Vehicle Inspection Device

Abstract

To accurately perform vehicle inspection processing.SOLUTION: A PLC 21 selects, from a plurality of learned models 33 that have learned a picked-up image indicating a normal state of an inspection point for every vehicle 10 and every inspection point of the vehicle 10, a learned model 33 to be used according to the vehicle 10 and inspection point, inputs the picked-up image at the inspection of the inspection point to the learned model 33, and executes inspection processing on the vehicle 10 on the basis of a result output from the learned model 33.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present disclosure relates to a vehicle inspection device. [Background technology]

[0002] BACKGROUND ART Conventionally, a vehicle inspection device is known that determines whether or not an item has been left behind by analyzing an image captured by a vehicle interior camera mounted on a vehicle door mirror (for example, Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent Publication No. 2021-46148 Summary of the Invention [Problem to be solved by the invention]

[0004] However, with the technology of Patent Document 1, there is a possibility that the accuracy will decrease if inspections are performed using a common standard for multiple vehicles and multiple inspection locations.

[0005] In view of the above, the technology disclosed herein aims to perform vehicle inspection processing with high accuracy. [Means for solving the problem]

[0006] In order to solve the above problem, according to one aspect of the present disclosure, A vehicle inspection device is provided that selects a trained model to be used for each vehicle and inspection location on the vehicle from among a plurality of trained models that have been trained using captured images showing the normal state of the inspection location, inputs captured images during inspection of the inspection location into the trained model, and performs inspection processing for the vehicle based on the results output from the trained model. [Effects of the Invention]

[0007] According to one aspect of the present disclosure, vehicle inspection processing can be performed with high accuracy. [Brief explanation of the drawings]

[0008] [Figure 1] 1A is a schematic configuration diagram of a vehicle inspection system according to an embodiment, and FIG. 1B is a control configuration diagram thereof. [Figure 2] A diagram showing the configuration of a learning model (a), the configuration of a trained model (b), and the vehicle inspection process (c) in one embodiment. [Figure 3] 10 is a flowchart illustrating a procedure for a vehicle inspection process according to an embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0009] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings. In each drawing, the same components are given the same reference numerals, and duplicated descriptions will be omitted as appropriate.

[0010] (Vehicle inspection system configuration) 1(a) is a schematic diagram of a vehicle inspection system 1 according to one embodiment. The vehicle inspection system 1 includes a vehicle 10 and a robot 20 that inspects the vehicle 10. The vehicle 10 and the robot 20 are configured to communicate directly, but may also be configured to communicate indirectly via an information processing device such as a server installed in an information processing center or the like.

[0011] Examples of the vehicle 10 include an engine vehicle, a hybrid vehicle, and an electric vehicle. An engine vehicle is equipped with only an engine (internal combustion engine) as a prime mover, a hybrid vehicle is equipped with an engine and an electric motor as a prime mover, and an electric vehicle is equipped with only an electric motor as a prime mover.

[0012] Examples of engine vehicles include gasoline engine vehicles and diesel engine vehicles. Examples of hybrid vehicles include HEVs (Hybrid Electric Vehicles), PHEVs (Plug-in Hybrid Electric Vehicles), and range extender EVs. Examples of electric vehicles include BEVs (Battery Electric Vehicles) and fuel cell vehicles.

[0013] The robot 20 is a service robot that inspects the vehicle 10. Inspection of the vehicle 10 includes checking for items left inside the vehicle, cleaning the interior or exterior of the vehicle, and inspecting for damage inside or outside the vehicle. The robot 20 is a mobile robot that can move within a workspace, but may also be a fixed robot installed at a fixed point. Examples of mobile robots include mobile, walking, and flying robots, while examples of fixed robots include robot arm types and humanoid types. The robot 20 is a mobile robot arm, but may also be a flying robot arm such as a drone.

[0014] Preferably, the robot 20 is an autonomous mobile robot. The autonomous mobile robot has, for example, a SLAM (Simultaneous Localization and Mapping) function, and generates a map using sensors that measure the surrounding environment while estimating its own position, detects a movement route, and moves autonomously. For example, when a rental car serving as the vehicle 10 is brought into a return location, the robot 20 recognizes the vehicle 10 and moves autonomously to the vicinity of the vehicle 10, and also moves autonomously to an imaging position to image an inspection location in response to a vehicle inspection request from the vehicle 10.

