Apparatus and method for updating learning models for image inspection

Abstract

To secure determination accuracy of image inspection using a learning model.SOLUTION: An apparatus for updating a learning model for image inspection includes: a light source information acquisition unit which acquires first light source information indicating one or both of a position and a direction of a light source in a real space acquired through a captured image; a light source information comparison unit which determines whether the first light source information is different from second light source information which indicates one or both of a position and a direction of a light source of a first learning image used in training of a learning model; a learning image generation unit which arranges, when a determination is made that the first light source information is different from the second light source information, a three-dimensional model obtained by simulating a subject and a virtual light source obtained by simulating a light source in accordance with the first light source information, in a virtual space, and generates a plurality of second learning images representing a three-dimensional model while changing one or both of a position and a direction of the three-dimensional model; and a learning model generation unit which updates the learning model using the second learning images.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present disclosure relates to an apparatus and method for updating a learning model for image inspection. [Background technology]

[0002] Patent Document 1 discloses a technique for changing the position of a light source in an image obtained by capturing an image of an object using a camera or the like. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2013-235537 Summary of the Invention [Problem to be solved by the invention]

[0004] A learning model generated by machine learning is sometimes used in image inspection. However, for example, when the position or direction of a light source illuminating an object to be inspected is changed due to a change in the layout of a factory, the accuracy of the judgment in the image inspection may be reduced. To solve this problem, it is possible to change the position of the light source in the image obtained by capturing an image of the object to be inspected with a camera, as in the above-mentioned document, but this alone does not ensure the accuracy of the judgment in the image inspection. [Means for solving the problem]

[0005] The present disclosure can be realized in the following forms.

[0006] (1) According to a first aspect of the present disclosure, there is provided an apparatus for updating a learning model for image testing, the apparatus comprising: a light source information acquisition unit that acquires first light source information representing one or both of a position and an orientation of a light source in a real space acquired by capturing an image of the real space in which an object to be tested for image testing is placed; a light source information comparison unit that determines whether second light source information representing one or both of the position and orientation of the light source in a first learning image used for training the learning model, the one or both of which are represented by the first light source information, differs from the first light source information; a learning image generation unit that, when it is determined by the light source information comparison unit that the first light source information and the second light source information differ, places a three-dimensional model simulating the object to be tested and a virtual light source simulating the light source in accordance with the first light source information in a virtual space, and generates a plurality of second learning images representing the three-dimensional model while changing one or both of a position and an orientation of the three-dimensional model in the virtual space; and a learning model generation unit that updates the learning model using the plurality of second learning images generated by the learning image generation unit. According to this aspect, when one or both of the position and orientation of the light source in real space is changed, a three-dimensional model illuminated by a virtual light source disposed at a position and orientation corresponding to the position and orientation of the light source in real space is used to generate a plurality of second training images in which one or both of the position and orientation of the three-dimensional model differ from each other, and the training model can be updated using the plurality of second training images, thereby ensuring the accuracy of judgment in image inspection of the object to be inspected. (2) The device of the above form may also include an image inspection unit that performs image inspection of the object to be inspected using an image obtained by imaging the object to be inspected after the learning model has been updated by the learning model generation unit and the learning model updated by the learning model generation unit. According to the device of this aspect, it is possible to perform image inspection of the object to be inspected using the updated learning model. (3) In the device of the above form, the light source information comparison unit may determine whether the first light source information and the second light source information are different when the error rate of image inspection by the image inspection unit exceeds a predetermined value. According to this type of device, if the error rate in image inspection increases due to a change in the position or orientation of the light source, the learning model can be automatically updated to reduce the error rate in image inspection. (4) The device of the above form may include a memory unit that stores the first learning image, and the light source information acquisition unit may acquire the second light source information by analyzing the first learning image stored in the memory unit. According to the device of this aspect, the second light source information can be acquired using the first learning image stored in the storage unit. (5) According to a second aspect of the present disclosure, there is provided a method for updating a learning model for image testing, the method comprising: a light source information acquisition step of acquiring first light source information representing one or both of information about a position and an orientation of a light source in a real space acquired by using an image of a real space in which an object to be tested for image testing is placed; a light source information comparison step of determining whether second light source information representing one or both of information about the position and orientation of the light source in a first learning image used for training the learning model, the second light source information representing the one or both of information represented by the first light source information, differs from the first light source information; a learning image generation step of, if it is determined in the light source information comparison step that the first light source information and the second light source information differ, placing a three-dimensional model simulating the object to be tested and a virtual light source simulating the light source in accordance with the first light source information in a virtual space, and generating a plurality of second learning images representing the three-dimensional model while changing one or both of the position and orientation of the three-dimensional model in the virtual space; and a learning model generation step of updating the learning model using the plurality of second learning images generated in the learning image generation step. According to this embodiment, when one or both of the position and orientation of a light source in real space is changed, a three-dimensional model illuminated by a virtual light source disposed at a position and orientation corresponding to the position and orientation of the light source in real space is used to generate a plurality of second training images of the three-dimensional model that differ from one another in one or both of the position and orientation, and the training model can be updated using the plurality of second training images, thereby ensuring the accuracy of judgment in image inspection of the object to be inspected. The present disclosure can be realized in various forms other than an apparatus for updating a learning model for image inspection or an apparatus for updating a learning model for image inspection, such as an image inspection system, an image inspection apparatus, or an image inspection method. [Brief explanation of the drawings]

