WO2023058445A1 - タイヤの状態の推定方法 - Google Patents
タイヤの状態の推定方法 Download PDFInfo
- Publication number
- WO2023058445A1 WO2023058445A1 PCT/JP2022/034916 JP2022034916W WO2023058445A1 WO 2023058445 A1 WO2023058445 A1 WO 2023058445A1 JP 2022034916 W JP2022034916 W JP 2022034916W WO 2023058445 A1 WO2023058445 A1 WO 2023058445A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- machine learning
- learning model
- tire
- tread
- output
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Ceased
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M17/00—Testing of vehicles
- G01M17/007—Wheeled or endless-tracked vehicles
- G01M17/02—Tyres
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/50—Depth or shape recovery
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60C—VEHICLE TYRES; TYRE INFLATION; TYRE CHANGING; CONNECTING VALVES TO INFLATABLE ELASTIC BODIES IN GENERAL; DEVICES OR ARRANGEMENTS RELATED TO TYRES
- B60C11/00—Tyre tread bands; Tread patterns; Anti-skid inserts
- B60C11/24—Wear-indicating arrangements
- B60C11/246—Tread wear monitoring systems
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60C—VEHICLE TYRES; TYRE INFLATION; TYRE CHANGING; CONNECTING VALVES TO INFLATABLE ELASTIC BODIES IN GENERAL; DEVICES OR ARRANGEMENTS RELATED TO TYRES
- B60C19/00—Tyre parts or constructions not otherwise provided for
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M17/00—Testing of vehicles
- G01M17/007—Wheeled or endless-tracked vehicles
- G01M17/02—Tyres
- G01M17/027—Tyres using light, e.g. infrared, ultraviolet or holographic techniques
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G07—CHECKING-DEVICES
- G07C—TIME OR ATTENDANCE REGISTERS; REGISTERING OR INDICATING THE WORKING OF MACHINES; GENERATING RANDOM NUMBERS; VOTING OR LOTTERY APPARATUS; ARRANGEMENTS, SYSTEMS OR APPARATUS FOR CHECKING NOT PROVIDED FOR ELSEWHERE
- G07C5/00—Registering or indicating the working of vehicles
- G07C5/006—Indicating maintenance
-
- G—PHYSICS
- G07—CHECKING-DEVICES
- G07C—TIME OR ATTENDANCE REGISTERS; REGISTERING OR INDICATING THE WORKING OF MACHINES; GENERATING RANDOM NUMBERS; VOTING OR LOTTERY APPARATUS; ARRANGEMENTS, SYSTEMS OR APPARATUS FOR CHECKING NOT PROVIDED FOR ELSEWHERE
- G07C5/00—Registering or indicating the working of vehicles
- G07C5/008—Registering or indicating the working of vehicles communicating information to a remotely located station
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Definitions
- the present invention relates to a tire state estimation method, an estimation device, an estimation system, an estimation program, and a learned model generation method.
- Patent Document 1 discloses a method of estimating a tire tread wear value using a machine learning model.
- the machine learning model collects and uses sensor data obtained from on-board sensors such as steering wheel angle, wheel speed, longitudinal acceleration, brake pressure, and total distance traveled to estimate the tread wear of each tire.
- the tread is not only worn uniformly over its entire surface, but may also wear concentrated in a specific portion or wear in a characteristic pattern along the circumferential direction of the tire.
- Such a phenomenon is also called uneven wear. If uneven wear is left unattended, tires are more likely to be damaged, more likely to slip, and it will be difficult for the vehicle to run comfortably. Therefore, it is important to detect uneven wear at an early stage and take countermeasures such as tire replacement and tire rotation. However, uneven wear may occur when the wear amount of the tread is less than the specified amount, and it may not be discovered by simply estimating the wear amount of the tread. This point is not taken into consideration in Patent Document 1.
- An object of the present disclosure is to provide a method, apparatus, system and program for estimating uneven tire wear as a tire condition, and a method for generating a trained model.
- a tire condition estimation method includes: Acquire an image of the tread of the target tire from the front so that the tread is continuous in a predetermined direction while including both ends of the tread of the target tire. Input the acquired image to the first machine learning model that has been trained. To derive an output from the trained first machine learning model. Note that the output of the learned first machine learning model corresponds to the estimated uneven wear of the target tire.
- the above method may further include determining whether the target tire needs to be replaced based on the output derived from the learned first machine learning model.
- the above method may further include generating a determination result display screen that displays a determination result as to whether the target tire needs to be replaced.
- the method may further comprise inputting the acquired image into a second trained machine learning model different from the first machine learning model and deriving an output from the second trained machine learning model. good.
- the output of the learned second machine learning model corresponds to the estimation result of the groove depth of the tread of the target tire. Further, determining whether the target tire needs to be replaced is based on the output derived from the learned first machine learning model and the output derived from the learned second machine learning model. may be used to determine whether or not the target tire needs to be replaced.
- the estimation result of uneven wear of the target tire is the estimation result of the degree of uneven wear of the tread, and may be represented by an index showing the degree of uneven wear in stages.
- the method may further include generating an estimation result display screen that displays the index corresponding to the output derived from the first machine learning model along with the overall position of the index indicating the degree of uneven wear in stages. .
- the estimation result of uneven wear of the target tire may be an estimation result of the type of uneven wear.
- the method includes obtaining information on the mounting position of the target tire in the vehicle, and recommending a tire based on the obtained mounting position information and the output derived from the first machine learning model. determining a recommended pattern, which is the rotation pattern of the .
- the above method may further include generating a rotation screen displaying the determined recommended pattern.
- the method includes making the data of the judgment result display screen accessible via the network, and generating a two-dimensional code for accessing the data of the judgment result display screen via the network. It may contain further.
- the method includes making the data of the estimation result display screen accessible via the network, and generating a two-dimensional code for accessing the data of the estimation result display screen via the network. It may contain further.
- a tire condition estimation device includes an image acquisition unit, a storage unit, and a derivation unit.
- the image acquisition unit acquires an image of the tread of the target tire including both ends of the tread and photographing the tread from the front such that the tread is continuous in a predetermined direction.
- the storage unit stores a learned machine learning model.
- the derivation unit inputs the acquired image to a trained machine learning model and derives an output from the trained machine learning model. It should be noted that the output of the learned machine learning model corresponds to the estimation result of the uneven wear of the target tire.
- the estimation device may further include a screen generator that generates an estimation result display screen that displays an estimation result corresponding to the output derived from the learned machine learning model.
- a tire condition estimation system may include an estimation device including the screen generation unit, a camera, and a display.
- a camera photographs the tread of the target tire.
- the display displays an estimation result display screen.
- a tire condition estimation program causes a computer to perform the following: Acquire an image of the tread of the target tire from the front so that the tread is continuous in a predetermined direction while including both ends of the tread of the target tire. Input the acquired image to the first machine learning model that has been trained. To derive an output from the trained first machine learning model. Note that the output of the learned first machine learning model corresponds to the estimated uneven wear of the target tire.
- a method of generating a trained model includes: Preparing learning data, which is a data set of correct data and learning images obtained by photographing the tread from the front so that the tread includes both ends of the tire tread and the tread is continuous in a predetermined direction.
- the machine outputs data corresponding to the correct data.
- the correct data is at least one of an index representing the degree of uneven tire wear in the learning image and a label corresponding to the type of uneven tire wear in the learning image.
- a tire condition estimation method includes: Obtaining an image including the tread of the target tire Inputting the obtained image to the trained first machine learning model Deriving the output from the trained first machine learning model Learning the obtained image inputting to a second trained machine learning model different from the first trained machine learning model; deriving an output from the second trained machine learning model; Determining whether the target tire needs to be replaced based on the output derived from the learned first machine learning model and the output derived from the learned second machine learning model.
- the output of the first learned machine learning model corresponds to the estimated uneven wear of the target tire
- the output of the second learned machine learning model corresponds to the depth of the tread groove of the target tire. corresponds to the estimation result of
- a tire condition estimation device includes an image acquisition unit, a storage unit, a derivation unit, and a determination unit.
- the image acquisition unit acquires an image including the tread of the target tire.
- the storage unit stores a trained first machine learning model and a trained second machine learning model different from the trained first machine learning model.
- the derivation unit inputs the acquired image to the learned first machine learning model and inputs the acquired image to the learned second machine learning model, and the learned first machine learning model and the learned second Derive the respective output from the machine learning model.
- the determination unit determines whether the target tire needs to be replaced based on the output derived from the learned first machine learning model and the output derived from the learned second machine learning model. Note that the output of the first learned machine learning model corresponds to the estimated uneven wear of the target tire, and the output of the second learned machine learning model corresponds to the depth of the tread groove of the target tire. corresponds to the estimation result of
- a tire condition estimation program causes a computer to perform the following: Obtaining an image including the tread of the target tire Inputting the obtained image to the trained first machine learning model Transferring the obtained image to the trained first machine learning model different from the trained first machine learning model 2 Input to the machine learning model Deriving output from the first trained machine learning model Deriving output from the second trained machine learning model Output derived from the first trained machine learning model and Determining whether or not the target tire needs to be replaced based on the output derived from the learned second machine learning model.
- the output of the first learned machine learning model corresponds to the estimated uneven wear of the target tire
- the output of the second learned machine learning model corresponds to the depth of the tread groove of the target tire. corresponds to the estimation result of
- a tire condition estimation method includes: Obtaining an image including the tread of the target tire Inputting the obtained image to the trained first machine learning model Deriving an output from the trained first machine learning model. Note that the output of the learned first machine learning model corresponds to the estimated uneven wear of the target tire.
- a tire condition estimation device includes an image acquisition unit, a storage unit, and a derivation unit.
- the image acquisition unit acquires an image including the tread of the target tire.
- the storage unit stores a learned machine learning model.
- the derivation unit inputs the acquired image to a trained machine learning model and derives an output from the trained machine learning model. It should be noted that the output of the learned machine learning model corresponds to the estimation result of the uneven wear of the target tire.
- a tire condition estimation program causes a computer to perform the following: Obtaining an image including the tread of the target tire Inputting the obtained image to the trained first machine learning model Deriving an output from the trained first machine learning model. Note that the output of the learned first machine learning model corresponds to the estimated uneven wear of the target tire.
- a method of generating a trained model includes: Prepare learning data that is a data set of learning images including tire treads and correct data. Adjust the parameters of the machine learning model so that the data is output.
- the correct data is at least one of an index representing the degree of uneven tire wear in the learning image and a label corresponding to the type of uneven tire wear in the learning image.
- FIG. 1 is a diagram showing the overall configuration of a tire state estimation system according to an embodiment
- FIG. FIG. 2 is a block diagram showing the electrical configuration of the estimating device
- FIG. 2 is a block diagram showing an electrical configuration of an information processing terminal on the user side
- 4 is a flow chart showing the flow of processing performed by an information processing terminal on the user side in the estimation system. 4 is a flow chart showing the flow of processing performed by the server device and the estimation device in the estimation system; An example of a user interface screen displayed on an information processing terminal on the user side. Another example of the user interface screen displayed on the user's information processing terminal. An example of a decision table stored in the estimating device.
- An example of a rotation screen generated by the estimator. 4 is a flowchart showing the flow of learning processing; The figure explaining one aspect
- the tire state estimation method, estimation device, estimation system, estimation program, and learned model generation method will be described below.
- the state of the tire includes the state of uneven wear of the tire.
- FIG. 1 is an overall configuration diagram of an estimation system 5 according to an embodiment of the present disclosure.
- the estimation system 5 is a system for estimating the uneven wear and wear amount of the tire T as the state of the tire T, and feeding back the estimated results and recommended actions (tire replacement, tire rotation, etc.) based on the estimated results.
