Environmental condition abnormality detection apparatus, method, program, and environmental condition abnormality detection system

Abstract

To detect an environmental condition abnormality of an imaging apparatus during capturing images used for inference by a trained model.SOLUTION: An environmental condition abnormality detection apparatus according to an embodiment of the present invention, includes: an image feature amount extraction unit for extracting feature amounts from both an image captured by an imaging device during capturing images used for inference by a trained model, and a correct answer image captured under environmental conditions during the training of the trained model; an image feature amount comparison unit for comparing a feature amount of the image captured by the imaging device with a feature amount of the correct answer image; and a control unit for detecting an abnormality in the environmental conditions based on a comparison result between the feature amount of the image captured by the imaging device and the feature amount of the correct answer image to modify the environmental conditions.SELECTED DRAWING: Figure 6

Description

[Technical Field]

[0001] The present invention relates to an apparatus, a method, a program, and a system for detecting an abnormal environmental condition. [Background technology]

[0002] Conventionally, artificial intelligence (AI) solutions using images captured by fixed cameras and other cameras have been known. Specifically, the images captured by the camera are input into a trained model, and various inferences are made based on the output results. Summary of the Invention [Problem to be solved by the invention]

[0003] However, if an object hits the stand on which the camera is installed, or if the mounting fixture loosens over time, the camera's angle of view or the light source may change.If these changes in environmental conditions, such as the camera's angle of view or light source, go unnoticed, images captured under environmental conditions different from those used when the trained model was trained may be input to the trained model, reducing the accuracy of inference.

[0004] Therefore, the present invention aims to prevent a decrease in the accuracy of inference using a trained model. [Means for solving the problem]

[0005] An environmental condition abnormality detection device according to one embodiment of the present invention includes an image feature extraction unit that extracts features from an image taken by an imaging device that is currently capturing an image to be used for inference using a trained model, and a correct image taken under the environmental conditions at the time of training of the trained model; an image feature comparison unit that compares the features of the image taken by the imaging device with the features of the correct image; and a control unit that corrects the environmental conditions when an abnormality in the environmental conditions is detected based on the comparison result between the features of the image taken by the imaging device and the features of the correct image. [Effects of the Invention]

[0006] According to the present invention, it is possible to detect and correct abnormalities in the environmental conditions of an imaging device while capturing images to be used for inference using a trained model. [Brief explanation of the drawings]

[0007] [Figure 1] FIG. 1 is a diagram for explaining an outline (correct image) of the present invention. [Figure 2] 1 is a diagram for explaining an outline of the present invention (when an abnormality is detected in an environmental condition (angle of view of a camera)). FIG. [Figure 3] FIG. 1 is a diagram for explaining an outline of the present invention (when an abnormality is detected in the environmental conditions (light source)). [Figure 4] 1 is a diagram illustrating an overall configuration according to an embodiment of the present invention. [Figure 5] 1 is a hardware configuration diagram of an environmental condition abnormality detection device according to an embodiment of the present invention. [Figure 6] 1 is a functional block diagram of an environmental condition abnormality detection system according to an embodiment of the present invention. [Figure 7] FIG. 4 is a sequence diagram illustrating a process for detecting an abnormality in an environmental condition according to an embodiment of the present invention. [Figure 8] 10 is an example of a notification according to an embodiment of the present invention. [Figure 9] 10 is an example of a notification according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0008] Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

[0009] <Summary> An overview of the present invention will be described with reference to Figures 1 to 3. For example, the present invention can be applied to a system that uses a trained model to estimate the work of a worker in a machine assembly operation and determine whether the worker is assembling the machine in the correct procedure. In this case, a camera is fixed so that it captures the worker's hands, and the trained model is generated by learning using video captured at a specific angle of view.