[0007] [Figure 1] FIG. 1 is an explanatory diagram illustrating a schematic configuration of an image inspection system. [Figure 2] 10 is a flowchart showing the contents of a learning model update process. [Figure 3] 10 is a flowchart showing the contents of a learning image generation process. [Figure 4] FIG. 1 is an explanatory diagram showing a three-dimensional model placed in a virtual space. DETAILED DESCRIPTION OF THE INVENTION

[0008] A. First embodiment: 1 is an explanatory diagram showing a schematic configuration of an image inspection system 10 according to a first embodiment. In this embodiment, the image inspection system 10 includes a stage 20, a light source 30, a camera 40, a camera controller 45, an information processing device 50, and a display device 60. The information processing device 50 may also be referred to as a device for updating a learning model GM for image inspection, or as an image inspection device.

[0009] The stage 20 is placed indoors, for example, in a factory. An object RM to be inspected by image inspection is placed on the stage 20. In this embodiment, the object RM to be inspected is an automobile bumper. Even for bumpers of the same vehicle model, various parts such as emblems and grilles attached to the bumper may differ depending on the grade, sales region, etc. The bumper as the object RM to be inspected is inspected by the image inspection system 10 to ensure that various parts are properly attached depending on the grade and sales region. The object RM to be inspected is transported, for example, by a robot arm (not shown) and placed on the stage 20.

[0010] The light source 30 irradiates light onto the inspection object RM placed on the stage 20. For example, an LED (Light Emitting Diode) illumination can be used as the light source 30. The light source 30 is fixed by a light source fixing jig (not shown).

[0011] The camera 40 captures an image of the object under inspection RM placed on the stage 20. The camera 40 captures an image of the object under inspection RM irradiated with light from the light source 30. The camera 40 may be, for example, a camera having a CCD (Charge-Coupled Device) image sensor or a camera having a COMS (Complementary Metal Oxide Semiconductor) image sensor. The camera 40 is fixed by a camera fixing jig (not shown). The camera 40 is connected to a camera controller 45 by wired or wireless communication, and an output signal from the camera 40 is transmitted to the camera controller 45.

[0012] The camera controller 45 generates a captured image RG by processing the output signal from the camera 40. Furthermore, in this embodiment, the camera controller 45 analyzes the captured image RG to estimate the position and orientation of the light source 30 and generate light source information LD representing the estimated position and orientation of the light source 30. The camera controller 45 estimates the position and orientation of the light source 30, for example, by a successive relaxation method that searches for reflection parameters, the position of the light source 30, and the orientation of the light source 30 that simultaneously satisfy the conditions that the diffuse reflection component of the reflected light represented in the captured image RG follows the Lambertian model and the specular reflection component of the reflected light represented in the captured image RG follows the Torrance-Sparrow reflection model. The camera controller 45 can estimate the position and orientation of the light source 30 from one captured image RG using the above-mentioned method. The camera controller 45 is connected to an information processing device 50 via wired or wireless communication, and the captured image RG and light source information LD generated by the camera controller 45 are transmitted to the information processing device 50.