- a main user of the estimation system 5 according to the present embodiment is, but not limited to, a person who provides tire inspection services to vehicle drivers.
- the estimation system 5 includes an estimation device 1 for estimating the state of the tire T, a camera 20 for photographing the tire T, and a display 21 for displaying the state estimation result and the like.
- the camera 20 is a photographing device that generates digital image data of a subject.
- the camera 20 is built in a portable information processing terminal connected to a network such as the Internet, such as a smart phone, tablet, laptop computer, smart watch, and mobile phone with a push button.
- the display 21 is a display device that displays various information including images captured by the camera 20 .
- the display 21 is, for example, a touch panel display, a liquid crystal display, an organic EL display, a liquid crystal display element, a plasma display, or the like provided in an information processing terminal in which the camera 20 is built.
- the camera 20 is a camera built into the smart phone 2 used by the user, and the display 21 is a touch panel display provided in the smart phone 2 .
- the uneven wear and the amount of wear of the tire T are estimated by the first machine learning model 130 and the second machine learning model 131 (Fig. 4).
- the output of the first machine learning model 130 corresponds to the estimation result of uneven wear of the tire T
- the output of the second machine learning model 131 corresponds to the estimation result of the wear amount of the tire T.
- the estimation device 1 comprehensively judges the outputs derived from the first and second machine learning models 130 and 131, and displays at least one of the estimation results that display the estimation results for the tires T and recommended actions for the driver.
- a feedback screen G3 (see FIG. 9A) to be displayed is generated.
- the feedback screen G3 is stored in the server device 4 as web data 7, which will be described later. Thereby, the feedback screen G3 is displayed on the display of the information processing terminal in response to a request from the information processing terminal via the network.
- the user By displaying the feedback screen G3 on the display 21, the user explains the current status of the tire T and recommended countermeasures while showing this to the driver, and proposes services that the user can provide as necessary. be able to.
- the driver has an information processing terminal (smartphone 3 in this embodiment) connected to the network, such as a smartphone, tablet, laptop computer, smart watch, desktop computer, or mobile phone with a push button, the driver can access the web data 7 from these information processing terminals via the network and display the feedback screen G3 on the display 31 of the information processing terminal.
- the driver can check the feedback screen G3 by himself/herself without being restricted by place or time.
- FIG. 2 is a block diagram showing the electrical configuration of the estimation device 1.
- the estimating device 1 is a general-purpose computer as hardware, and is implemented as, for example, a desktop personal computer, a laptop personal computer, a tablet, or a smart phone.
- the estimation device 1 is manufactured by installing a program 132 in a general-purpose computer from a computer-readable storage medium 133 such as a CD-ROM, USB memory, or via a network.
- the program 132 estimates the state of the tire T shown in the image based on the image data 201 (see FIG. 4) sent from the smartphone 2, and determines whether or not some action is required based on the estimation result.
- the estimation device 1 includes a control unit 10, a display unit 11, an input unit 12, a storage unit 13, and a communication unit 14. These units 10 to 14 are connected to each other via a bus line 15 and can communicate with each other.
- the display unit 11 can be composed of a liquid crystal display or the like, and displays the code of the machine learning model, which will be described later, errors in the learning process of the machine learning model, and the like. This display can be used mainly by those who train machine learning models and generate trained machine learning models 130 and 131 .
- the input unit 12 can be composed of a mouse, a keyboard, a touch panel, or the like, and receives operations on the estimation device 1 .
- the storage unit 13 can be composed of a non-volatile memory such as a hard disk and flash memory. In addition to storing the program 132 in the storage unit 13, data including the image data 201 sent from the smartphone 2 is stored as appropriate.
- the storage unit 13 also stores information defining a first machine learning model 130 and a second machine learning model 131 learned in a learning process described later and used in an estimation process described later. Further, the storage unit 13 stores a judgment table 134 for comprehensively judging the estimation results by these machine learning models and determining recommended measures. Details of the determination table 134 will be described later.
- the control unit 10 can be configured with a CPU (Central Processing Unit), GPU (Graphics Processing Unit), ROM, RAM, and the like. By reading and executing the program 132 in the storage unit 13, the control unit 10 virtually operates as the image acquisition unit 10A, the derivation unit 10B, the determination unit 10C, the screen generation unit 10D, and the learning unit 10E.
- the image acquisition unit 10A acquires image data 201 input to the first machine learning model 130 and the second machine learning model 131 .
- the derivation unit 10B inputs the image data 201 to the first machine learning model 130 and the second machine learning model 131, and derives the output from each machine learning model.
- Determination section 10C determines whether tire replacement or tire rotation is necessary based on the output derived by derivation section 10B and determination table 134 .
- the screen generation unit 10D generates a feedback screen that displays the determination results of the determination unit 10C.
- the communication unit 14 functions as a communication interface that performs data communication via a network.
- the learning unit 10E will be described later.
- FIG. 3 is a block diagram showing an electrical configuration of an information processing terminal including a camera 20 and a display 21 that constitute the estimation system 5.
- the information processing terminal is a device for a user to provide a tire inspection service, and as described above, is configured as the smartphone 2 in this embodiment.
- the smartphone 2 is a general-purpose smartphone, and includes a control unit 22 , a storage unit 23 and a communication unit 24 in addition to a camera 20 and a display 21 . These units 20 to 24 are connected to each other via a bus line 25 and can communicate with each other.
- the display 21 is a touch panel display as described above, and is configured to receive user operations and display various types of information.
- the communication unit 24 functions as a communication interface that performs data communication via a network.
- the control unit 22 can be configured with a CPU, GPU, ROM, RAM, and the like.
- the storage unit 23 can be composed of a non-volatile memory such as a flash memory.
- the storage unit 23 stores image data 201 captured by the camera 20, and also stores a dedicated application program 200 (hereinafter also simply referred to as "application 200") for the estimation system 5.
- the application 200 is installed in the smart phone 2 via a network, for example. When the user starts application 200 , camera 20 , display 21 and the like are controlled to operate as part of estimation system 5 .
- the application 200 cooperates with the web data 7 as necessary, displays each piece of information on the display 21, and supports the user's operation for executing a process of estimating the state of the tire T, which will be described later.
- the server device 4 is a general-purpose computer as hardware and has a non-volatile and rewritable storage device.
- the storage device stores web data 7 that constitutes a dedicated website related to the estimation system 5 in the network to which the estimation device 1, the smartphone 2, and the smartphone 3 are connected.
- the web data 7 is data including website content data such as screen data displayed on the display of each information processing terminal.
- the web data 7 may be accessible from smart phones, tablets, laptops, smart watches, desktops and mobile phones with push buttons via a generic web browser.
- the storage device of server device 4 further includes a vehicle information database.
- the vehicle information is identification information issued by the user to the target vehicle when performing the process of estimating the state of the tire T, and the format is not particularly limited as long as the information can distinguish the vehicle in the estimation system 5. .
- the vehicle information may be, for example, the number on the license plate of the vehicle, the number assigned to the driver for service provision, the date and time when the estimation process was started, and a combination of these numbers.
- the server device 4 acquires vehicle information via an application 200 activated by the smartphone 2, as will be described later.
- the server device 4 stores the acquired vehicle information in its own storage device and constructs a vehicle information database.
- the server device 4 acquires the feedback screen G3 generated by the estimation device 1 in the estimation process described later.
- the server device 4 assigns a unique URL (Uniform Resource Locator) to the acquired feedback screen G3, associates it with the corresponding vehicle information stored in the vehicle information database, and saves it as web data 7.
- This enables access to the feedback screen G3 from a general information processing terminal such as the smartphone 3 via the network.
- the feedback screen cannot be reached unless the dedicated website is first accessed from the information processing terminal and the target vehicle information is entered in the form screen displayed on the display of the information processing terminal. It has become.
- a business card-sized card 6 containing the driver's vehicle information and a URL for accessing a dedicated website is sent from the user to the driver.
- the URL is written on the card 6 in the form of a code that can be read by an electronic device, such as a QR code (registered trademark).
- the smartphone 3 is an example of an information processing terminal owned by the driver.
- the smart phone 3 is a general-purpose smart phone similar to the smart phone 2 and includes a camera 30 and a display 31 .
- the display 31 is configured as a touch panel display.
- the smartphone 3 does not need to have the application 200 installed.
- the driver When the estimation process for the tires T included in the vehicle in which the driver himself/herself rides is finished, the driver is presented with a feedback screen G3 displayed on the display 21 of the smartphone 2 by the user, while explaining the current state of the tires T and making recommendations. Receive proposals for measures to be taken. Then, the user receives a card 6 on which his or her own vehicle information is written. After that, when the driver checks the feedback screen G3 again, he/she refers to the card 6 and reads the code 60 with a barcode reader application using the camera 30 of the smart phone 3 . Thereby, the driver can access the dedicated website from the smartphone 3 without manually inputting the URL of the dedicated website. At this time, the top page of the dedicated website displayed on the display 31 becomes a form screen for inputting vehicle information.
- the driver operates the touch panel display 31 to input his/her own vehicle information written on the card 6 into the input fields on the form screen.
- the server device 4 compares it with the vehicle information database and searches for matching vehicle information.
- the server device 4 transmits to the smart phone 3 the data of the feedback screen G3 associated with the vehicle information found by verification. As a result, the display 31 displays the feedback screen G3.
- the first machine learning model 130 receives an image of the tire T and outputs a value corresponding to the uneven wear of the tire T as an output.
- the input image of the tire T is an image obtained by photographing the tread including both ends of the tire T from the front. is an image taken from the front, and has a predetermined number of pixels of H ⁇ W.
- the estimation result of uneven wear is an index indicating the degree of uneven wear of the tire T in stages. be either Uneven wear level 1 indicates that uneven wear has not occurred or hardly occurred. Uneven wear level 2 indicates that moderate uneven wear has occurred. Uneven wear level 3 indicates that severe uneven wear has occurred.
- the first machine learning model 130 is a convolutional neural network (CNN) in this embodiment, and is built for each type of tire (summer tire, winter tire, all-season tire). That is, the first machine learning model 130 includes three machine learning models: the first machine learning model 130A for summer tires, the first machine learning model 130B for winter tires, and the first machine learning model 130C for all season tires. It is a general term for models.
- Each of the first machine learning models 130A-C is generated by learning using different learning data, and parameters are optimized for each tire type. That is, the layer configurations of the first machine learning models 130A to 130C are common as shown in FIG. 4, but the parameters that define them are different. Therefore, the layer configuration of the first machine learning model 130A will be described below, and the layer configuration of the first machine learning models 130B and 130C will be omitted.
- the first machine learning model 130A comprises alternately arranged convolution layers and pooling layers, a fully connected layer connected to the output side of these layers, and an output layer.
- a convolution layer and a pooling layer are layers for extracting features of an input image.
- the input image is convolved with a large number of weight filters of smaller size.
- M feature maps are generated, which is the same number as the number M of weighting filters (M is an integer of 2 or more and is determined as appropriate). More specifically, the feature map is calculated by repeatedly calculating the inner product of the input image and each weighting filter by raster scanning, and convolving the input image with the weighting filter.
- a weighting filter is an array of values for detecting and enhancing certain patterns (features) contained in the input image.
- a feature map is an image (or array of values) that responds to the features of a weighting filter and emphasizes the features of the weighting filter in the input image. The value of each weighting filter is optimized by learning processing, which will be described later.
- the pooling process is a process of converting the previous feature map into a new feature map by outputting a response value representing a small area included in the previous feature map.
- This pooling process can reduce the size of the previous feature map.