[0015] The robot 20 includes, for example, a robot arm 20a and a traveling unit 20b to which the robot arm 20a is attached. The robot arm 20a is, for example, an articulated robot, and each joint is provided with at least a servo motor, etc. The traveling unit 20b is, for example, provided with at least a wheeled, crawler, or legged traveling mechanism and a servo motor, etc.

[0016] (Vehicle inspection system control configuration) 1(b) is a control configuration diagram of a vehicle inspection system 1 according to one embodiment. A vehicle 10 is provided with at least a control device such as an ECU 11 (Electronic Control Unit).

[0017] The ECU 11 is mainly composed of a microcomputer. The ECU 11 includes, for example, a CPU (Central Processing Unit), a ROM (Read Only Memory), a RAM (Random Access Memory), an input circuit, and an output circuit. The ECU 11 performs various processes by, for example, loading a program stored in the ROM into the RAM and executing the program with the CPU. The ROM may be, for example, a rewritable nonvolatile memory such as a flash memory.

[0018] Various detection signals are input to the ECU 11 from the sensor 12. Various command signals are output from the ECU 11 to the in-vehicle actuator 13. The ECU 11 also causes the in-vehicle communication device 14 to transmit various pieces of information, and detects reception of various pieces of information by the in-vehicle communication device 14. The ECU 11 may be configured with multiple ECUs that can communicate via a CAN (Controller Area Network).

[0019] Examples of sensors 12 include a load sensor that detects seat occupancy, a seat belt sensor that detects whether a seat belt is fastened, a courtesy switch that detects whether a door is opened or closed, an ETC (electronic toll collection) card detection sensor, and a proximity sensor that detects the surrounding environment.

[0020] The various detection signals from the sensor 12 include, for example, a seat occupancy detection signal, a seat belt fastening detection signal, a door opening / closing detection signal, an ETC card detection signal, or a surrounding environment approach detection signal. When the various detection signals from the sensor 12 are input to the ECU 11, the ECU 11 records the usage history of the vehicle 10, such as the seat occupancy history, seat belt fastening history, door opening / closing history, ETC card usage history, or surrounding environment approach history, in a non-volatile memory.

[0021] Examples of the in-vehicle actuator 13 include a door actuator that opens and closes a door, a door lock actuator that locks and unlocks a door, and a door actuator that opens and closes a door such as a fuel filler or charging port.

[0022] The various command signals to the in-vehicle actuator 13 include, for example, a door opening / closing command signal, a door locking / unlocking command signal, a door opening / closing command signal, etc. When the various command signals from the ECU 11 are output to the in-vehicle actuator 13, the ECU 11 records the usage history of the vehicle 10, such as the door opening / closing history, the door locking / unlocking history, or the door opening / closing history, in a non-volatile memory.

[0023] The usage history of the vehicle 10 is used to identify areas to be inspected for lost items, areas to be inspected for cleaning, areas to be inspected for damage, etc.

[0024] The ECU 11 causes the in-vehicle communication device 14 to transmit vehicle information to the robot 20. The vehicle information includes the ID (identification information) of the vehicle 10 and the usage history of the vehicle 10. The ID of the vehicle 10 may be, for example, a VIN (Vehicle Identification Number), a model code, or an identification number of the vehicle 10 set by a business company. Examples of business companies include rental car companies, shared car companies, bus companies, taxi companies, and vehicle sales companies.

[0025] Examples of the in-vehicle communication device 14 include a DCM (Data Communication Module), a wireless LAN (Local Area Network) communication module, a short-range wireless communication module, and a combination thereof. A Wi-Fi (registered trademark) module is an example of a wireless LAN communication module, and a Bluetooth (registered trademark) module is an example of a short-range wireless communication module.

[0026] The robot 20 is provided with at least a control device such as a PLC 21 (Programmable Logic Controller). The PLC 21 is an example of a vehicle inspection device, and executes inspection processing of the vehicle 10. Note that the inspection processing of the vehicle 10 may be executed not by the PLC 21 of the robot 20, but by another control device such as an ECU 11 mounted on the vehicle 10, a processor mounted on a server, or a processor mounted on a terminal carried by a user, manager, or owner of the vehicle 10.

[0027] The PLC 21 includes, for example, a CPU, a ROM, a RAM, an input circuit, and an output circuit. The PLC 21 executes various processes by, for example, loading a program stored in the ROM into the RAM and executing the program with the CPU. The ROM may be, for example, a rewritable nonvolatile memory such as a flash memory.