[0013] The information processing device 50 includes a CPU 101, a memory 102, and an input / output interface 103. In this embodiment, the information processing device 50 includes a captured image acquisition unit 110, a light source information acquisition unit 120, a light source information comparison unit 130, a learning image generation unit 140, a learning model generation unit 150, and an image inspection unit 160. The captured image acquisition unit 110, the light source information acquisition unit 120, the light source information comparison unit 130, the learning image generation unit 140, the learning model generation unit 150, and the image inspection unit 160 are realized in software by the CPU 101 executing a computer program stored in the memory 102. Note that at least one of the captured image acquisition unit 110, the light source information acquisition unit 120, the light source information comparison unit 130, the learning image generation unit 140, the learning model generation unit 150, and the image inspection unit 160 may be realized by a hardware circuit. The memory 102 may also be referred to as a storage unit.

[0014] The captured image acquisition unit 110 acquires the captured image RG. In this embodiment, the captured image acquisition unit 110 acquires the captured image RG transmitted from the camera controller 45. The captured image RG acquired by the captured image acquisition unit 110 is stored in the memory 102.

[0015] The light source information acquisition unit 120 acquires light source information LD that represents information about the position and orientation of the light source 30. In this embodiment, the light source information acquisition unit 120 acquires the light source information LD transmitted from the camera controller 45. The light source information LD acquired by the light source information acquisition unit 120 is stored in the memory 102.

[0016] The light source information comparison unit 130 compares the two pieces of light source information LD to determine whether or not at least one of the position and orientation of the light source 30 represented in the two pieces of light source information LD is different.

[0017] The training image generation unit 140 generates training images VG to be used in machine learning by the learning model generation unit 150. In this embodiment, the training image generation unit 140 is configured by a 3D simulator. The training images VG generated by the training image generation unit 140 are stored in the memory 102.

[0018] The learning model generation unit 150 generates the learning model GM by executing machine learning using the learning image VG. The learning model GM generated by the learning model generation unit 150 is stored in the memory 102.

[0019] The image inspection unit 160 performs image inspection of the object RM to be inspected, using the captured image RG obtained by capturing an image of the object RM with the camera 40 and the learning model GM generated by the learning model generation unit 150. The inspection result RD of the image inspection by the image inspection unit 160 is stored in the memory 102.

[0020] The display device 60 is configured, for example, with a liquid crystal display or an organic electroluminescence (EL) display. The display device 60 displays the inspection result RD of the image inspection by the image inspection unit 160. For example, if the image inspection unit 160 determines that the various parts corresponding to the grade and sales region are properly attached to the bumper, which is the object RM to be inspected, the display device 60 displays the word "OK." If the various parts corresponding to the grade and sales region are not properly attached to the bumper, the display device 60 displays the word "NG." The display device 60 is, for example, disposed near the stage 20 in a factory. By checking the inspection result RD displayed on the display device 60, an inspector can determine whether the various parts corresponding to the grade and sales region are properly attached to the bumper. If the word "NG" is displayed on the display device 60, a visual inspection is performed by the inspector. If the visual inspection results indicate that the various parts corresponding to the grade and sales region are properly attached to the bumper, the inspector inputs into the information processing device 50 that the image inspection unit 160 made an incorrect judgment. The inspection result RD includes information on the error judgment rate.

[0021] In the image inspection system 10 of this embodiment, the image inspection unit 160 performs image inspection using the learning model GM, thereby reducing the burden on the inspector compared to a visual inspection of the inspection object RM by an inspector. Furthermore, image inspection using the learning model GM, as in this embodiment, generally requires less time than visual inspection by an inspector. However, for example, if the position or orientation of the light source 30 is changed due to a change in the layout within a factory, the accuracy of the image inspection performed by the image inspection unit 160 may decrease. A decrease in the accuracy of the image inspection means an increase in the rate of misidentification. A decrease in the accuracy of the image inspection occurs, for example, when the color of the inspection object RM displayed in the captured image RG changes before and after a change in the position or orientation of the light source 30. In this embodiment, even if the position or orientation of the light source 30 is changed, the information processing device 50 executes a learning model update process to update the learning model GM used in the image inspection, thereby reducing the rate of misidentification in the image inspection.