- the position sensitivity can be reduced to accommodate positional shifts of the features to be detected in the input image. More specifically, in the pooling process, each feature map is divided into small regions, and one pixel value as a response value is determined based on the pixel values included in each small region.
- the method of determining the response value can be set as appropriate, and may be the average value of the pixel values in the small region or the maximum value.
- N is an integer equal to or greater than 2 and determined as appropriate
- new weighting filters are prepared, and convolution processing is performed in the same manner as in the first convolution layer.
- N feature maps are generated.
- the values of each weighting filter are optimized by learning processing, which will be described later.
- each of the N feature maps is input to the second pooling layer.
- a second pooling process is performed to transform the N feature maps into new feature maps.
- the feature map (image) output from the second pooling layer is converted into a one-dimensional vector and input to the fully connected layer.
- the fully connected layer is a multi-layer perceptron, and classifies the tire T of the input image into one of uneven wear level 1, uneven wear level 2, and uneven wear level 3 indexes.
- a fully connected layer has an input layer, an intermediate layer and an output layer each having a plurality of units. Weighting coefficients and biases that connect units in each layer are optimized by learning processing, which will be described later.
- Each unit of the output layer in the fully bonded layer corresponds to three indexes (levels) indicating the degree of uneven wear.
- the softmax function is applied to the three output values from the output layer of the fully connected layer, and the uneven wear of the tire T becomes an index of uneven wear level 1, uneven wear level 2, and uneven wear level 3.
- Output each corresponding probability equivalent value The index corresponding to the unit with the highest probability equivalent value corresponds to the estimation result of uneven wear of the tire T.
- the second machine learning model 131 receives the image of the tire T as an input, and outputs a value corresponding to the estimated remaining depth of the main groove of the tread of the tire T, in other words, the wear amount of the tread. .
- the input image of the tire T is the same image as the image input to the first machine learning model 130 .
- the estimation result of the wear amount of the tread is an index indicating the remaining depth of the main groove of the tread step by step.
- wear amount level 3 Wear amount level 1 indicates that the depth of the main groove is sufficient and no or almost no wear occurs.
- Wear amount level 2 indicates that the depth of the main groove has decreased moderately and that moderate wear has occurred.
- Wear amount level 3 indicates that the depth of the main groove is considerably reduced and severe wear is occurring.
- the second machine learning model 131 of this embodiment is also a convolutional neural network (CNN) similar to the first machine learning model 130, and is built for each type of tire (summer tire, winter tire, all-season tire). That is, the second machine learning model 131 includes three machine learning models: a second machine learning model 131A for summer tires, a second machine learning model 131B for winter tires, and a second machine learning model 131C for all season tires. It is a general term for models. Each of the second machine learning models 131A to 131C is generated by learning using different learning data, and parameters are optimized for each tire type. Therefore, although the second machine learning models 131A to 131C have a common layer structure, the parameters that define them are different.
- CNN convolutional neural network
- the layer configuration of the second machine learning models 131A to 131C is common to the first machine learning model 130 shown in FIG. Therefore, since the above description also applies to the second machine learning models 131A-C, the description of the layer configuration for the second machine learning models 131A-C is omitted.
- the trained first machine learning models 130A-C and the trained second machine learning models 131A-C are generated by the learning unit 10E incorporated in the estimation device 1 as described later. be.
- the learning function for learning the machine learning model and generating the learned first and second machine learning models 130 and 131 is based on these learned models 130 and 131 for uneven tire wear and main groove It may be independent of the ability to estimate depth.
- the estimating device 1 is equipped with only an estimating function, and the first machine learning model 130 and the second machine learning model 131 that have been trained by different computers are loaded into the storage unit 13, respectively. can be
- FIG. 5A is a flowchart showing the flow of processing performed by the smartphone 2
- FIG. 5B is a flowchart showing the flow of processing performed by the server device 4 and the estimation device 1.
- a start screen is displayed on the display 21 by the application 200 .
- the start screen displays, for example, an input field for inputting vehicle information and a graphic confirmation button for confirming the input.
- the user operates the touch panel display 21 to input vehicle information in the input field.
- the application 200 accepts input of vehicle information and stores it in the storage unit 23 (step S21).
- the application 200 displays on the display 21 a photographing instruction screen G1 prompting the user to photograph the tire T mounted on each wheel (step S22).
- the photographing instruction screen G1 displays four boxes B1 to B4 corresponding to the left front wheel, right front wheel, left rear wheel, and right rear wheel of the vehicle in correspondence with the positional relationship of the wheels of the vehicle. indicate.
- the position of a wheel is displayed above each box, and a camera icon C1 is displayed in each box. More specifically, above box B1 is labeled "front left”, above box B2 is labeled "front right”, above box B3 is labeled "rear left”, and above box B4 is labeled "rear right”.
- the positional relationship of the boxes B1 to B4 displayed on the photographing instruction screen G1 corresponds to the positional relationship of the wheels of the vehicle.
- FIG. A monitor screen G2 includes a camera frame C2.
- the camera frame C2 is a frame having the same number of H ⁇ W pixels as the input image input to the first and second machine learning models 130 and 131, and image data is generated for the subject of the camera frame C2. configured as
- the user positions himself on the side of the tire T that is attached to the position corresponding to the selected box, and photographs the tire T so that both ends of the tread of the tire T fit within the camera frame C2.
- the user looks at the captured image on the monitor screen G2 and confirms whether the image is suitable as data to be used for estimation processing.
- the image is suitable as data to be used for estimation processing.
- For this confirmation for example, is the main groove of the tread including both ends of the tread clearly photographed? Is the tread photographed from the front? Is the tread continuous in the vertical direction of the image? (whether the contact patch is in the image), whether the image of the tire T is in focus, whether there is any other reflection that interferes with the image of the tire T, and so on.
- selection boxes B5 to B7 are displayed for three types of tires, for example, summer tires, winter tires, and all-season tires.
- the user determines whether the tire T is a summer tire, a winter tire, or an all-season tire by visually confirming the tire T or hearing from the driver, and selects one of the selection boxes B5 to B7 according to this determination.
- Tap As a result, the tapped selection box becomes selected.
- the application 200 brightly displays the selected selection box, for example, and grays out the other selection boxes.
- the user taps the "Save” button graphic displayed on the monitor screen G2 to confirm the captured image. Accordingly, the image data 201 of the tire T associated with the selected tire T type is generated by the application 200 and stored in the storage unit 23 of the smartphone 2 . On the other hand, if the image is determined to be inappropriate, the user taps the graphic of the "redo" button displayed on the monitor screen G2, for example, and takes the image again with the camera 20. FIG. The user can retake the image any number of times until the image data 201 of the tire T corresponding to the selected boxes B1 to B4 is saved.
- the application 200 displays the shooting instruction screen G1 on the display 21 again.
- the photographing instruction screen G1 to be displayed next the image data stored in the storage unit 23 is displayed in the box corresponding to the wheel for which the user has already photographed and confirmed the image.
- 201 thumbnail images are displayed. This allows the user to grasp the wheels for which the image data 201 has not yet been saved.
- the user looks at the photographing instruction screen G1, selects a box whose thumbnail is not displayed, and photographs the tire T corresponding to the selected box in the same procedure as above.
- four pieces of image data 201 including the treads of the four tires T are generated and stored in the storage unit 23 .
- the image data 201 are automatically associated with the positions of the respective wheels and stored in the storage unit 23 .
- the application 200 determines whether the image data 201 of the tire T has been saved for all the wheels of the vehicle (step S23). ). When it is determined that the image data 201 of the tire T has been saved for all the wheels, that is, the four image data 201 of the tire T have been collected (YES), the application 200 displays a start confirmation screen, which will be described later, on the display 21. (Step S24). If it is determined that the image data 201 of the tire T is not stored for all the wheels, that is, the image data 201 of the tire T is less than four (NO), as described above, the application 200 displays the photographing instruction screen G1. is displayed again, and the user is urged to photograph the remaining tires T.
- the start confirmation screen displayed on the display 21 in step S24 performs estimation processing based on the first machine learning model 130 and the second machine learning model 131 using the four pieces of image data 201 determined to be complete in step S23. This is the screen for asking the user whether to start.
- the start confirmation screen includes a message asking the user to start the estimation process, such as "Do you want to start AI diagnosis?", a "start” button graphic, and a "return” button graphic. can be done.
- the application 200 sends the four image data 201 and the type of tire T saved in step S22, and the vehicle information saved in step S21 to the server device. 4 (step S25).
- step S24 the application 200 may be configured to display the shooting instruction screen G1 again.
- the user looks at the thumbnails displayed on the shooting instruction screen G1, for example, and selects the box of the wheel for which the image data 201 is to be modified.
- the application 200 may be configured to accept this operation, activate the camera 20 again, and display the monitor screen G2.
- step S25 when the image data 201, the vehicle information, and the type of tire T are normally transmitted to the server device 4, the application 200 receives the data of the feedback screen G3 generated by the estimation device 1 from the server device 4. , transition to standby mode. From here, the process shown in FIG. 5B starts.
- the server device 4 When the server device 4 receives the image data 201, the vehicle information, and the type of tire T from the smartphone 2 in step S25, it registers the received vehicle information and type of tire T in its own vehicle information database (step S41). In addition, the server device 4 performs the estimation process together with the image data 201, the vehicle information, and the type of the tire T, and transmits to the estimation device 1 a command requesting to send the data of the feedback screen G3 (step S42).
- the estimation device 1 receives the image data 201, the vehicle information, the type of the tire T, and the request for the feedback screen G3 transmitted by the server device 4 in step S42, and starts the process of estimating uneven wear and wear amount of the tire T. (Step S11). In step S11, the image acquisition unit 10A reads (acquires) the received image data 201. FIG.
- the derivation unit 10B selects an appropriate model for each tire T from among the first machine learning models 130A to 130C and the second machine learning models 131A to 131C based on the type of tire T received. do. For example, when the tires T are summer tires, the derivation unit 10B selects the first machine learning model 130A and the second machine learning model 131A. Since this selection is made for each tire T, different machine learning models may be selected even for tires of the same vehicle.
- the derivation unit 10B inputs the image data 201 of the tire T to the first machine learning model 130 selected in step S12, and inputs the image data 201 of the tire T to the second machine learning model 131 selected in step S12. Input image data 201 .
- appropriate tire T image data 201 is input to each machine learning model.
- the derivation unit 10B derives the output for each image data 201 from the first machine learning model 130 and stores it in the storage unit 23.
- the output from the first machine learning model 130 is, as described above, the probability equivalent values corresponding to the three indices indicating the degree of uneven wear.
- the derivation unit 10B derives an output for each image data 201 from the second machine learning model 131 and stores it in the storage unit 23.
- FIG. The output from the second machine learning model 131 is a probability equivalent value corresponding to one of the three indices representing the amount of wear or remaining depth of the main groove of the tread.
- the determination unit 10C uses the output of the first machine learning model 130 and the second machine learning model 131 derived in step S14 as an index indicating the degree of uneven wear of the tire T of each image data 201. and an index indicating the amount of wear.
- the determination unit 10C takes the index having the largest probability equivalent value as the uneven wear estimation result.
- the determination unit 10C takes the index having the largest probability equivalent value among the outputs from the second machine learning model 131 as the wear amount estimation result.
- the determination unit 10C reads the determination table 134 from the storage unit 13, and searches for a cell that matches the estimation result of uneven wear and the estimation result of the amount of wear. As shown in FIG.
- the determination table 134 in this embodiment is a table having 3 ⁇ 3 cells with uneven wear levels as rows and wear amount levels as columns. It has data of A1 to A3 indicating the determination of.