[0010] The present invention is not limited to the estimation of the work of the workers described above, but can also be applied to security monitoring systems (systems that use fixed cameras to detect abnormalities for the security of important facilities and infrastructure), traffic monitoring systems (systems that monitor road traffic conditions and accidents, and detect traffic violations), monitoring systems (systems that monitor patients and pets in hospitals and notify monitors when an abnormality is detected), disaster monitoring systems (systems that monitor targets in areas where natural disasters may occur, and issue an alert if an abnormality is detected), etc.

[0011] Fig. 1 is a diagram for explaining an overview (correct image) of the present invention. Fig. 1 shows a correct image captured under environmental conditions during training of a trained model. For example, the correct image is assumed to be captured with the camera angle of view and light source (position and direction of the light source) shown in Fig. 1 so that the image captures the hands of a worker on a workbench.

[0012] FIG. 2 is a diagram for explaining an overview of the present invention (when an abnormality is detected in the environmental conditions (camera angle of view)). In FIG. 2, the camera angle of view has changed since the correct image in FIG. 1 was captured. Therefore, when inference is performed using the image in FIG. 2 (i.e., when the image in FIG. 2 is input into a trained model), the accuracy of the inference decreases because the environmental conditions of the image in FIG. 2 and the environmental conditions at the time of training the trained model are different.

[0013] FIG. 3 is a diagram for explaining an overview of the present invention (when an abnormality is detected in the environmental conditions (light source)). In FIG. 3, the light source (the position of the light source and the direction of the light) has changed since the correct image in FIG. 1 was captured. Therefore, when inference is performed using the image in FIG. 3 (i.e., when the image in FIG. 3 is input into a trained model), the accuracy of the inference decreases because the environmental conditions in the image in FIG. 3 and the environmental conditions at the time of training the trained model are different.

[0014] <Overall structure> 4 is an overall configuration diagram according to one embodiment of the present invention. The abnormal environmental condition detection system 1 may include an abnormal environmental condition detection device 10, an inference device 20, an administrator terminal 30, a camera 40, a light source 50, and a device 60. The abnormal environmental condition detection device 10 is communicably connected to the inference device 20, the administrator terminal 30, the camera 40, the light source 50, and the device 60 via an arbitrary network.

[0015] <<Environmental Condition Abnormality Detection Device>> The environmental condition abnormality detection device (e.g., a server) 10 is a device that detects abnormalities in environmental conditions during inference by the inference device 20. Specifically, the environmental condition abnormality detection device 10 detects abnormalities in the environmental conditions based on the results of a comparison between the feature amounts of an image captured by a camera 40 currently capturing an image used for inference using a trained model and the feature amounts of a correct image captured under the environmental conditions at the time of training the trained model. The environmental condition abnormality detection device 10 is one or more computers.

[0016] [Environmental conditions] Here, environmental conditions will be described. Environmental conditions are factors that affect the accuracy of various inferences made based on the results output from an image captured by the camera 40 that is input into a trained model. For example, environmental conditions include at least one of the angle of view of the camera 40 and the light source 50 (e.g., including at least one of the position, light direction, and light intensity of the light source 50).

[0017] <<Inference device>> The inference device 20 is a device that inputs an image captured by the camera 40 into a trained model and performs various inferences based on the output results. The inference device 20 is one or more computers.

[0018] Although the environmental condition abnormality detection device 10 and the inference device 20 are described as separate devices in FIG. 1, the environmental condition abnormality detection device 10 and the inference device 20 may be implemented as a single device.

[0019] <<Administrator terminal>> The administrator terminal 30 is a terminal (e.g., a personal computer, a smartphone, etc.) operated by an administrator (e.g., an administrator of equipment such as the camera 40). For example, when the administrator terminal 30 receives a notification from the environmental condition abnormality detection device 10 that an abnormality in the environmental condition has been detected, the administrator adjusts the camera 40, the light source 50, and the device 60.