[0022] Fig. 2 is a flowchart showing the contents of the learning model update process. Fig. 3 is a flowchart showing the contents of the learning image generation process executed in the learning model update process. Fig. 4 is an explanatory diagram showing a three-dimensional model VM placed in a virtual space VS used in the learning image generation process.

[0023] 2, in this embodiment, the learning model update process is started by the CPU 101 of the information processing device 50 when the erroneous determination rate of the image inspection represented in the inspection result RD stored in the memory 102 exceeds a predetermined value. When the learning model update process is started, first, in step S110, the captured image acquisition unit 110 controls the camera 40 via the camera controller 45 to capture an image of the inspection object RM using the camera 40, and acquires the captured image RG of the inspection object RM from the camera controller 45.

[0024] Next, in step S120, the light source information acquisition unit 120 acquires light source information LD representing information about the position and orientation of the light source 30. In this embodiment, the camera controller 45 generates the light source information LD by analyzing the captured image RG to estimate the position and orientation of the light source 30, and the light source information acquisition unit 120 acquires the light source information LD generated by the camera controller 45. The light source information LD acquired by the light source information acquisition unit 120 is stored in the memory 102. Note that step S120 may also be referred to as a light source information acquisition step of a method for updating a learning model GM for image inspection, or a light source information acquisition step of an image inspection method.

[0025] In step S130, the light source information comparison unit 130 determines whether at least one of the position and orientation of the light source 30 has changed by comparing the first light source information LD1, which is light source information LD acquired in step S120 and represents the current position and orientation of the light source 30, with the second light source information LD2, which is light source information LD representing the past position and orientation of the light source 30 stored in the memory 102. In this embodiment, the light source information comparison unit 130 determines that at least one of the position and orientation of the light source 30 has changed when at least one of the position and orientation of the light source 30 represented in the first light source information LD1 and at least one of the position and orientation of the light source 30 represented in the second light source information LD2 deviate from each other. The deviation between at least one of the position and orientation of the light source 30 represented in the first light source information LD1 and at least one of the position and orientation of the light source 30 represented in the second light source information LD2 refers to at least one of the following states: the position of the light source 30 represented in the first light source information LD1 is separated from the position of the light source 30 represented in the second light source information LD2 by a predetermined distance or more; and the orientation of the light source 30 represented in the first light source information LD1 is separated from the orientation of the light source 30 represented in the second light source information LD2 by a predetermined angle or more. Note that step S130 may also be referred to as a light source information comparison step in a method for updating a learning model GM for image inspection or a light source information comparison step in an image inspection method. The light source information comparison unit 130 uses, as the second light source information LD2, light source information LD representing the position and orientation of the light source 30 in the training images VG used to train the training model GM used in image inspection before the learning model update process began. The learning images VG used to learn the learning model GM that was used in the image inspection before the start of the learning model update process may be referred to as first learning images VG1.

[0026] If it is determined in step S130 that at least one of the position and orientation of the light source 30 has not been changed, the information processing device 50 ends this process. If it is determined in step S130 that at least one of the position and orientation of the light source 30 has been changed, in step S140, the learning image generation unit 140 starts a learning image generation process.

[0027] As shown in FIG. 3, when the learning image generation process starts, first, in step S141, the learning image generation unit 140 places a three-dimensional model VM that simulates the object RM to be inspected in the virtual space VS shown in FIG. 4. In this embodiment, the learning image generation unit 140 reads the three-dimensional model VM pre-stored in the memory 102 and places the three-dimensional model VM in the virtual space VS. The three-dimensional model VM simulates not only the three-dimensional shape of the object RM to be inspected, but also the material of the object RM to be inspected. The actual space in which the object RM to be inspected is placed may be referred to as the real space.