- A1 means that there is no problem with uneven wear and the amount of wear, and no action is required under the present circumstances.
- A2 means that at least one of the uneven wear and the amount of wear is moderate, and tire rotation is recommended as a countermeasure.
- A3 means determination that at least one of the uneven wear and the amount of wear is severe, and replacement of the tire T is recommended as a countermeasure.
- the uneven wear estimation result based on the output of the first machine learning model 130 is uneven wear level 2
- the wear amount estimation result based on the output of the second machine learning model 131 is wear amount level 1.
- the data in the corresponding cell is "A2”, and tire rotation is recommended in this case.
- the uneven wear estimation result based on the output of the first machine learning model 130 is uneven wear level 3
- the wear amount estimation result based on the output of the second machine learning model 131 is wear amount level 1.
- the data of the cell corresponding to this is "A3", and in this case replacement of the tire is recommended.
- the estimation result of uneven wear and the estimation result of the amount of wear are comprehensively considered, and even if one is level 1, the other is level 2 or higher. In some cases, it is determined that a response tailored to the greater degree is recommended. As a result, even if only one of the uneven wear and the amount of wear can be estimated without causing a problem, early countermeasures can be recommended, and appropriate maintenance and management of the tire T can be achieved.
- the screen generator 10D generates the feedback screen G3 based on each index and determination determined in step S15.
- Feedback screen G3 includes, for example, four boxes B8-B11, thumbnails of image data 201 displayed in boxes B8-B11, and graphics C8-C18, as shown in FIG. 9A. Boxes B8 to B11 are displayed in a positional relationship corresponding to the left front wheel, right front wheel, left rear wheel, and right rear wheel of the vehicle, similar to the photographing instruction screen G1. Thumbnails of the image data 201 of the tire T are displayed in boxes B8 to B11 corresponding to the wheels with which the image data 201 is associated.
- Graphics C8 to C11 include figures in which any of the numbers 1 to 3 are displayed and arranged in order.
- the numbers in the figure represent the estimated uneven wear levels 1-3 corresponding to the output derived from the first machine learning model 130, respectively.
- the screen generation unit 10D brightly displays the figures corresponding to the numbers of the uneven wear level determined in step S15 for each of the tires T displayed in the boxes B8 to B11, and brightly displays the figures corresponding to the other numbers. By graying out, the uneven wear level can be easily grasped visually.
- graphics C12 to C15 include figures in which any of the numbers 1 to 3 are displayed and arranged in order.
- the numbers in the figure represent estimated wear amount levels 1 to 3 corresponding to the output derived from the second machine learning model 131, respectively.
- the screen generation unit 10D brightly displays the figure corresponding to the number of the wear amount level determined in step S14 for each tire T displayed in the boxes B8 to B11, and brightly displays the figure corresponding to the other number. By graying out, the level of wear amount can be easily grasped visually. In this way, on the feedback screen G3, as the estimation result of the uneven wear and the amount of wear, indicators showing the degree of uneven wear and wear in stages are displayed together with the overall positioning of the indicators. As a result, the user and the driver who check the feedback screen G3 can intuitively understand the uneven wear and the progress of the wear.
- Graphics C16 to C19 show the determination results based on the determination table 134, and can include different graphics according to the determination results A1 to A3.
- graphic C16 includes a figure of a double circle and corresponds to determination result A1.
- Graphics C17 and C18 include triangular figures and correspond to determination result A2.
- a graphic C19 includes a figure of a cross and corresponds to the determination result A3. That is, the graphics C16 to C19 are graphics that display the result of determination as to whether or not each tire T needs to be dealt with, and recommended actions.
- On the side of graphics C16 to C19 for example, "no response required", “tire rotation required”, “tire replacement required”, etc. A message indicating the may be written together.
- the feedback screen G3 of the present embodiment is an example of a determination result display screen and an uneven wear estimation result display screen.
- the feedback screen G3 illustrated in FIG. 9A displays all of the uneven wear estimation result, the wear amount estimation result, and the determination result, but any one of them may be omitted.
- the determination result display screen and the uneven wear estimation result display screen may not be generated as the same screen data, but may be generated as data of a plurality of screens for each tire, or for each estimation result or determination result.
- the determination unit 10C selects a tire rotation pattern (hereinafter referred to as "recommended pattern") recommended according to the determination based on the determination table 134. ) may be further determined. Determination of the recommended pattern is performed by, for example, incorporating an algorithm for determining the recommended pattern based on determination based on the determination table 134 and wheel position information into the program 132 in advance, and executing this in step S15. may Further, the determination unit 10C may determine a generally recommended rotation pattern as a recommended pattern without depending on an algorithm or the like for determining a recommended pattern.
- the screen generator 10D may further generate data for the rotation screen G4 indicating the recommended pattern in addition to the data for the feedback screen G3.
- An example of the rotation screen G4 is shown in FIG. 9B.
- the rotation screen G4 includes, for example, a graphic C20 of a vehicle with tires T mounted on each wheel, graphics C21 to C23 indicating recommended patterns, and a graphic C24 indicating replacement.
- the graphics C21 to C23 are arrow figures, for example, and indicate that the wheel position indicated by the end of the arrow is recommended as the replacement position of the tire T at the starting point of the arrow.
- the graphic C24 is, for example, a figure of circular arrows, and is displayed on the graphic of the tire T or on the side thereof to indicate that replacement of the tire T is recommended.
- the graphics of each tire T may change in pattern, color, etc. according to the determination result based on the determination table 134, for example.
- the screen generator 10D transmits the data of the generated feedback screen G3 to the server device 4 (step S17).
- the screen generation unit 10D also transmits this in the same manner.
- the estimating device 1 completes the tire state estimating process for one set of image data 201 .
- the server device 4 receives the data of the feedback screen G3 transmitted from the estimation device 1, associates it with the vehicle information database, and stores it in its own storage device (step S43).
- the server device 4 assigns a unique URL to the data of the feedback screen G3 and uses it as web data 7 .
- the server device 4 transmits the data of the feedback screen G3 to the smartphone 2 (step S44). Also when the data of the rotation screen G4 is transmitted, the same processing as that of the data of the feedback screen G3 is performed, and the data of the rotation screen G4 is also made accessible via the network. As a result, the server device 4 completes the processing of the set of image data 201 .
- the smartphone 2 receives the data of the feedback screen G3 transmitted by the server device 4 in step S44.
- the application 200 displays the feedback screen G3 on the display 21 (step S26).
- the user While presenting the feedback screen G3 displayed on the display 21 to the driver, the user explains the situation of uneven wear and the amount of wear estimated for the four-wheel tires, and the recommended measures.
- a case where the feedback screen G3 is the screen illustrated in FIG. 9A will be described below as an example.
- the user shows graphics C8 and C12 and explains that both the uneven wear level and the wear amount level are estimated to be "1". Can explain that there seems to be no problem (it is determined that no action is required).
- the user For the right front wheel (front right), the user shows graphics C9 and C13 and explains that the uneven wear level is estimated to be "2" but the wear amount level is estimated to be "1".
- While showing graphic C17 it can be explained that tire rotation is recommended in order to suppress further progression of uneven wear.
- the user explains that the uneven wear level is estimated to be "1" and the wear amount level is estimated to be “2" while showing graphics C10 and C14, and then While showing graphic C18, it can be explained that tire rotation is recommended in order to smooth out the amount of wear.
- the uneven wear level is estimated to be "3” and the wear amount level is estimated to be "2” while showing graphics C11 and C15.
- C19 is shown, it can be explained that tire replacement is recommended because even if the amount of wear appears to be moderate, uneven wear is severe.
- the images of the four wheels are listed as thumbnails, and the estimated uneven wear level and wear amount level are displayed in a graphic format together with the overall stage.
- the driver who receives the explanation from the user does not check the tires T of the four wheels and does not go around, the driver can grasp the estimated situation of the tires T of his own vehicle with a sense of satisfaction.
- any one of three types of graphics is displayed in accordance with the recommended measures for the larger one of the uneven wear and the amount of wear. This allows the driver to plan early tire rotation or tire replacement. For example, a driver presented with a feedback screen G3 as shown in FIG.
- the application 200 can switch the feedback screen G3 to the rotation screen G4 according to the user's operation and display it on the display 21.
- the user can easily recognize the recommended pattern by checking the rotation screen G4.
- the recommended patterns can be explained to the driver while presenting the rotation screen G4, and the driver's sense of satisfaction can be further enhanced.
- the first machine learning models 130A to 130C of this embodiment are generated for each tire type. For this reason, learning data for the first machine learning models 130A to 130C are also prepared for each type of tire T.
- FIG. In the present embodiment, the learning data for the first machine learning model 130A is a large number of data sets in which learning images including treads of summer tires and correct data are combined.
- Data for learning is a large number of data sets in which learning images including treads of winter tires and correct data are combined
- data for learning of the first machine learning model 130C is data for learning including treads of all-season tires. It is a large number of data sets in which images and correct data are combined.
- the correct data is a label of an index representing the degree of uneven wear.
- the correct data is a label of one of uneven wear level 1, uneven wear level 2, and uneven wear level 3.
- the training image and these labels are combined by a person who has checked the actual tire in the training image.
- the learning process for generating the trained first machine learning model 130A will be described below, but the trained first machine learning models 130B and 130C are also generated by similar learning processes. Therefore, description of the learning process of the first machine learning models 130B and 130C is omitted.
- step S51 learning data including a large number of data sets in which learning images including tire treads are combined with correct data are prepared, and stored in the storage unit 13 by the learning unit 10E of the estimation device 1. be done.
- the learning image is an image of the tread photographed from the front so that the tread of the tire is continuous in the longitudinal direction while including both ends of the tread of the tire.
- the learning unit 10E divides the learning data into training data and test data in advance and stores them. The ratio of both can be set as appropriate.
- the learning unit 10E randomly selects K data sets as sample data from the training data.
- K is a value also called a batch size, and can be set as appropriate.
- the learning unit 10E inputs the K learning images included in the sample data to the first machine learning model 130A, and derives the output from the first machine learning model 130A.
- the output is data corresponding to the correct data combined with each of the input K learning images. This is the corresponding probability equivalent value.
- the learning unit 10E adjusts the parameters so that the value of the error function between the output derived in step S53 and the correct data combined with the learning image input in step S53 is minimized. do. More specifically, the learning unit 10E adjusts and updates the values of the weight coefficients and biases in the fully connected layers and the weight filters in the convolution layers of the first machine learning model 130A using backpropagation.
- the learning unit 10E determines whether or not one epoch of learning has been completed. In this embodiment, it is determined that one epoch of learning has been completed when the processing from step S52 to step S55 is performed for the same number of sample data as the number of training data. If it is determined that one epoch of learning has not been completed, the learning unit 10E returns to step S52 after step S55. That is, the learning unit 10E randomly selects sample data again, and repeats the procedure from step S53 to step S55 using the newly selected sample data. On the other hand, if it is determined that one epoch of learning has been completed, it is determined in step S56 whether or not all epochs of learning have been completed.
- step S56 determines in step S56 that the learning of all epochs has not been completed, the learning unit 10E executes step S52 again to perform the learning of the next epoch.
- the total number of epochs is not particularly limited and can be set as appropriate.
- step S56 determines in step S56 that learning of all epochs has ended.
- the learning unit 10E ends learning of the first machine learning model 130A.
- the learning unit 10E stores the latest parameters of the first machine learning model 130A in the storage unit 13, and uses this as the learned first machine learning model 130A. That is, the learned first machine learning model 130A is generated by the above procedure.