[0020] <<Camera (imaging device)>> The camera (imaging device) 40 captures an image of any object and transmits the image (e.g., a video (video) including multiple frames (still images)) to the abnormal environmental condition detection device 10 and the inference device 20. The angle of view of the camera 40 is variable. For example, the camera 40 can change the angle of view according to an instruction from the abnormal environmental condition detection device 10 (e.g., via an API). Note that the camera 40 may have at least some of the functions of the abnormal environmental condition detection device 10 described in this specification.

[0021] <<Light source>> The light source 50 is any light source. At least one of the position, the direction of light, and the amount of light of the light source 50 is variable. For example, the light source 50 can change at least one of the position, the direction of light, and the amount of light of the light source 50 in accordance with an instruction from the environmental condition abnormality detection device 10 (for example, via an API). Note that the light source is not limited to a light source external to the camera 40, and may be a light source internal to the camera 40.

[0022] <<Device>> The device 60 is any device that affects the environmental conditions when the camera 40 is capturing an image. For example, the device 60 can operate according to instructions from the environmental condition anomaly detection apparatus 10 (e.g., via an API).

[0023] The devices described in the examples are merely illustrative of one of several computing environments for implementing the embodiments disclosed herein. In one embodiment, the environmental condition abnormality detection device (e.g., server) 10 includes multiple computing devices, such as a server cluster, configured to communicate with each other via any type of communication link, including a network, shared memory, etc., and to perform the processes disclosed herein.

[0024] <Hardware configuration> 5 is a hardware configuration diagram of an environmental condition abnormality detection device (for example, a server) 10 according to one embodiment of the present invention. The same applies to the inference device 20.

[0025] As shown in FIG. 5, the abnormal environmental condition detection device 10 is constructed by a computer, and as shown in FIG. 5, it is equipped with a CPU 1001, a ROM 1002, a RAM 1003, a HD 1004, an HDD (Hard Disk Drive) controller 1005, a display 1006, an external device connection I / F (Interface) 1007, a network I / F 1008, a data bus 1009, a keyboard 1010, a pointing device 1011, a DVD-RW (Digital Versatile Disk Rewritable) drive 1013, and a media I / F 1015.

[0026] Of these, the CPU 1001 controls the overall operation of the abnormal environmental condition detection device 10. The ROM 1002 stores programs, such as an IPL, used to drive the CPU 1001. The RAM 1003 is used as a work area for the CPU 1001. The HD 1004 stores various data, such as programs. The HDD controller 1005 controls the reading and writing of various data from and to the HD 1004 under the control of the CPU 1001. The display 1006 displays various information, such as a cursor, menu, window, text, or image. The external device connection I / F 1007 is an interface for connecting various external devices. In this case, the external devices are, for example, USB (Universal Serial Bus) memory or a printer. The network I / F 1008 is an interface for data communication using a communication network. The bus line 1009 is an address bus, a data bus, or the like, for electrically connecting the components, such as the CPU 1001, shown in FIG. 5.

[0027] The keyboard 1010 is a type of input means having multiple keys for inputting characters, numbers, various instructions, etc. The pointing device 1011 is a type of input means for selecting and executing various instructions, selecting a processing target, moving a cursor, etc. The DVD-RW drive 1013 controls reading and writing of various data from a DVD-RW 1012, which is an example of a removable recording medium. Note that this is not limited to a DVD-RW, and may be a DVD-R, etc. The media I / F 1015 controls reading and writing (storing) of data from a recording medium 1014, such as a flash memory.

[0028] <Functional configuration> FIG. 6 is a functional block diagram of an abnormal environmental condition detection system 1 according to one embodiment of the present invention.

[0029] The abnormal environmental condition detection device 10 can include a control unit 101, an image feature extraction unit 102, an image feature comparison unit 103, a notification unit 104, a camera control unit 105, a light source control unit 106, a device control unit 107, and a correct image storage unit 108. The abnormal environmental condition detection device 10 can function as the control unit 101, the image feature extraction unit 102, the image feature comparison unit 103, the notification unit 104, the camera control unit 105, the light source control unit 106, and the device control unit 107 by executing a program.