[0028] The PLC 21 causes the robot communication device 24 to transmit various types of information, and detects reception of various types of information by the robot communication device 24. Images captured by the camera 22 are input to the PLC 21. Various command signals to the robot actuator 23 are output from the PLC 21 to the robot actuator 23.

[0029] When the PLC 21 detects that the robot communication device 24 has received the vehicle information, the PLC 21 identifies the inspection points of the vehicle 10 based on the ID of the vehicle 10 and the usage history of the vehicle 10 .

[0030] For example, if the occupant is seated in the driver's seat or fastening the seat belt, the PLC 21 identifies the driver's seat door pocket, center console pocket, etc. Similarly, if the occupant is seated in the passenger seat or fastening the seat belt, the PLC 21 identifies the passenger's seat door pocket or glove pocket, etc., as the inspection location for left behind items.

[0031] For example, if the door opening / closing history or the door locking / unlocking history is the trunk, the PLC 21 identifies, for example, the trunk or trunk pocket as the inspection location for lost items.

[0032] For example, if there is a history of ETC card usage, the PLC 21 identifies, for example, the ETC card on-board unit as the inspection location for lost items.

[0033] For example, if the surrounding environment approach history is the front bumper or the rear bumper, the PLC 21 identifies the front bumper or the rear bumper as the location to be inspected for damage. Similarly, if there is a door opening / closing history, the PLC 21 identifies a door, such as a fuel filler or charging port, as the location to be inspected for damage.

[0034] After identifying the inspection location of the vehicle 10, the PLC21 selects a trained model to use from among a plurality of trained models that have been trained using captured images showing the normal state of the inspection location, based on the ID of the vehicle 10 and the inspection location of the vehicle 10.

[0035] Furthermore, when the inspection location of the vehicle 10 is inside the vehicle, the PLC 21 causes the robot communication device 24 to transmit a window-open command signal to the vehicle 10 in order to allow the camera 22 to enter the vehicle. When the ECU 11 detects that the in-vehicle communication device 14 has received the window-open command signal, it outputs the window-open command signal to the in-vehicle actuator 13 to cause the in-vehicle actuator 13 to open the window, and causes the in-vehicle communication device 14 to transmit a window-open completion signal to the robot 20.

[0036] When the PLC 21 detects reception of the window opening completion signal from the robot communication device 24, it outputs an operation command signal for the robot arm 20a to the robot actuator 23 to cause the camera 22 attached to the tip of the robot arm 20a to enter the vehicle interior. Then, the PLC 21 moves the camera 22 to an imaging position where it will image the inspection location.

[0037] Examples of the robot actuator 23 include a servo motor and a drive circuit that drives each wheel of the traveling unit 20b, and a servo motor and a drive circuit that drives each axis of the robot arm 20a.

[0038] Examples of the camera 22 include a two-dimensional camera and a three-dimensional camera. If the robot 20 is an autonomous mobile robot, a three-dimensional camera such as a LiDAR (Light Detection And Ranging) or a stereo camera is preferred. The camera 22 can use visible light, infrared light, ultrasonic waves, radio waves, X-rays, etc.

[0039] The camera 22 is attached to the tip of the robot arm 20a, for example, but may be installed on a movable device different from the robot 20, or may be provided on the vehicle 10. The camera 22 is an example of an imaging device.

[0040] The PLC 21 causes the camera 22 to capture an image of an inspection location of the vehicle 10. The PLC 21 inputs the captured image of the inspection location from the camera 22. The PLC 21 inputs the captured image at the time of inspection of the inspection location into a trained model, and executes inspection processing of the vehicle 10, such as inspection for lost items, cleaning, or damage, based on the results output from the trained model.

[0041] For example, if an item is left behind in the vehicle 10, the PLC 21 causes the robot communication device 24 to transmit inspection information including information about the item left behind to a terminal of the user, manager, or owner of the vehicle 10. The information about the item left behind preferably includes a captured image of the item.

[0042] For example, if dirt or stains remain on the vehicle 10, the PLC 21 causes the robot communication device 24 to transmit inspection information including cleaning information to a terminal of the user, manager, or owner of the vehicle 10. The cleaning information preferably includes a captured image of the dirt or stains.

[0043] For example, if there is damage or missing parts in the interior or exterior of the vehicle 10, the PLC 21 causes the robot communication device 24 to transmit inspection information including the damage information to a terminal of the user, manager, or owner of the vehicle 10. The damage information preferably includes captured images of the damage or missing parts.

[0044] The PLC 21 may output the inspection information to another output device such as a display, a speaker, or a printer, instead of the robot communication device 24.