- the learning unit 10E inputs the test data to the first machine learning model 130A each time one epoch of learning is completed, calculates the error between the output and the test data with respect to the correct data, and displays the calculation result. It may be displayed in section 11. Also, if it is considered that the error in the output of the first machine learning model 130A converges within a predetermined range before the learning of all epochs is completed, the learning may be terminated at that point.
- the learning unit 10E performs the above-described learning process for the first machine learning models 130A-C, and also performs the same learning process for the second machine learning models 131A-C. That is, the processing of steps S51 to S56 described above also applies to the learning processing of the second machine learning models 131A to 131C. Therefore, the detailed description of the learning process of the second machine learning models 131A to 131C will be omitted, and the learning data of the second machine learning models 131A to 131C will be described below.
- the second machine learning models 131A to 131C of this embodiment are generated for each type of tire T. For this reason, learning data for the second machine learning models 131A to 131C are also prepared for each type of tire T.
- FIG. In the present embodiment, the learning data for the second machine learning model 131A is a large number of data sets in which learning images including the treads of summer tires and correct data are combined.
- the data for learning is a large number of data sets in which learning images including treads of winter tires and correct data are combined
- the data for learning of the second machine learning model 131C is for learning including treads of all-season tires It is a large number of data sets in which images and correct data are combined.
- the learning image is an image of the tire tread taken from the front so that both ends of the tire tread are included and the tread is continuous in the longitudinal direction.
- the correct data is the label of the index representing the degree of wear.
- the correct answer data is a label of one of wear amount level 1, wear amount level 2, and wear amount level 3.
- FIG. Note that the learning image may be data common to the learning image of the first machine learning model 130 .
- the wear amount level is determined according to the remaining depth of the main groove of the tread for each type of tire T, and the depth of the main groove measured in the actual tire of the learning image is determined. , an appropriate wear level is selected and a label representing this is combined with the training images.
- a person checks the actual tire, selects the most worn (shallowest) portion of the main groove, and measures the depth of the groove at that portion.
- the uneven wear and the amount of wear of the tire T are estimated by a machine learning model that has already been learned. Therefore, variations in judgment by people are reduced, and the user's tire inspection service is supported.
- the algorithm for determining the necessity of response and the type of response of the estimated degree of uneven wear and the amount of wear, the more severe one is weighted, so a more appropriate response is recommended. It also enables early tire rotation and replacement.
- the user can present these data to the driver, and more effectively explain and propose services. management can be encouraged. Since the driver can check the feedback screen G3 and the rotation screen G4 at any time by using the card 6 provided by the user, the driver can present these to the user at a later date and receive necessary services.
- the CNN model was used as the first and second machine learning models 130, 131, but the machine learning model is not limited to this, support vector machine (SVM), neural network (NN) model, Other machine learning models such as K-NN models, clustering, k-means, decision trees, and models combining these may be used.
- the learning data for the second machine learning model 131 may be a data set in which a learning image and the depth of the main groove as correct data are combined. That is, the output of the second machine learning model 131 may be an estimate of the remaining depth of the main groove of the tire in the input image.
- the learning method of the first and second machine learning models 130 and 131 is not limited to the above embodiment, and a known parameter optimization algorithm such as stochastic gradient descent can be applied.
- the loss function is not limited to the above embodiment, and can be changed as appropriate according to the properties of the data to be output.
- the images including the tire treads input to the first and second machine learning models 130 and 131 and the learning images for learning these machine learning models have treads that are not continuous in the vertical direction. , the tread may be continuous in the lateral direction, and the tread may be continuous in other directions.
- the uneven wear and the amount of wear are estimated together as the state of the tire T, but only the uneven wear may be estimated, and the recommended measures may be determined based on this estimation result.
- the degree of uneven wear was estimated as the estimation of uneven wear, but in addition to or instead of this, the type of uneven wear may be estimated.
- the types of uneven wear are, for example, patterns shown in FIGS. 11A to 11D.
- FIG. 11A shows "unilateral wear” in which one side of the tread in the width direction of the tire T (the direction in which the rotation axis of the tire T extends) is concentrated and worn. "Uneven wear” is further subdivided into either “inner wear” or “outer wear” depending on which side of the vehicle the tire T is mounted on, left or right.
- FIG. 11B shows a state in which both sides of the tread in the width direction of the tire T are worn intensively, which is a state of "worn both sides”.
- FIG. 11A shows "unilateral wear” in which one side of the tread in the width direction of the tire T (the direction in which the rotation axis of the tire T extends) is concentrated and worn. "Uneven wear” is further subdivided into either “in
- FIG. 11C shows "center wear” in which the center of the tread in the width direction of the tire T is worn intensively.
- FIG. 11D shows "heel-toe wear", which is a state in which each block of the tread wears in one direction to form a sawtooth shape when the tire T is viewed from the circumferential direction.
- the type of uneven wear can be estimated, for example, by implementing a learned machine learning model in the estimation device 1, in which the input is a tire image and the output is each pattern of uneven wear and the classification of no uneven wear. .
- the structure of this machine learning model is not particularly limited.
- the learning data for the machine learning model can be a large number of data sets in which learning images of the tire T and correct data are combined.
- the correct data is a label corresponding to each type of uneven wear, including, for example, a state in which uneven wear does not occur.
- the determination unit 10C of the estimation device 1 may determine the recommended response in more detail, and the information on the mounting position of the tire T may also be acquired, and the recommended pattern may be further subdivided and determined. For example, when it is estimated that the tire T is "reduced on both sides", the air pressure of the tire T is not appropriate and may be decompressed, so for example, the driver is recommended to review the air pressure of the tire T. can do. On the other hand, when it is estimated that "center wear” has occurred in the tire T, the air pressure of the tire T is not appropriate and may be high, so for example, the driver is recommended to review the air pressure of the tire T. can do. Furthermore, when it is estimated that the tire T is unevenly worn, the determination unit 10C can determine a rotation pattern in which the mounting position of the tire T is left-right reversed as a recommended pattern.
- the screen generation unit 10D of the estimation device 1 when the type of uneven wear is estimated in the estimation system 5, the screen generation unit 10D of the estimation device 1 generates a screen showing the type of uneven wear estimated for each tire T as an estimation result display screen. good too.
- the type of uneven wear may be displayed as text information, or may be displayed graphically as shown in FIGS. 11A to 11D, for example.
- the screen generator 10D may further generate a rotation screen that displays recommended patterns.
- the code 60 displayed on the card 6 was generated for the URL of the top page of the dedicated website.
- the code 60 may be generated for at least one of the URL of the feedback screen G3 and the URL of the rotation screen G4 and displayed on the card 6. good.
- the display of the URL may be simply a character notation, or may be a one-dimensional code other than the QR code (registered trademark), a stack-type two-dimensional code, or a matrix-type two-dimensional code.
- the card 6 is not limited to a physical card, and may be screen data simulating this, and may be generated by the application 200, for example.
- the data of the screen imitating the card 6 may be transmitted from the smart phone 2 to the driver's information processing terminal.
- the user may display a screen imitating the card 6 on the display 21, and the driver may photograph this with his own information processing terminal or camera.
- the estimation device 1 and the server device 4 are configured as separate devices, but these devices may be configured integrally.
- the smartphone 2 may include at least a part of the functions of the estimation device 1, and the smartphone 2 may perform at least one of uneven wear and wear amount of the tire T.
- the application 200 is configured as a program that incorporates the functions of the learned first machine learning model 130, the second machine learning model 131, and the program 132, and by installing the application 200 on the smartphone 2, the estimation device 1 A smartphone 2 may be manufactured that includes the functions of Further, in the above embodiment, the user uses the smartphone 2 on which the application 200 is installed to provide services to the driver.
- the estimation system 5 is such that the driver installs the application 200 on his/her own smartphone 3, photographs the tire T of the vehicle himself, and executes the process of estimating at least one of the uneven wear and the amount of wear of the tire T.
- the camera that takes an image of the tire T and generates the image data 201 is not limited to a camera built into various information processing terminals, and may be a digital camera, a video camera, or the like dedicated to shooting.
- the control unit 10 of the estimation device 1 may be configured to include a vector processor, FPGA (Field Programmable Gate Array), ASIC (Application Specific Integrated Circuit), other artificial intelligence chips, etc., in addition to the CPU and GPU.
- FPGA Field Programmable Gate Array
- ASIC Application Specific Integrated Circuit
- the first machine learning models 130A-C and the second machine learning models 131A-C are generated according to the tire type, but the first machine learning model 130 and the second machine learning model 131 are one may be generated for only one type, or conversely, more types may be generated for more types.
- the estimation device 1 may be configured to appropriately perform image processing such as resizing and trimming when inputting the image data 201 to the machine learning model.
- the start screen, the shooting instruction screen G1, the monitor screen G2, and the start confirmation screen displayed on the display 21 in steps S21 to S24 of the above embodiment may be screens incorporated in the application 200, or web data 7. It may be a screen configured as Also, the configuration of these user interface screens can be changed as appropriate.
- the shooting instruction screen G1 and the feedback screen G3 may not be divided into four by boxes, and information about one tire T may be displayed on each screen, and the tire T may be switched by tabs. .
- graphics C8 to C11 representing uneven wear estimation results
- graphics C12 to C15 representing wear amount estimation results
- graphics C16 to C19 representing determination results
- vehicle graphics C20 and rotation and replacement graphics C21 to C21 C24 may also be changed as appropriate, and may be configured in a moving image format.