[0030] The inference device 20 can include an inference unit 201 and a model storage unit 202. The inference device 20 can function as the inference unit 201 by executing a program.

[0031] <<Environmental Condition Abnormality Detection Device>> The control unit 101 performs processing to detect abnormalities in environmental conditions.

[0032] Specifically, the control unit 101 acquires an image captured by the camera 40 (i.e., the camera capturing an image to be used for inference using the trained model stored in the trained model storage unit 202) (note that, since it is preferable that no unnecessary objects are captured when extracting features from an image captured by the camera 40, it is preferable that the image be captured before work begins or during a break, etc.). The control unit 101 also acquires the correct image stored in the correct image storage unit 108 (the correct image specified by the administrator terminal 30 in the initial setting).

[0033] The control unit 101 causes the image feature extraction unit 102 to extract the feature amounts of the image captured by the camera 40 and the correct image, and causes the image feature comparison unit 103 to compare the feature amounts of the image captured by the camera 40 with the feature amounts of the correct image.

[0034] When the control unit 101 detects an abnormality in the environmental conditions based on the comparison result between the feature amounts of the image captured by the camera 40 and the feature amounts of the authentic image, the control unit 101 corrects the environmental conditions. Specifically, the control unit 101 determines that an abnormality in the environmental conditions has occurred when the difference between the feature amounts of the image captured by the camera 40 and the feature amounts of the authentic image exceeds a threshold. When an abnormality in the environmental conditions is detected, the control unit 101 causes the notification unit 104 to notify the administrator terminal 30 of the detection of the abnormality in the environmental conditions, and causes the camera control unit 105, the light source control unit 106, and the device control unit 107 to perform control.

[0035] The image feature amount extraction unit 102 extracts feature amounts (for example, feature amounts of the camera's angle of view) of the image captured by the camera 40 and the correct image.

[0036] [Image features] Here, the feature amount of an image will be described. The feature amount of an image is a feature amount that indicates the environmental conditions. For example, the feature amount that indicates the environmental conditions includes at least one of the angle of view of the camera and the light source.

[0037] Any algorithm may be used to extract image features. For example, when detecting an abnormality in the camera's angle of view, SIFT (Scale-Invariant Feature Transform) is used to extract the SIFT features of the image captured by camera 40 and the SIFT features of the correct image.

[0038] The image feature amount comparison unit 103 compares the feature amount of the image captured by the camera 40 (for example, the feature amount of the camera's angle of view) extracted by the image feature amount extraction unit 102 with the feature amount of the correct image (for example, the feature amount of the camera's angle of view). Specifically, the image feature amount comparison unit 103 calculates the difference between the feature amount of the image captured by the camera 40 extracted by the image feature amount extraction unit 102 and the feature amount of the correct image.

[0039] [Detection of abnormalities in camera angle of view and light source] Here, a case will be described in which first, whether or not there is an abnormality in the angle of view of the camera is determined, and if there is no abnormality in the angle of view of the camera, whether or not there is an abnormality in the light source is determined.

[0040] Image feature extraction unit 102 extracts feature amounts of the camera's angle of view, image feature comparison unit 103 compares the feature amounts of the camera's angle of view, and control unit 101 determines whether or not there is an abnormality in the camera's angle of view. If no abnormality in the camera's angle of view is detected, image feature extraction unit 102 extracts feature amounts of the light source, image feature comparison unit 103 compares the feature amounts of the light source, and if control unit 101 detects an abnormality in light source 50, it corrects light source 50.

[0041] The notification unit 104 notifies the administrator terminal 30 that an abnormality in the environmental condition has been detected. For example, when an abnormality in the environmental condition is detected, the notification unit 104 notifies the administrator terminal 30 of a procedure for correcting the environmental condition.