[0028] Next, in step S142, learning image generation unit 140 changes the position and orientation of virtual light source VL arranged in virtual space VS in accordance with the position and orientation of light source 30 represented in first light source information LD1 acquired in step S120 of the learning model update process. Note that the order of steps S141 and S142 may be reversed.

[0029] In step S143, the training image generation unit 140 generates training images VG representing the three-dimensional model VM illuminated by the virtual light source VL by capturing the three-dimensional model VM placed in the virtual space VS with a virtual camera VC simulating the viewpoint of the camera 40. The training images VG represent the three-dimensional model VM from the same viewpoint as the viewpoint from which the camera 40 captures the image of the object RM to be inspected. The training images VG generated by the training image generation unit 140 are stored in the memory 102.

[0030] In step S144, the learning image generation unit 140 determines whether a predetermined number of learning images VG have been generated. For example, the learning image generation unit 140 determines whether 1000 learning images VG have been generated.

[0031] If it is not determined in step S144 that the predetermined number of training images VG have been generated, in step S145, training image generation unit 140 changes at least one of the position and orientation of three-dimensional model VM in virtual space VS without changing the position and orientation of virtual light source VL, and then returns the process to step S143 to generate training images VG. By repeating the processes from step S143 to step S145 until it is determined in step S144 that the predetermined number of training images VG have been acquired, training image generation unit 140 generates a predetermined number of training images VG in which at least one of the position and orientation of three-dimensional model VM differs from each other.

[0032] If it is determined in step S144 that the predetermined number of training images VG have been acquired, the training image generation unit 140 determines in step S146 whether training images VG have been generated for all three-dimensional models VM. If it is not determined in step S146 that training images VG have been generated for all three-dimensional models VM, the training image generation unit 140 switches the three-dimensional model VM to be placed in the virtual space VS in step S147. In this embodiment, multiple types of three-dimensional models VM are pre-stored in the memory 102. Each three-dimensional model VM simulates multiple types of test objects RM, i.e., multiple types of bumpers with different grades and sales regions. The training image generation unit 140 switches the three-dimensional model VM placed in the virtual space VS by loading a different type of three-dimensional model VM from the three-dimensional model VM for which the predetermined number of training images VG have been generated, and then returns to step S143 to generate training images VG.

[0033] The training image generation unit 140 generates a predetermined number of training images VG for each 3D model VM by repeating the processes from step S143 to step S146 until it is determined in step S146 that training images VG have been generated for all 3D models VM. If it is determined in step S146 that training images VG have been generated for all 3D models VM, the training image generation unit 140 terminates the training image generation process. Note that the training image generation process may also be referred to as the training image generation process of the method for updating the training model GM for image inspection, or the training image generation process of the image inspection method. The training images VG generated by the training image generation process may also be referred to as second training images VG2.

[0034] As shown in FIG. 3, after the training image generation process is completed, in step S150 of the training model update process, the training model generation unit 150 generates a training dataset by assigning labels representing grades and sales regions to the training images VG. In step S160, the training model generation unit 150 reads the training dataset and performs machine learning to generate a training model GM, thereby updating the training model GM stored in memory 102. In this embodiment, the training model generation unit 150 generates the training model GM by performing machine learning of a convolutional neural network. Thereafter, the information processing device 50 terminates the training model update process. Note that step S160 may also be referred to as a training model generation step of a method for updating a training model GM for image inspection, or a learning model generation step of an image inspection method.

[0035] According to the image inspection system 10 of the present embodiment described above, when at least one of the position and orientation of the light source 30 in real space is changed, the information processing device 50 uses a three-dimensional model VM illuminated by a virtual light source VL positioned at a position and orientation corresponding to the position and orientation of the light source 30 in real space to generate multiple training images VG in which at least one of the positions and orientations of the three-dimensional model VM differ from each other, and updates the training model GM using the multiple training images VG. Therefore, even when at least one of the position and orientation of the light source 30 in real space is changed, the accuracy of the image inspection of the object RM can be ensured. In particular, in the present embodiment, the training model GM is updated using multiple training images VG in which at least one of the positions and orientations of the three-dimensional model VM differ from each other. Therefore, the accuracy of the image inspection of the object RM can be ensured even when the position or orientation of the object RM placed on the stage 20 is shifted from a reference position or orientation. Furthermore, in this embodiment, an image generated using the three-dimensional model VM is used as the training image VG, so that the training image VG can be collected in a shorter time than in a configuration in which the captured image RG obtained by capturing an image of the subject RM using the camera 40 is used as the training image VG.