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Mechanical Engineering (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Theoretical Computer Science (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
- Tires In General (AREA)
Abstract
Description
対象となるタイヤのトレッドの両端を含むとともに、所定方向に該トレッドが連続するように該トレッドを正面から撮影した画像を取得すること
取得した画像を、学習済みの第1機械学習モデルに入力すること
学習済みの第1機械学習モデルから出力を導出すること。
なお、学習済みの第1機械学習モデルの出力は、対象となるタイヤの偏摩耗の推定結果に対応する。
対象となるタイヤのトレッドの両端を含むとともに、所定方向に該トレッドが連続するように該トレッドを正面から撮影した画像を取得すること
取得した画像を、学習済みの第1機械学習モデルに入力すること
学習済みの第1機械学習モデルから出力を導出すること。
なお、学習済みの第1機械学習モデルの出力は、対象となるタイヤの偏摩耗の推定結果に対応する。
タイヤのトレッドの両端を含むとともに、所定方向に該トレッドが連続するように該トレッドを正面から撮影した学習用画像と、正解データとのデータセットである学習用データを用意すること
学習用データを用いて、対象となるタイヤのトレッドの両端を含むとともに、所定方向に該トレッドが連続するように該トレッドを正面から撮影した画像を入力すると、正解データに対応するデータが出力されるように機械学習モデルのパラメータを調整すること。
なお、正解データは、学習用画像のタイヤの偏摩耗の程度を表す指標及び学習用画像のタイヤの偏摩耗の類型に対応するラベルの少なくとも一方である。
対象となるタイヤのトレッドを含む画像を取得すること
取得した画像を、学習済みの第1機械学習モデルに入力すること
学習済みの第1機械学習モデルから出力を導出すること
取得した画像を、学習済みの第1機械学習モデルとは異なる学習済みの第2機械学習モデルに入力すること
学習済みの第2機械学習モデルから出力を導出することと、
学習済みの第1機械学習モデルから導出された出力及び学習済みの第2機械学習モデルから導出された出力に基づいて、対象となるタイヤの交換が必要か否かを判定すること。
なお、学習済みの第1機械学習モデルの出力は、対象となるタイヤの偏摩耗の推定結果に対応し、学習済みの第2機械学習モデルの出力は、対象となるタイヤのトレッドの溝の深さの推定結果に対応する。
なお、学習済みの第1機械学習モデルの出力は、対象となるタイヤの偏摩耗の推定結果に対応し、学習済みの第2機械学習モデルの出力は、対象となるタイヤのトレッドの溝の深さの推定結果に対応する。
対象となるタイヤのトレッドを含む画像を取得すること
取得した画像を、学習済みの第1機械学習モデルに入力すること
取得した画像を、学習済みの第1機械学習モデルとは異なる学習済みの第2機械学習モデルに入力すること
学習済みの第1機械学習モデルから出力を導出すること
学習済みの第2機械学習モデルから出力を導出すること
学習済みの第1機械学習モデルから導出された出力及び学習済みの第2機械学習モデルから導出された出力に基づいて対象となるタイヤの交換が必要か否かを判定すること。
なお、学習済みの第1機械学習モデルの出力は、対象となるタイヤの偏摩耗の推定結果に対応し、学習済みの第2機械学習モデルの出力は、対象となるタイヤのトレッドの溝の深さの推定結果に対応する。
対象となるタイヤのトレッドを含む画像を取得すること
取得した画像を、学習済みの第1機械学習モデルに入力すること
学習済みの第1機械学習モデルから出力を導出すること。
なお、学習済みの第1機械学習モデルの出力は、対象となるタイヤの偏摩耗の推定結果に対応する。
対象となるタイヤのトレッドを含む画像を取得すること
取得した画像を、学習済みの第1機械学習モデルに入力すること
学習済みの第1機械学習モデルから出力を導出すること。
なお、学習済みの第1機械学習モデルの出力は、対象となるタイヤの偏摩耗の推定結果に対応する。
タイヤのトレッドを含む学習用画像と、正解データとのデータセットである学習用データを用意すること
学習用データを用いて、対象となるタイヤのトレッドを含む画像を入力すると、正解データに対応するデータが出力されるように機械学習モデルのパラメータを調整すること。
なお、正解データは、学習用画像のタイヤの偏摩耗の程度を表す指標及び学習用画像のタイヤの偏摩耗の類型に対応するラベルの少なくとも一方である。
図1は、本開示の一実施形態に係る推定システム5の全体構成図である。推定システム5は、タイヤTの状態としてタイヤTの偏摩耗及び摩耗量を推定し、推定結果やこれに基づいて推奨される対応(タイヤ交換及びタイヤローテーション等)をフィードバックするためのシステムである。本実施形態に係る推定システム5の主たるユーザとして想定されるのは、これに限られないが、車両のドライバーに対し、タイヤの点検サービスを提供する者である。
[推定装置]
図2は、推定装置1の電気的構成を示すブロック図である。推定装置1は、ハードウェアとしては汎用のコンピュータであり、例えば、デスクトップパソコン、ラップトップパソコン、タブレット、スマートフォンとして実現される。推定装置1は、CD-ROM、USBメモリ等のコンピュータで読み取り可能な記憶媒体133から、或いはネットワークを介して、プログラム132を汎用のコンピュータにインストールすることにより製造される。プログラム132は、スマートフォン2から送られてくる画像データ201(図4参照)に基づいて画像に写るタイヤTの状態を推定するとともに、推定結果に基づいて、何らかの対応が必要か否かを判定し、推奨される対応を決定するソフトウェアであり、推定装置1に後述する動作を実行させる。
図3は、推定システム5を構成するカメラ20と、ディスプレイ21とを備える情報処理端末の電気的構成を示すブロック図である。情報処理端末は、ユーザがタイヤの点検サービスを提供するための装置であり、上述した通り、本実施形態ではスマートフォン2として構成される。本実施形態では、スマートフォン2は汎用のスマートフォンであり、カメラ20及びディスプレイ21に加えて、制御部22、記憶部23、及び通信部24を備える。これらの部20~24は、互いにバス線25を介して接続されており、相互に通信可能である。ディスプレイ21は、上述した通りタッチパネルディスプレイであり、ユーザによる操作を受け付けるとともに、各種の情報を表示するように構成される。通信部24は、ネットワークを介したデータ通信を行う通信インターフェースとして機能する。
サーバー装置4は、ハードウェアとしては汎用のコンピュータであり、不揮発性で書き換え可能な記憶装置を有する。記憶装置には、推定装置1、スマートフォン2及びスマートフォン3が接続されるネットワークにおいて、推定システム5に関連する専用のウェブサイトを構成するウェブデータ7が格納される。ウェブデータ7は、各情報処理端末のディスプレイで表示される画面のデータ等、ウェブサイトのコンテンツデータが含まれたデータである。ウェブデータ7は、スマートフォン、タブレット、ラップトップパソコン、スマートウォッチ、デスクトップパソコン及びプッシュボタン付き携帯電話から、汎用のウェブブラウザを介してアクセス可能であってよい。
本実施形態では、スマートフォン3等からフィードバック画面へのアクセスを容易にすべく、ドライバーの車両情報と、専用ウェブサイトにアクセスするためのURLとが記載された名刺サイズのカード6が、ユーザからドライバーに提供される。URLは、例えばQRコード(登録商標)等の、電子機器により読み取り可能なコード形式でカード6に記載される。
スマートフォン3は、ドライバーが所有する情報処理端末の一例である。スマートフォン3は、スマートフォン2と同様に汎用のスマートフォンであり、カメラ30とディスプレイ31とを備える。ディスプレイ31は、タッチパネルディスプレイとして構成される。スマートフォン3は、スマートフォン2とは異なり、アプリ200がインストールされる必要はない。
[第1機械学習モデル]
次に、図4を参照しつつ、後述する推定処理の中で使用される第1機械学習モデル130の構成について説明する。第1機械学習モデル130は、上述したように、タイヤTの画像を入力とし、タイヤTの偏摩耗の推定結果に対応する値を出力とする。入力されるタイヤTの画像は、タイヤTの正面から両端を含むトレッドを撮影した画像、より詳細には、タイヤTのトレッドの両端を含むとともに、画像の縦方向にトレッドが連続するようにトレッドを正面から撮影した画像であり、予め決められたH×Wのピクセル数を有する。偏摩耗の推定結果は、本実施形態では、タイヤTの偏摩耗の程度を段階的に示す指標であり、偏摩耗の程度が少ない順に、偏摩耗レベル1、偏摩耗レベル2、偏摩耗レベル3のいずれかとなる。偏摩耗レベル1は、偏摩耗が生じていないか、殆ど生じていないことを表す。偏摩耗レベル2は、中程度の偏摩耗が生じていることを表す。偏摩耗レベル3は、重度の偏摩耗が生じていることを示す。
第2機械学習モデル131は、上述したように、タイヤTの画像を入力とし、タイヤTのトレッドの主溝の残り深さ、言い換えるとトレッドの摩耗量の推定結果に対応する値を出力とする。入力されるタイヤTの画像は、第1機械学習モデル130に入力される画像と同一の画像である。トレッドの摩耗量の推定結果は、本実施形態では、トレッドの主溝の残り深さを段階的に示す指標であり、主溝の残り深さが深い順に、摩耗量レベル1、摩耗量レベル2、摩耗量レベル3のいずれかとなる。摩耗量レベル1は、主溝の深さが充分であり、摩耗が生じていないか、殆ど生じていないことを表す。摩耗量レベル2は、主溝の深さが中程度に減っており、中程度の摩耗が生じていることを表す。摩耗量レベル3は、主溝の深さが相当減っており、重度の摩耗が生じていることを示す。
次に、図5~図9を参照しつつ、タイヤTの偏摩耗及び摩耗量を推定する推定システム5の動作について説明する。タイヤTの偏摩耗及び摩耗量を推定する推定処理は、推定装置1によって実行される。しかし、推定システム5全体は、推定装置1、スマートフォン2及びサーバー装置4が協働することにより動作する。従って、以下ではスマートフォン2等に表示されるユーザーインターフェース画面についても適宜触れながら、推定装置1による推定処理を含む推定システム5の動作について説明する。この動作は、例えばユーザがスマートフォン2のアプリ200を起動したときにスタートする。図5Aは、スマートフォン2で行われる処理の流れを示すフローチャートであり、図5Bは、サーバー装置4及び推定装置1で行われる処理の流れを示すフローチャートである。
以下、図10を参照しつつ、学習済みの第1機械学習モデル130及び第2機械学習モデル131を生成するための方法、つまり、学習部10Eにより実行される第1及び第2機械学習モデルの学習方法について説明する。
上述したように、本実施形態の第1機械学習モデル130A~Cは、タイヤTの種類別に生成される。このため、第1機械学習モデル130A~Cの学習用データもタイヤTの種類別に用意される。本実施形態では、第1機械学習モデル130Aの学習用データは、夏タイヤのトレッドを含む学習用画像と、正解データとが組み合わせられた多数のデータセットであり、第1機械学習モデル130Bの学習用データは、冬タイヤのトレッドを含む学習用画像と、正解データとが組み合わせられた多数のデータセットであり、第1機械学習モデル130Cの学習用データは、オールシーズンタイヤのトレッドを含む学習用画像と、正解データとが組み合わせられた多数のデータセットである。これらのいずれの学習用データにおいても、正解データは、偏摩耗の程度を表す指標のラベルである。つまり、正解データは、偏摩耗レベル1、偏摩耗レベル2、偏摩耗レベル3のいずれかのラベルである。本実施形態では、学習用画像のタイヤ現物を確認した人によって、学習用画像とこれらのラベルとが組み合わせられる。
学習部10Eは、第1機械学習モデル130A~Cについて以上の学習処理を行うとともに、第2機械学習モデル131A~Cについても同様の学習処理を行う。すなわち、上述したステップS51~S56までの処理は、第2機械学習モデル131A~Cの学習処理についても該当する。このため、第2機械学習モデル131A~Cの学習処理の詳細については説明を省略し、以下では第2機械学習モデル131A~Cの学習用データについて説明する。
以上の推定システム5によれば、学習済みの機械学習モデルによりタイヤTの偏摩耗及び摩耗量が推定されるので、人による判断のばらつきが少なくなり、ユーザのタイヤ点検サービスが支援される。また、対応の要否や対応の種類を判定するアルゴリズムでは、推定される偏摩耗の程度及び摩耗量の程度のうち、より重度である方に重み付けがなされているため、より適切な対応が推奨されるとともに、早期にタイヤのローテーションや交換を行うことができる。さらに、フィードバック画面G3やローテーション画面G4のデータが生成されることにより、ユーザはこれらを提示し、ドライバーに対してより効果的に説明やサービスの提案を行うことができ、ドライバーによる適切なタイヤTの管理を促すことができる。ドライバーは、ユーザから提供されるカード6により、好きなタイミングでフィードバック画面G3やローテーション画面G4を確認できるため、後日これらをユーザに提示して、必要なサービスの提供を受けることが可能である。
以上、本発明の一実施形態について説明したが、本発明は上記実施形態に限定されるものではなく、その趣旨を逸脱しない限りにおいて、種々の変更が可能である。以下に示す変形例の要旨は、適宜組み合わせることができる。
上記実施形態では、第1及び第2機械学習モデル130,131としてCNNモデルが用いられたが、機械学習モデルはこれに限定されず、サポートベクタ―マシン(SVM)、ニューラルネットワーク(NN)モデル、K-NNモデル、クラスタリング、k-means、決定木等、その他の機械学習モデル及びこれらを組み合わせたモデルが用いられてもよい。また、第2機械学習モデル131の学習用データは、学習用画像と、正解データとして主溝の深さとが組み合わされたデータセットであってもよい。すなわち、第2機械学習モデル131の出力は、入力画像のタイヤの主溝の残り深さの推定値であってもよい。また、1及び第2機械学習モデル130,131の学習方法は上記実施形態に限定されず、確率的勾配降下法等、公知のパラメータ最適化アルゴリズムを適用することができる。また、損失関数も上記実施形態に限定されず、出力されるデータの性質に応じて適宜変更することができる。また、第1及び第2機械学習モデル130,131に入力されるタイヤのトレッドを含む画像、並びにこれらの機械学習モデルを学習させるための学習用画像は、縦方向にトレッドが連続するのではなく、横方向にトレッドが連続してもよく、その他の方向にトレッドが連続してもよい。