[0042] When an abnormality in the camera 40 (for example, an abnormality (deviation) in the camera's angle of view) is detected, the camera control unit 105 controls the camera 40 (for example, adjusts the camera's angle of view) so that the camera 40 returns to a normal state (for example, the camera's angle of view at the time of learning the trained model stored in the model memory unit 202).

[0043] When an abnormality in the light source 50 (e.g., an abnormality in the position of the light source, the direction of light, the amount of light, etc.) is detected, the light source control unit 106 controls the light source 50 (e.g., adjusts the position of the light source, the direction of light, the amount of light, etc.) so that the light source 50 becomes normal (e.g., the light source at the time of learning the trained model stored in the model storage unit 202).

[0044] The device control unit 107 controls the device 60 so that the environmental conditions are normal (i.e., the environmental conditions at the time of learning of the learned model stored in the model storage unit 202).

[0045] In this way, if the abnormality in the environmental conditions can be automatically corrected, the abnormality in the environmental conditions is automatically corrected, and if the abnormality in the environmental conditions cannot be automatically corrected, the procedure for correcting the environmental conditions is notified to the administrator terminal 30.

[0046] The correct image storage unit 108 stores a correct image specified by the administrator terminal 30. The correct image is an image captured under the environmental conditions at the time of learning of the trained model stored in the model storage unit 202 (for example, the same camera angle of view, light source, etc. as those at the time of learning).

[0047] <<Inference device>> The inference unit 201 acquires an image captured by the camera 40 (for example, a video (video) including multiple frames (still images)), inputs the image into a trained model stored in the model storage unit 202, and performs various inferences based on the output results.

[0048] A trained model is stored in the model storage unit 202. The trained model is a model that has been machine-learned so that when an image captured by the camera 40 is input, arbitrary information is output.

[0049] <Method> FIG. 7 is a sequence diagram showing a process for detecting an abnormality in an environmental condition according to an embodiment of the present invention.

[0050] [Initial Settings] In steps 1 and 2, initial settings are made.

[0051] In step 1 (S1), the administrator terminal 30 transmits to the environmental condition abnormality detection device 10 (control unit 101) information on the initial setting that the administrator inputted to the administrator terminal 30. For example, the information on the initial setting includes the timing of detecting an abnormality (e.g., 9:00 every Monday, 9:00 on the first day of each month, etc.), the designation of the correct image, and the threshold for detecting an abnormality (specifically, the threshold for the difference between the feature amount of the image captured by the camera 40 and the feature amount of the correct image. Note that a threshold may be set for each feature amount (e.g., the angle of view of the camera, the position of the light source, the direction of light, and the amount of light)).

[0052] In step 2 (S2), the control unit 101 of the environmental condition abnormality detection device 10 stores the correct image designated in S1 in the correct image storage unit .

[0053] After the initial setup is complete, inference processing and anomaly detection processing are performed.

[0054] [Inference processing] In step 3 (S3), the administrator terminal 30 instructs the inference device 20 (inference unit 201) to start the inference process.

[0055] In step 4 (S4), the inference unit 201 acquires an image captured by the camera 40 (for example, a video (moving image) including a plurality of frames (still images)).

[0056] In step 5 (S5), the inference unit 201 inputs the image acquired in S4 into the trained model stored in the model storage unit 202, thereby performing inference processing.

[0057] S4 and S5 are repeated.

[0058] [Abnormality detection processing] In step 6 (S6), the administrator terminal 30 instructs the environmental condition abnormality detection device 10 (control unit 101) to start the process of detecting an abnormality.

[0059] In step 7 (S7), the control unit 101 acquires an image captured by the camera 40 (for example, a video (moving image) including a plurality of frames (still images)).

[0060] In step 8 (S8), the control unit 101 acquires the correct image stored in the correct image storage unit 108 (that is, the correct image designated by the administrator terminal 30 in S1 and S2).