[0045] (Learning model configuration) Next, we will explain the configuration of a learning model that learns the normal state of an inspection point of the vehicle 10. Fig. 2(a) is a diagram showing the configuration of a learning model 30 according to one embodiment.

[0046] Examples of the learning model 30 include a support vector machine, a logistic regression, a decision tree (classification tree), a random forest, and a neural network. Learning methods include machine learning such as supervised learning, unsupervised learning, semi-supervised learning, and reinforcement learning. The learning model 30 is configured as an arithmetic circuit and a memory, or as a computer-readable program.

[0047] In the following, unsupervised learning using a neural network, particularly an autoencoder, will be described as an example of the learning model 30. The autoencoder includes an encoder 30a that receives an input image 31 and a decoder 30b that outputs an output image 32.

[0048] The encoder 30a reduces the dimension of the input image 31 and outputs a low-dimensional image (for example, an image of 5×5 pixels) that includes the feature amounts (edges, texture, etc.) of the input image 31. The decoder 30b outputs an output image 32 that has been restored to the same dimension as the input image 31 from the low-dimensional image that includes the feature amounts.

[0049] That is, the autoencoder is a convolutional autoencoder, with the encoder 30a having the functions of dimensionality reduction and feature extraction, and the decoder 30b having the function of image generation, restoring an output image 32 that is identical or similar to an input image 31 from low-dimensional features. Both the encoder 30a and the decoder 30b have multiple layers each having one or more nodes and edges connecting the nodes in each layer, with each edge having its own weight.

[0050] The autoencoder adjusts the weight of each edge in the decoder 30b and the encoder 30a so that the output image 32 matches the input image 31. For example, backpropagation (error backpropagation) is used to adjust the weights (i.e., learning).

[0051] The learning process of the autoencoder may be executed by the PLC 21 mounted on the robot 20, but is preferably executed by a processor such as a GPU (Graphics Processing Unit) mounted on a server. By distributing the learning process in the learning model 30 and the inference process in the trained model 33, the load or battery capacity of the PLC 21 can be reduced.

[0052] The flow of the learning process using the learning model 30 is, for example, as follows. First, the PLC 21 outputs an operation command signal to the robot actuator 23 to move the camera 22 attached to the tip of the robot arm 20a to an image capturing position of the inspection point. The PLC 21 outputs an image capturing command signal to the camera 22 to capture an image of the inspection point in a normal state. After acquiring captured images of all the inspection points, the PLC 21 causes the robot communication device 24 to transmit learning information including the ID of the vehicle 10 and the captured images of the inspection points of the vehicle 10 to the server.

[0053] The processor of the server prepares a learning model 30 for each ID of the vehicle 10 and for each inspection location of the vehicle 10, and inputs a captured image showing the normal state of the inspection location into the learning model 30 as an input image 31. The processor adjusts the weight of each edge of the learning model 30 so that the output image 32 matches the input image 31, thereby generating a trained model 33 that outputs a captured image showing the normal state of the inspection location. The processor generates multiple trained models 33 corresponding to the ID of the vehicle 10 and the inspection locations of the vehicle 10, and transmits them to the PLC 21.

[0054] The PLC 21 associates a plurality of trained models 33 with the ID of the vehicle 10 and inspection locations of the vehicle 10 and stores them in advance in non-volatile memory, thereby completing preparation for the inspection process of the vehicle 10. Note that the plurality of trained models 33 may be stored in non-volatile memory on the server side, the vehicle side, or the terminal side, instead of the non-volatile memory of the robot 20. In response to a vehicle inspection request, the PLC 21 can acquire the trained model 33 corresponding to the ID of the vehicle 10 and the inspection locations of the vehicle 10 from the server, the vehicle, or the terminal.

[0055] (Configuration of trained model) 2(b) is a diagram showing the configuration of a trained model 33 according to one embodiment. The trained model 33 is a model in which the weights of each edge of the training model 30 have been adjusted and training has converged or been completed. Like the training model 30, the trained model 33 is configured as an arithmetic circuit and memory, or as a computer-readable program.

[0056] When the trained model 33 receives an input image 31 of an inspection point of the vehicle 10, it outputs an output image 32 that is identical to or similar to the captured image showing the normal state of the inspection point. Therefore, even if a lost item 34 is captured in the input image 31, for example, the trained model 33 does not extract the lost item 34 as a feature, and therefore an abnormal part such as the lost item 34 is not restored in the output image 32. In other words, the autoencoder serving as the trained model 33 always outputs, as the output image 32, an output image 32 that is identical to or similar to the captured image showing the normal state of the inspection point.