上記実施形態では、タイヤTの状態として、偏摩耗と摩耗量とが合わせて推定されたが、偏摩耗のみを推定し、この推定結果に基づいて推奨される対応が判定されてもよい。
上記実施形態では、偏摩耗の推定として、偏摩耗の程度が推定されたが、これに加えてまたはこれに替えて、偏摩耗の類型が推定されてもよい。偏摩耗の類型とは、例えば図11A~Dに示すパターンである。図11Aは、タイヤTの幅方向(タイヤTの回転軸が延びる方向)において、トレッドの片側が集中して摩耗した状態である、「片減り」である。「片減り」は、タイヤTが車両の左右のいずれの側に装着されているかにより、「内減り」または「外減り」のいずれかにさらに細分化される。図11Bは、タイヤTの幅方向において、トレッドの両側が集中して摩耗した状態である、「両減り」である。図11Cは、タイヤTの幅方向において、トレッドの中央が集中して摩耗した状態である、「センター摩耗」である。図11Dは、タイヤTを周方向から見たときに、トレッドの各ブロックが一方向に摩耗し、のこぎりの歯状となった状態である「ヒールトー摩耗」である。偏摩耗の類型の推定は、例えば入力をタイヤ画像とし、出力を偏摩耗の各パターン及び偏摩耗無しの分類とする、学習済みの機械学習モデルを推定装置1に実装することにより行うことができる。この機械学習モデルの構造は特に限定されない。また、機械学習モデルの学習用データは、タイヤTの学習用画像と、正解データとが組み合わせられた多数のデータセットとすることができる。正解データは、例えば偏摩耗が生じていない状態を含む、偏摩耗の各類型に対応するラベルである。
上記実施形態では、カード6に表示されるコード60は、専用ウェブサイトのトップページのURLについて生成されたものであった。しかし、トップページのURLに加えてまたはこれに替えて、フィードバック画面G3のURL、及びローテーション画面G4のURLのうち少なくとも一方に対してもコード60を生成し、これをカード6に表示してもよい。これにより、ドライバーは、スマートフォン3等から専用ウェブサイトにアクセスしたときに、フォーム画面に車両情報の入力を要求されることなく、ダイレクトにフィードバック画面G3やローテーション画面G4にアクセスすることができる。また、URLの表示は単に文字表記であってもよく、QRコード(登録商標)以外の1次元コード、スタック型2次元コード、マトリクス型2次元コードであってもよい。さらに、カード6は物理的なカードに限定されず、これを模した画面のデータであってもよく、例えばアプリ200により生成されてもよい。この場合、カード6を模した画面のデータは、スマートフォン2からドライバーの情報処理端末に送信されてもよい。あるいは、ユーザがディスプレイ21にカード6を模した画面を表示し、ドライバーが自身の情報処理端末やカメラによりこれを撮影してもよい。
上記実施形態では、推定装置1とサーバー装置4とが別の装置として構成されていたが、これらの装置は一体的に構成されてもよい。あるいは、スマートフォン2が推定装置1の少なくとも一部の機能を包含してもよく、スマートフォン2がタイヤTの偏摩耗及び摩耗量の少なくとも一方の推定処理を行ってもよい。この場合、アプリ200が学習済みの第1機械学習モデル130、第2機械学習モデル131及びプログラム132の機能が組み込まれたプログラムとして構成され、スマートフォン2にアプリ200をインストールすることで、推定装置1の機能を包括するスマートフォン2が製造されてもよい。また、上記実施形態ではユーザがアプリ200がインストールされたスマートフォン2を使用して、ドライバーに対するサービスを提供した。しかし、推定システム5は、ドライバーがアプリ200を自身のスマートフォン3にインストールして、自身で車両のタイヤTを撮影し、タイヤTの偏摩耗及び摩耗量の少なくとも一方の推定処理を実行するように構成されてもよい。また、タイヤTを撮影し、画像データ201を生成するカメラは、各種情報処理端末に内蔵されるカメラに限られず、撮影専用のデジタルカメラやビデオカメラ等であってもよい。
推定装置1の制御部10は、CPUやGPUの他、ベクトルプロセッサ、FPGA(Field Programmable Gate Array)、ASIC(Application Specific Integrated Circuit)、その他人工知能専用チップ等を含んで構成されてもよい。
上記実施形態では、タイヤの種類に応じて第1機械学習モデル130A~C及び第2機械学習モデル131A~Cが生成されたが、第1機械学習モデル130及び第2機械学習モデル131は1つだけ生成されてもよいし、反対にさらに多くの種類について生成されてもよい。また、推定装置1は、画像データ201を機械学習モデルに入力するにあたり、リサイズやトリミング等、適宜画像処理を行うように構成されてもよい。
上記実施形態のステップS21~S24でディスプレイ21に表示されるスタート画面、撮影指示画面G1、モニタ画面G2、開始確認画面は、アプリ200に組み込まれている画面であってもよいし、ウェブデータ7として構成された画面であってもよい。また、これらのユーザーインターフェース画面の構成は適宜変更することができる。撮影指示画面G1及びフィードバック画面G3は、例えば、ボックスにより4分割されていなくてもよく、1つの画面につき、1つのタイヤTについての情報が表示され、タブによりタイヤTを切り替えられる構成としてもよい。また、偏摩耗の推定結果を表すグラフィックC8~C11、摩耗量の推定結果を表すグラフィックC12~C15、判定結果を表すグラフィックC16~C19、車両を表すグラフィックC20、及びローテーションや交換を表すグラフィックC21~C24も適宜変更されてもよく、動画形式で構成されてもよい。
2 スマートフォン
3 スマートフォン
4 サーバー装置
5 推定システム
10 制御部
10A 画像取得部
10B 導出部
10C 判定部
10D 画面生成部
130 第1機械学習モデル
131 第2機械学習モデル
Claims (19)
- 対象となるタイヤのトレッドの両端を含むとともに、所定方向に該トレッドが連続するように該トレッドを正面から撮影した画像を取得することと、
前記取得した画像を、学習済みの第1機械学習モデルに入力することと、
前記学習済みの第1機械学習モデルから出力を導出することと
を含み、
前記学習済みの第1機械学習モデルの出力は、前記対象となるタイヤの偏摩耗の推定結果に対応する、
タイヤの状態の推定方法。 - 前記学習済みの第1機械学習モデルから導出された出力に基づいて、前記対象となるタイヤの交換が必要か否かを判定すること
をさらに含む、
請求項1に記載のタイヤの状態の推定方法。 - 前記対象となるタイヤの交換が必要か否かの判定結果を表示する判定結果表示画面を生成すること
をさらに含む、
請求項2に記載のタイヤの状態の推定方法。 - 前記取得した画像を、前記第1機械学習モデルとは異なる学習済みの第2機械学習モデルに入力することと、
前記学習済みの第2機械学習モデルから出力を導出することと
をさらに含み、
前記学習済みの第2機械学習モデルの出力は、前記対象となるタイヤのトレッドの溝の深さの推定結果に対応し、
前記対象となるタイヤの交換が必要か否かを判定することは、前記学習済みの第1機械学習モデルから導出された出力及び前記学習済みの第2機械学習モデルから導出された出力に基づいて前記対象となるタイヤの交換が必要か否かを判定することである、
請求項2または3に記載のタイヤの状態の推定方法。 - 前記対象となるタイヤの偏摩耗の推定結果は、トレッドの偏摩耗の程度についての推定結果であり、偏摩耗の程度を段階的に示す指標で表される、
請求項1から4のいずれか1項に記載のタイヤの状態の推定方法。 - 前記第1機械学習モデルから導出された出力に対応する前記指標を、前記偏摩耗の程度を段階的に示す指標の全体に対する位置付けとともに表示する推定結果表示画面を生成すること
をさらに含む、
請求項5に記載のタイヤの状態の推定方法。 - 前記対象となるタイヤの偏摩耗の推定結果は、偏摩耗の類型についての推定結果である、
請求項1から4のいずれか1項に記載のタイヤの状態の推定方法。 - 前記対象となるタイヤの、車両における取付位置の情報を取得することと、
前記取得された取付位置の情報と、前記第1機械学習モデルから導出された出力とに基づいて、推奨されるタイヤのローテーションパターンである、推奨パターンを決定することと
をさらに含む、
請求項1から7のいずれか1項に記載のタイヤの状態の推定方法。 - 前記決定された推奨パターンを表示するローテーション画面を生成すること
をさらに含む、
請求項8に記載のタイヤの状態の推定方法。 - 前記判定結果表示画面のデータを、ネットワークを介してアクセス可能な状態にすることと、
前記判定結果表示画面のデータに、ネットワークを介してアクセスするための2次元コードを生成することと、
をさらに含む、
請求項3に記載のタイヤの状態の推定方法。 - 前記推定結果表示画面のデータを、ネットワークを介してアクセス可能な状態にすることと、
前記推定結果表示画面のデータに、ネットワークを介してアクセスするための2次元コードを生成することと、
をさらに含む、
請求項6に記載のタイヤの状態の推定方法。 - 対象となるタイヤのトレッドの両端を含むとともに、所定方向に該トレッドが連続するように該トレッドを正面から撮影した画像を取得する画像取得部と、
学習済みの機械学習モデルを記憶する記憶部と、
前記取得した画像を前記学習済みの機械学習モデルに入力し、前記学習済みの機械学習モデルから出力を導出する導出部と、
を備え、
前記学習済みの機械学習モデルの出力は、前記対象となるタイヤの偏摩耗の推定結果に対応する、
タイヤの状態の推定装置。 - 前記学習済みの機械学習モデルから導出された出力に対応する前記推定結果を表示する推定結果表示画面を生成する画面生成部
をさらに備える、
請求項12に記載のタイヤの状態の推定装置。 - 請求項13に記載の推定装置と、
前記対象となるタイヤのトレッドを撮影するカメラと、
前記推定結果表示画面を表示するディスプレイと
を備える、
タイヤの状態の推定システム。 - 対象となるタイヤのトレッドの両端を含むとともに、所定方向に該トレッドが連続するように該トレッドを正面から撮影した画像を取得することと、
前記取得した画像を、学習済みの第1機械学習モデルに入力することと、
前記学習済みの第1機械学習モデルから出力を導出することと
をコンピュータに実行させ、
前記学習済みの第1機械学習モデルの出力は、前記対象となるタイヤの偏摩耗の推定結果に対応する、
タイヤの状態の推定プログラム。 - タイヤのトレッドの両端を含むとともに、所定方向に該トレッドが連続するように該トレッドを正面から撮影した学習用画像と、正解データとのデータセットである学習用データを用意することと、
前記学習用データを用いて、対象となるタイヤのトレッドの両端を含むとともに、所定方向に該トレッドが連続するように該トレッドを正面から撮影した画像を入力すると、前記正解データに対応するデータが出力されるように機械学習モデルのパラメータを調整することと
を含み、
前記正解データは、前記学習用画像のタイヤの偏摩耗の程度を表す指標及び前記学習用画像のタイヤの偏摩耗の類型に対応するラベルの少なくとも一方である、
学習済みモデルの生成方法。 - 対象となるタイヤのトレッドを含む画像を取得することと、
前記取得した画像を、学習済みの第1機械学習モデルに入力することと、
前記学習済みの第1機械学習モデルから出力を導出することと、
前記取得した画像を、前記学習済みの第1機械学習モデルとは異なる学習済みの第2機械学習モデルに入力することと、
前記学習済みの第2機械学習モデルから出力を導出することと、
前記学習済みの第1機械学習モデルから導出された出力及び前記学習済みの第2機械学習モデルから導出された出力に基づいて、前記対象となるタイヤの交換が必要か否かを判定することと、
を含み、
前記学習済みの第1機械学習モデルの出力は、前記対象となるタイヤの偏摩耗の推定結果に対応し、
前記学習済みの第2機械学習モデルの出力は、前記対象となるタイヤのトレッドの溝の深さの推定結果に対応する、
タイヤの状態の推定方法。 - 対象となるタイヤのトレッドを含む画像を取得する画像取得部と、
学習済みの第1機械学習モデルと、前記学習済みの第1機械学習モデルとは異なる学習済みの第2機械学習モデルとを記憶する記憶部と、
前記取得した画像を前記学習済みの第1機械学習モデルに入力するとともに前記取得した画像を前記学習済みの第2機械学習モデルに入力し、前記学習済みの第1機械学習モデル及び前記学習済みの第2機械学習モデルからそれぞれ出力を導出する導出部と、
前記学習済みの第1機械学習モデルから導出された出力及び前記学習済みの第2機械学習モデルから導出された出力に基づいて、前記対象となるタイヤの交換が必要か否かを判定する判定部と
を備え、
前記学習済みの第1機械学習モデルの出力は、前記対象となるタイヤの偏摩耗の推定結果に対応し、
前記学習済みの第2機械学習モデルの出力は、前記対象となるタイヤのトレッドの溝の深さの推定結果に対応する、
タイヤの状態の推定装置。 - 対象となるタイヤのトレッドを含む画像を取得することと、
前記取得した画像を、学習済みの第1機械学習モデルに入力することと、
前記取得した画像を、前記学習済みの第1機械学習モデルとは異なる学習済みの第2機械学習モデルに入力することと、
前記学習済みの第1機械学習モデルから出力を導出することと、
前記学習済みの第2機械学習モデルから出力を導出することと、
前記学習済みの第1機械学習モデルから導出された出力及び前記学習済みの第2機械学習モデルから導出された出力に基づいて前記対象となるタイヤの交換が必要か否かを判定することと、
をコンピュータに実行させ、
前記学習済みの第1機械学習モデルの出力は、前記対象となるタイヤの偏摩耗の推定結果に対応し、
