state a rule-based type, that is, learning according to a specified rule. Therefore, since the first learning stage (means) for learning the indicator is not machine learning, the total learning effort until the object in the proper work completion state is machine-learned can be further reduced. In addition, this rule-based detection can also be used when detecting an object from an image after the actual work is completed.

[0050] Incidentally, for example, the pointing gesture learned in the above embodiment can be used as an indicator of an object in a proper work completion state in a work other than the above connector fitting work, and if the object in the proper work completion state is pointed to with a pointing gesture, the object related to the pointing gesture can be similarly learned in the procedure after step S3 in the flowchart of Fig. 4. The same applies to other indicators. [Explanation of symbols]

[0051] 1 Rear gate 3 (Mating) Connector (target body) 4 Harness (wiring material) 5. PC (Personal Computer) 6 Wearable camera (imaging means, appropriate image providing means)

Claims

1. A work suitability determination device that trains an artificial intelligence on an image of an object in a properly completed state when a work is properly completed, and determines that the work has been performed properly when an image of an object that substantially matches the learned image of the object in the properly completed state is detected in an image after the work is completed, a first learning means for causing the artificial intelligence to learn image information of an indicator indicating the presence of an image of the object in the properly completed state; a proper image providing means for providing an image that may include an image of the object in the proper completion state and an image of the indicator; a second learning means for causing the artificial intelligence to detect an image in which image information of the index exists from among images provided by the appropriate image providing means after the first learning means has learned the index, and for causing the artificial intelligence to learn image information of the object in the image in which the index exists; an imaging means capable of imaging an area including the target object and the indicator after the work; and a work suitability judgment means for judging that the work has been performed properly when the artificial intelligence detects image information of an object in an image captured by the imaging means that approximately matches image information of the object in the properly completed state that has been learned.

2. 2. The task suitability determination device according to claim 1, wherein the first learning means causes the artificial intelligence to learn image information of the indicator from the image provided by the suitability image providing means.

3. The first learning means causes the artificial intelligence to learn a predetermined area in which image information of the indicator is present in the image provided by the appropriate image providing means or a predetermined area in the vicinity of the image information of the indicator; 3. The task suitability determination device according to claim 2, wherein the second learning means causes the artificial intelligence to learn image information of an object within the predetermined area of ​​the image in which the indicator is present.

4. The work suitability determination device according to claim 3, characterized in that the work suitability determination means detects, by the artificial intelligence, an image in which image information of the index is captured from among the images captured by the imaging means, and determines that the work has been performed properly when the artificial intelligence detects image information of an object that approximately matches the object in the properly completed state learned within the specified area in the image in which the index is captured.

5. 5. The work suitability determination device according to claim 4, wherein the work suitability determination means includes a notification means for notifying a user that the work has been performed improperly.

Images