[0057] (Vehicle inspection processing) 2(c) is a diagram showing a vehicle inspection process according to one embodiment. For example, even if a left-behind item 34 is captured in input image 31, trained model 33 outputs output image 32 that is identical to or similar to a captured image showing the normal state of the inspection location. Therefore, a difference image 35 is generated by subtracting input image 31 capturing the inspection location from output image 32 showing the inspection location in a normal state. If a foreign object is detected in difference image 35, it can be determined that a left-behind item 34 is present.

[0058] In other vehicle inspection processes such as cleaning inspection or damage inspection, foreign objects such as dirt, stains, damage, or missing parts can be detected by performing a similar inspection process. The vehicle inspection process is executed by the PLC 21 of the robot 20, but may also be executed by the ECU 11 of the vehicle 10, or by a processor of a server, or by a terminal carried by the user, manager, or owner of the vehicle 10.

[0059] The flow of the inference process using the trained model 33 is, for example, as follows: When an inspection request is received from the vehicle 10, the PLC 21 reads from the non-volatile memory and selects the trained model 33 to be used according to the ID of the vehicle 10 and the inspection location of the vehicle 10. The PLC 21 outputs an operation command signal to the robot actuator 23 to move the camera 22 attached to the tip of the robot arm 20a to the inspection location, and outputs an image capturing command signal to the camera 22 to capture an input image 31 of the inspection location.

[0060] The PLC 21 inputs an input image 31 taken at the time of inspection of an inspection location into a trained model 33, and acquires an output image 32 output from the trained model 33. The PLC 21 generates a differential image 35 by subtracting the input image 31 taken at the time of inspection at the inspection location from the output image 32 output from the trained model 33, and performs a labeling process on the differential image 35.

[0061] The PLC 21 detects foreign objects based on various parameters such as the number of pixels, perimeter, major axis, minor axis, center of gravity position, or a combination of these, of the object labeled in the differential image 35. The threshold values ​​for the various parameters are determined in advance depending on the type of inspection, such as a lost item inspection, cleaning inspection, or damage inspection.

[0062] If the PLC 21 determines that the foreign object is a lost item 34, it outputs inspection information including the lost item information. For example, the PLC 21 causes the robot communication device 24 to transmit the inspection information to a terminal of the user, manager, or owner of the vehicle 10. Note that the PLC 21 may output the inspection information to another output device such as a display, speaker, or printer, instead of the robot communication device 24.

[0063] In addition, other trained models such as a neural network that outputs whether the inspection point of the vehicle 10 is in a normal state or an abnormal state may be used as the trained model 33.

[0064] (Vehicle inspection procedure) Next, the procedure for the vehicle inspection process will be described. Fig. 3 is a flowchart showing the procedure for the vehicle inspection process according to one embodiment. The following describes an example of the procedure for the left-behind item inspection process when returning a rental car, which is an example of the vehicle 10.

[0065] <Step S10> The ECU 11 detects the arrival of the vehicle 10 at the return location based on, for example, travel route information from a navigation system or measurement information from a GPS (Global Positioning System) receiver.

[0066] <Step S11> The ECU 11 detects that there is no occupant in the vehicle 10 based on the seat occupancy detection signal from the sensor 12. When the ECU 11 detects the presence of an occupant (NO in step S11), the ECU 11 repeats the detection of the presence of an occupant until there is no occupant.

[0067] <Step S12> When the ECU 11 detects the absence of an occupant (YES in step S11), it causes the in-vehicle communication device 14 to transmit the vehicle information together with a vehicle inspection request to the robot 20. The vehicle information includes the ID of the vehicle 10 and the usage history of the vehicle 10.

[0068] <Step S13> The PLC 21 detects the reception of the vehicle information by the robot communication device 24 .

[0069] <Step S14> The PLC 21 identifies the inspection locations for the left behind items 34 based on the usage history of the vehicle 10. When a cleaning inspection or a damage inspection is performed, the PLC 21 identifies the inspection locations for cleaning or damage based on the usage history of the vehicle 10.

[0070] <Step S15> The PLC 21 selects a trained model 33 to use from among a plurality of trained models 33 that have learned images showing the normal state of the inspection location, based on the ID of the vehicle 10 and the inspection location of the vehicle 10.

[0071] <Step S16> When the camera 22 is to enter the vehicle interior, the PLC 21 causes the robot communication device 24 to transmit a window opening command signal to the vehicle 10.