[0061] The control unit 101 causes the image feature extraction unit 102 to extract the features of the image taken by the camera 40 in S7 and the correct image in S8, and causes the image feature comparison unit 103 to compare the features of the image taken by the camera 40 in S7 with the features of the correct image in S8.

[0062] In step 9 (S9), the image feature extraction unit 102 extracts feature amounts from the image captured by the camera 40 in S7 and the correct image in S8.

[0063] In step 10 (S10), the image feature amount comparison unit 103 compares the feature amount of the image captured by the camera 40 in S7, extracted in S9, with the feature amount of the correct image in S8.

[0064] In step 11 (S11), the control unit 101 detects an abnormality in the environmental conditions based on the comparison result of S 10. Specifically, the control unit 101 determines that an abnormality in the environmental conditions has occurred when the difference between the feature amount of the image captured by the camera 40 in S7 and the feature amount of the correct image in S8 exceeds a threshold value.

[0065] If an abnormality in the environmental conditions is detected, steps 12 to 15 are executed. The control unit 101 causes the notification unit 104 to notify the administrator terminal 30 of the detection of the abnormality in the environmental conditions, and causes the camera control unit 105, the light source control unit 106, and the device control unit 107 to perform control.

[0066] In step 12 (S12), the notification unit 104 notifies the administrator terminal 30 that an abnormality in the environmental conditions has been detected.

[0067] In step 13 (S13), the camera control unit 105 controls the camera 40 so that the camera 40 is normal (for example, adjusts the angle of view of the camera).

[0068] In step 14 (S14), the light source control unit 106 controls the light source 50 so that the light source 50 is normal (for example, adjusts the position of the light source, the direction of the light, the amount of light, etc.).

[0069] In step 15 (S15), the device control unit 107 controls the device 60 so that the environmental conditions become normal.

[0070] S7 to S15 are repeated.

[0071] <User Interface> 8 is an example of a notification according to one embodiment of the present invention. This figure illustrates a case in which the camera angle of view differs between the image captured by camera 40 and the correct image. In this case, text or audio indicating a correction procedure, such as "The angle of view of the image (right) taken when an abnormality was detected is different from the correct image (left) is. Please rotate the camera 15 degrees to the right to adjust the angle of view," is transmitted from the environmental condition abnormality detection device 10 to the administrator terminal 30.

[0072] 9 is an example of a notification according to an embodiment of the present invention. This figure illustrates a case in which the position of the light source differs between the image captured by the camera 40 and the correct image. In this case, text or audio indicating correction procedures, such as "The position of the light source has changed compared to the correct image (left). Please adjust the position of the light source so that the light from the left is stronger," is transmitted from the abnormal environmental condition detection device 10 to the administrator terminal 30.

[0073] <Effects> Thus, in one embodiment of the present invention, if a camera capturing an image used for inference using a trained model captures the image under environmental conditions that differ from the environmental conditions at the time the trained model was trained, the environmental conditions are corrected, thereby preventing a decrease in the accuracy of the inference.

[0074] Each function of the above-described embodiments can be realized by one or more processing circuits. Here, the term "processing circuit" in this specification includes a processor programmed to perform each function by software, such as a processor implemented by an electronic circuit, as well as devices such as an ASIC (Application Specific Integrated Circuit), a DSP (Digital Signal Processor), an FPGA (Field Programmable Gate Array), or a conventional circuit module designed to perform each function described above. [Explanation of symbols]

[0075] 1. Environmental condition abnormality detection system 10. Environmental condition abnormality detection device 20 Reasoning device 30 Administrator terminal 40 Camera 50 light source 60 devices 101 Control section 102 Image feature extraction unit 103 Image feature comparison unit 104 Notification Department 105 Camera control unit 106 Light source control unit 107 Device control unit 108 Correct image memory section 201 Reasoning Department 202 Model Memory Unit [Prior art documents] [Patent documents]