[0036] Furthermore, in this embodiment, the information processing device 50 includes an image inspection unit 160 that performs image inspection of the object RM to be inspected using a captured image RG obtained by imaging the object RM after the learning model GM has been updated by the learning model generation unit 150 and the learning model GM updated by the learning model generation unit 150. Therefore, the information processing device 50 can perform image inspection of the object RM to be inspected using the updated learning model GM.

[0037] Furthermore, in this embodiment, the information processing device 50 starts a learning model update process when the misjudgment rate of the image inspection by the image inspection unit 160 exceeds a predetermined value. Therefore, when the misjudgment rate of the image inspection increases, the learning model GM is automatically updated to reduce the misjudgment rate of the image inspection. In particular, in this embodiment, even if the inspector is unaware that the position or orientation of the light source 30 has changed or that the misjudgment rate of the image inspection has increased, the learning model GM can be automatically updated to reduce the misjudgment rate of the image inspection.

[0038] B. Other Embodiments: (B1) In the image inspection system 10 of the above-described embodiment, the camera controller 45 has a function of receiving an output signal from the camera 40 and generating a captured image RG, and a function of analyzing the captured image RG to estimate the position and orientation of the light source 30 and generate light source information LD. In contrast, the information processing device 50 may have at least some of the functions of the camera controller 45 described above. For example, the image inspection system 10 may not be provided with the camera controller 45, and the information processing device 50 may have all of the functions of the camera controller 45 described above. Alternatively, in the image inspection system 10, the camera controller 45 may receive an output signal from the camera 40 and generate a captured image RG, and the light source information acquisition unit 120 of the information processing device 50 may analyze the captured image RG transmitted from the camera controller 45 to estimate the position and orientation of the light source 30 and generate the light source information LD.

[0039] (B2) In the image inspection system 10 of the embodiment described above, the second light source information LD2 is stored in the memory 102 of the information processing device 50. In contrast, the second light source information LD2 does not have to be stored in the memory 102 of the information processing device 50. In this case, for example, the light source information acquisition unit 120 may be configured to have a function of estimating the position and orientation of the light source 30 by analyzing the captured image RG and generating the light source information LD, and the light source information acquisition unit 120 may generate the second light source information LD2 by reading from the memory 102 and analyzing the captured image RG that was captured before the position or orientation of the light source 30 was changed.

[0040] (B3) In the image inspection system 10 of each of the above-described embodiments, the information processing device 50 is provided with an image inspection unit 160. In contrast, the image inspection unit 160 may not be provided in the information processing device 50, but may be provided in, for example, the camera controller 45. In this case, an image inspection of the object to be inspected RM may be performed using a captured image RG captured using the camera 40 and a learning model GM acquired from the information processing device 50.

[0041] (B4) In the image inspection system 10 of each of the above-described embodiments, when the misclassification rate represented in the inspection results RD of the image inspection stored in the memory 102 of the information processing device 50 exceeds a predetermined value, the information processing device 50 automatically starts the learning model update process. In contrast, the information processing device 50 does not have to automatically start the learning model update process. For example, during regular inspection of the image inspection system 10, the information processing device 50 may start the learning model update process when a predetermined start operation is performed on the information processing device 50 by an operator performing the regular inspection.

[0042] (B5) In the image inspection system 10 of each of the above-described embodiments, the training image generation unit 140 generates training images VG using multiple types of three-dimensional models VM in the training image generation process. In contrast, the training image generation unit 140 may generate training images VG using only one type of three-dimensional model VM in the training image generation process.