[0072] <Step S17> The ECU 11 detects the reception of the window opening command signal by the in-vehicle communication device 14 .

[0073] <Step S18> The ECU 11 outputs a window opening command signal to the in-vehicle actuator 13 to cause the in-vehicle actuator 13 to open the window of the vehicle 10 .

[0074] <Step S19> The ECU 11 causes the in-vehicle communication device 14 to transmit a window opening completion signal to the robot 20 .

[0075] <Step S20> The PLC 21 detects the reception of the window opening completion signal by the robot communication device 24 .

[0076] <Step S21> The PLC 21 outputs an operation command signal to the robot actuator 23 to move the robot arm 20a, and causes the camera 22 attached to the tip of the robot arm 20a to enter the vehicle interior and move the camera 22 to a position for capturing an image of the inspection location.

[0077] <Step S22> The PLC 21 outputs an imaging command signal to the camera 22 to capture an image of the inspection location. The PLC 21 inputs an input image 31, which is an image of the inspection location captured at the time of inspection, to the trained model 33, and acquires an output image 32 output from the trained model 33.

[0078] The PLC 21 detects the presence or absence of any left behind items 34 based on a differential image 35 obtained by subtracting an input image 31 of an inspection location from an output image 32 showing the normal state of the inspection location. In the case of a cleaning inspection or damage inspection, the PLC 21 detects the presence or absence of trash, or the presence or absence of damage or missing parts based on the differential image 35.

[0079] <Step S23> The PLC 21 determines whether there is any left-behind property 34. If the PLC 21 determines that there is no left-behind property 34 (NO in step S23), the PLC 21 ends the inspection process for the vehicle 10.

[0080] <Step S24> If the PLC 21 determines that there is a left-behind item 34 (YES in step S23), it causes the robot communication device 24 to notify the left-behind item information to the terminal of the user, manager, or owner of the vehicle 10. In the case of a cleaning inspection or damage inspection, the PLC 21 causes the robot communication device 24 to notify the cleaning information or damage information to the terminal of the user, manager, or owner of the vehicle 10. This completes the inspection process.

[0081] The AI ​​inspection process in steps S14 and S15 may be performed by the ECU 11 of the vehicle 10, or by a processor of a server installed in an information processing center, or by a terminal of the user, administrator, or owner of the vehicle 10.

[0082] Furthermore, when inspecting the outside of the vehicle, such as for cleaning inspection or damage inspection, the processes of opening the windows and inserting the camera 22 into the vehicle in steps S16 to S21 are not required.

[0083] The above flowchart can be applied not only to the inspection process when returning a rental car, but also to the inspection process for lost items, cleaning, or damage to various vehicles 10 such as privately owned cars, shared cars, taxis, buses, sales test drive cars, or railway vehicles.

[0084] (Action and effect) According to the above embodiment, the trained model 33 to be used can be selected depending on the vehicle and the inspection location, thereby enabling inspection processing of the vehicle 10 to be performed with high accuracy.

[0085] Furthermore, while some conventional techniques involve preparing a reference image in advance and performing inspection processing by subtracting the reference image from a comparison image, the above-described embodiment eliminates the need to prepare a reference image. In other words, it is sufficient to prepare a trained model 33 configured as a program or an arithmetic circuit, which significantly reduces the memory space required by the PLC 21.

[0086] The various functions described in the above embodiments can be realized by one or more processing circuits, such as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array) designed to perform the various functions.

[0087] The program may also be provided in a form recorded on a non-transitory recording medium such as a CD-ROM (Compact Disk Read Only Memory), or may be provided in a form downloaded from an external device such as a server via a network.

[0088] In the above embodiments, when the number, quantity, unit, range, etc. of each element are mentioned, they are not limited to those mentioned unless otherwise specified or specified in principle. Furthermore, the structures, etc. described in the embodiments are not necessarily required unless otherwise specified or specified in principle. [Explanation of symbols]

[0089] 10 vehicles 21 PLC 22 Camera 31 input images 32 output images 33 trained models

Claims

[Claim 1] A vehicle inspection device that selects a trained model to be used for the vehicle and the inspection location identified based on the vehicle's usage history from among a plurality of trained models that have been trained for each vehicle and each inspection location on the vehicle using captured images that show the normal state of the inspection location, inputs captured images at the time of inspection of the inspection location into the trained model, and performs inspection processing for the vehicle based on the results output from the trained model.

Images