[0076] [Patent Document 1] Japanese Patent Application Publication No. 2017-138808

Claims

1. an image feature extraction unit that extracts features from each of an image captured by an imaging device capturing an image used for inference using the trained model and a correct image captured under environmental conditions during training of the trained model; an image feature comparison unit that compares a feature of the image captured by the imaging device with a feature of the correct image; a control unit that corrects the environmental conditions when detecting an abnormality in the environmental conditions based on a comparison result between a feature amount of the image captured by the imaging device and a feature amount of the correct image; An environmental condition abnormality detection device equipped with the device.

2. the feature amount is a feature amount indicating the environmental condition, The abnormal environmental condition detection device according to claim 1 , wherein the characteristic amount indicating the environmental condition includes at least one of an angle of view of the imaging device and a light source.

3. the image feature extraction unit extracts a feature of an angle of view of the imaging device; the image feature amount comparison unit compares feature amounts of the angles of view of the imaging devices; The environmental condition abnormality detection device according to claim 1 , wherein the control unit corrects the angle of view of the imaging device when detecting an abnormality in the angle of view of the imaging device.

4. If no abnormality in the angle of view of the imaging device is detected, the image feature extraction unit extracts a feature of a light source; the image feature amount comparison unit compares the feature amounts of the light sources; The environmental condition abnormality detection device according to claim 3 , wherein the control unit corrects the light source when an abnormality in the light source is detected.

5. The environmental condition abnormality detection device according to claim 3 , further comprising an imaging device control unit that adjusts the angle of view of the imaging device when an abnormality in the angle of view of the imaging device is detected.

6. The environmental condition abnormality detection device according to claim 4 , further comprising a light source adjustment unit that adjusts the light source when an abnormality in the light source is detected.

7. The environmental condition abnormality detection device according to claim 1 , further comprising a notification unit that notifies an administrator terminal of a procedure for correcting the environmental condition when an abnormality in the environmental condition is detected.

8. If the abnormality in the environmental condition can be automatically corrected, the abnormality in the environmental condition is automatically corrected; 2. The environmental condition abnormality detection device according to claim 1, wherein, when the abnormality in the environmental condition cannot be automatically corrected, a procedure for correcting the environmental condition is notified to an administrator terminal.

9. A method performed by an environmental condition abnormality detection device, comprising: A step of extracting features from each of an image captured by an imaging device capturing an image used for inference using the trained model and a correct image captured under environmental conditions during training of the trained model; a step of comparing a feature amount of an image captured by the imaging device with a feature amount of the correct image; detecting an abnormality in the environmental conditions based on a comparison result between a feature amount of the image captured by the imaging device and a feature amount of the correct image, and correcting the environmental conditions; A method comprising:

10. Environmental condition abnormality detection device, A process of extracting features from each of an image captured by an imaging device capturing an image used for inference using a trained model and a correct image captured under environmental conditions during training of the trained model; a process of comparing the feature amount of the image captured by the imaging device with the feature amount of the correct image; A process of detecting an abnormality in the environmental conditions based on a comparison result between the feature amount of the image captured by the imaging device and the feature amount of the correct image, and correcting the environmental conditions. A program to execute.

11. An abnormal environmental condition detection system including an abnormal environmental condition detection device and an administrator terminal, The environmental condition abnormality detection device is an image feature extraction unit that extracts features from each of an image captured by an imaging device capturing an image used for inference using the trained model and a correct image captured under environmental conditions during training of the trained model; an image feature comparison unit that compares a feature of the image captured by the imaging device with a feature of the correct image; a control unit that detects an abnormality in the environmental conditions based on a comparison result between a feature amount of the image captured by the imaging device and a feature amount of the correct image; a notification unit that notifies the administrator terminal that an abnormality in the environmental conditions has been detected; An environmental condition abnormality detection system equipped with

Images