[0043] (B6) In the image inspection system 10 of each of the above-described embodiments, the first light source information LD1 and the second light source information LD2 represent information about the position and orientation of the light source 30. In contrast, the first light source information LD1 and the second light source information LD2 do not necessarily represent information about the orientation of the light source 30. In this case, if the position of the light source 30 represented in the first light source information LD1 is different from the position of the light source 30 represented in the second light source information LD2, the training image generation unit 140 may change the position of the virtual light source VL according to the first light source information LD1 and generate training images VG representing the three-dimensional model VM illuminated by the virtual light source VL whose position has been changed. Furthermore, the first light source information LD1 and the second light source information LD2 do not necessarily represent information about the position of the light source 30. In this case, if the orientation of the light source 30 represented in the first light source information LD1 differs from the orientation of the light source 30 represented in the second light source information LD2, the training image generation unit 140 may change the orientation of the virtual light source VL according to the first light source information LD1 and generate a training image VG representing the three-dimensional model VM illuminated by the virtual light source VL whose orientation has been changed.

[0044] The present disclosure is not limited to the above-described embodiments and can be realized in various configurations without departing from the spirit thereof. For example, the technical features in the embodiments corresponding to the technical features in each aspect described in the Summary of the Invention section can be appropriately replaced or combined to solve some or all of the above-described problems or achieve some or all of the above-described effects. Furthermore, if a technical feature is not described as essential in this specification, it can be appropriately deleted. [Explanation of symbols]

[0045] 10...image inspection system, 20...stage, 30...light source, 40...camera, 45...camera controller, 50...information processing device, 60...display device, 101...CPU, 102...memory, 103...input / output interface, 110...captured image acquisition unit, 120...light source information acquisition unit, 130...light source information comparison unit, 140...learning image generation unit, 150...learning model generation unit, 160...image inspection unit

Claims

1. An apparatus for updating a learning model for image inspection, comprising: a light source information acquisition unit that acquires first light source information representing information on one or both of a position and an orientation of a light source in a real space, the first light source information being acquired using an image obtained by imaging a real space in which an object to be inspected is placed; a light source information comparison unit that determines whether second light source information representing information on one or both of the position and orientation of the light source of the first learning image used in learning the learning model differs from the first light source information; a training image generation unit that, when the light source information comparison unit determines that the first light source information and the second light source information are different, places a three-dimensional model that simulates the object under test and a virtual light source that simulates the light source in accordance with the first light source information in a virtual space, and generates a plurality of second training images that represent the three-dimensional model while changing one or both of the position and orientation of the three-dimensional model in the virtual space; a learning model generation unit that updates the learning model using the plurality of second learning images generated by the learning image generation unit; An apparatus comprising:

2. 10. The apparatus of claim 1, An apparatus comprising an image inspection unit that performs image inspection of the object to be inspected using an image obtained by imaging the object to be inspected after the learning model generation unit has updated the learning model and the learning model updated by the learning model generation unit.

3. 3. The apparatus of claim 2, The light source information comparison unit determines whether the first light source information and the second light source information are different when the error rate of image inspection by the image inspection unit exceeds a predetermined value.

4. 4. The device according to claim 1, further comprising: a storage unit that stores the first learning image; The light source information acquisition unit acquires the second light source information by analyzing the first learning image stored in the storage unit.

5. 1. A method for updating a learning model for image inspection, comprising: a light source information acquisition step of acquiring first light source information representing information on one or both of the position and orientation of a light source in a real space, the first light source information being acquired using an image obtained by imaging a real space in which an object to be inspected is placed; a light source information comparison step of determining whether second light source information representing information on one or both of the position and orientation of the light source of the first learning image used in learning the learning model is different from the first light source information; a training image generating step of, when it is determined in the light source information comparing step that the first light source information and the second light source information are different, arranging a three-dimensional model simulating the object under test and a virtual light source simulating the light source in accordance with the first light source information in a virtual space, and generating a plurality of second training images in which the three-dimensional model is represented while varying one or both of the position and orientation of the three-dimensional model in the virtual space; a learning model generation step of updating the learning model using the plurality of second learning images generated in the learning image generation step; A method comprising:

Images