前記学習済みの第2機械学習モデルの出力は、前記対象となるタイヤのトレッドの溝の深さの推定結果に対応する、
タイヤの状態の推定プログラム。
Priority Applications (3)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202280062758.0A CN117957433A (zh) | 2021-10-06 | 2022-09-20 | 轮胎的状态的推测方法 |
| EP22878316.3A EP4390361B1 (en) | 2021-10-06 | 2022-09-20 | Tire state estimation method |
| US18/698,381 US12602800B2 (en) | 2021-10-06 | 2022-09-20 | Tire state estimation method |
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JP2021-164969 | 2021-10-06 | ||
| JP2021164969A JP7020581B1 (ja) | 2021-10-06 | 2021-10-06 | タイヤの状態の推定方法 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2023058445A1 true WO2023058445A1 (ja) | 2023-04-13 |
Family
ID=80948521
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/JP2022/034916 Ceased WO2023058445A1 (ja) | 2021-10-06 | 2022-09-20 | タイヤの状態の推定方法 |
Country Status (5)
| Country | Link |
|---|---|
| US (1) | US12602800B2 (ja) |
| EP (1) | EP4390361B1 (ja) |
| JP (2) | JP7020581B1 (ja) |
| CN (1) | CN117957433A (ja) |
| WO (1) | WO2023058445A1 (ja) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2024254667A1 (pt) * | 2023-06-14 | 2024-12-19 | Robert Bosch Limitada | Método e sistema para determinação de uma condição de pneu, equipamento, método para treinar uma rede neural para estimar a profundidade de um sulco de pneu e memórias lidas por computador correspondentes |
| WO2025074765A1 (ja) * | 2023-10-02 | 2025-04-10 | 株式会社ブリヂストン | 非接触測定装置、非接触測定方法、及びプログラム |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP7697339B2 (ja) * | 2021-10-01 | 2025-06-24 | 住友ゴム工業株式会社 | タイヤ異常判定システム、タイヤ異常判定装置、タイヤ異常判定方法、及びプログラム |
| JP7515189B2 (ja) * | 2022-06-16 | 2024-07-12 | オートペディア カンパニー リミテッド | 深層人工ニューラルネットワークを用いたタイヤトレッド面摩耗判断システム及び方法 |
| JP7841470B2 (ja) * | 2023-04-07 | 2026-04-07 | トヨタ自動車株式会社 | 推定装置 |
| JP2024171796A (ja) * | 2023-05-30 | 2024-12-12 | 株式会社ブリヂストン | リスク予測システム、リスク予測装置およびリスク予測方法 |
| CN120513381A (zh) | 2023-12-13 | 2025-08-19 | 日本精工株式会社 | 状态诊断系统、状态诊断方法以及程序 |
| KR102739087B1 (ko) * | 2024-03-11 | 2024-12-04 | 삼성화재해상보험 주식회사 | 타이어 이미지를 이용한 자동차 보험료 산정 방법 및 시스템 |
| JP2026064895A (ja) | 2024-10-02 | 2026-04-14 | 住友ゴム工業株式会社 | タイヤ残り寿命予測システム、タイヤ残り寿命予測方法、プログラム |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2009102009A (ja) * | 2009-01-23 | 2009-05-14 | Yokohama Rubber Co Ltd:The | タイヤ偏摩耗管理方法 |
| US20170254727A1 (en) * | 2016-03-04 | 2017-09-07 | TIREAUDIT.COM, Inc. | Mesh Registration System and Method for Diagnosing Tread Wear |
| JP2018020752A (ja) * | 2016-07-22 | 2018-02-08 | 横浜ゴム株式会社 | メンテナンス時期提示装置、タイヤ販売管理システム、メンテナンス時期提示方法、タイヤ販売管理方法、メンテナンス時期提示プログラムおよびタイヤ販売管理プログラム |
| JP2019035626A (ja) * | 2017-08-10 | 2019-03-07 | 株式会社ブリヂストン | タイヤ画像の認識方法及びタイヤ画像の認識装置 |
| JP2019184405A (ja) * | 2018-04-09 | 2019-10-24 | 株式会社豊田自動織機 | タイヤ管理システム及びタイヤ管理方法 |
| CN111257019A (zh) * | 2020-01-17 | 2020-06-09 | 杭州中车数字科技有限公司 | 一种跨座式单轨列车部件磨耗检测设备及其检测方法 |
| JP2021063703A (ja) * | 2019-10-11 | 2021-04-22 | 横浜ゴム株式会社 | タイヤ摩耗度推定装置、タイヤ摩耗度学習装置、タイヤ摩耗度推定方法、学習済モデルの生成方法及びプログラム |
| JP2021524040A (ja) | 2018-06-28 | 2021-09-09 | ニッサン ノース アメリカ,インク | ハイブリッド機械学習システム及び方法を用いたタイヤ摩耗推定 |
| JP2021165915A (ja) * | 2020-04-06 | 2021-10-14 | 横浜ゴム株式会社 | 消耗状態学習装置、学習済モデルの生成方法及びプログラム |
Family Cites Families (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9805697B1 (en) * | 2012-06-01 | 2017-10-31 | Hunter Engineering Company | Method for tire tread depth modeling and image annotation |
| DE102012112016A1 (de) | 2012-12-10 | 2014-06-12 | Continental Reifen Deutschland Gmbh | Fahrzeugluftreifen |
| US20150075271A1 (en) * | 2013-09-18 | 2015-03-19 | Tire Profiles, Inc | Modular tire tread analyzing station |
| WO2015097635A1 (en) * | 2013-12-23 | 2015-07-02 | Pirelli Tyre S.P.A. | Method and apparatus for detecting defects on tyres in a tyre production process |
| US9868482B2 (en) * | 2014-10-29 | 2018-01-16 | Caterpillar Inc. | Track roller assembly with a wear measurement system |
| RU2640673C1 (ru) * | 2014-12-22 | 2018-01-11 | Пирелли Тайр С.П.А. | Способ и устройство для детектирования дефектов на шинах в процессе производства шин |
| CA2978482C (en) * | 2015-03-04 | 2020-05-26 | Camso Inc. | Track system for traction of a vehicle |
| CA3042780C (en) * | 2015-06-29 | 2022-06-21 | Camso Inc. | Systems and methods for monitoring a track system for traction of a vehicle |
| WO2017001969A1 (en) * | 2015-06-30 | 2017-01-05 | Pirelli Tyre S.P.A. | Method and apparatus for analysing a surface of a tyre |
| US20180265145A1 (en) * | 2015-09-21 | 2018-09-20 | Soucy International Inc. | Smart track system having embedded sensors and method of using the same |
| US10295333B2 (en) * | 2015-12-30 | 2019-05-21 | Bosch Automotive Service Solutions Inc. | Tire tread depth measurement |
| US9880075B2 (en) * | 2016-02-11 | 2018-01-30 | Caterpillar Inc. | Wear measurement system using a computer model |
| JP6846239B2 (ja) | 2017-03-07 | 2021-03-24 | 株式会社ブリヂストン | タイヤ位置交換提案プログラム及びタイヤ位置交換提案方法 |
| JP2019067221A (ja) | 2017-10-03 | 2019-04-25 | 株式会社ライフ技術研究所 | タイヤからの情報取得システム |
| KR102079862B1 (ko) * | 2018-05-10 | 2020-02-20 | 이도규 | 타이어의 트레드 이미지를 이용한 마모도 측정 시스템 및 방법 |
| CA3111042A1 (en) * | 2018-08-30 | 2020-03-05 | Camso Inc. | Systems and methods for monitoring vehicles with tires |
| US20200070906A1 (en) * | 2018-08-30 | 2020-03-05 | Camso Inc. | Systems and methods for monitoring vehicles |
| US10987977B2 (en) | 2018-10-26 | 2021-04-27 | Toyota Motor North America, Inc. | Systems and methods for measuring tire wear using embedded image sensors |
| WO2020205640A1 (en) * | 2019-04-01 | 2020-10-08 | Exxonmobil Chemical Patents Inc. | System for identifying vehicles and detecting tire characteristics |
| JP7341875B2 (ja) | 2019-12-02 | 2023-09-11 | Toyo Tire株式会社 | タイヤ力表示システムおよびタイヤ力表示方法 |
| JP7495596B2 (ja) | 2020-02-17 | 2024-06-05 | 横浜ゴム株式会社 | 交換支援システム、交換支援方法およびプログラム |
| US20220055699A1 (en) * | 2020-08-18 | 2022-02-24 | Camso Inc. | Monitoring of track system for traction of a vehicle |
| FR3146830B1 (fr) * | 2023-03-21 | 2026-05-01 | Cie Generale Des Etablissements Michelin | Système et procédé de nettoyage automatique de fond de sculpture de pneumatique |
-
2021
- 2021-10-06 JP JP2021164969A patent/JP7020581B1/ja active Active
-
2022
- 2022-01-26 JP JP2022010310A patent/JP2023055602A/ja active Pending
- 2022-09-20 EP EP22878316.3A patent/EP4390361B1/en active Active
- 2022-09-20 US US18/698,381 patent/US12602800B2/en active Active
- 2022-09-20 WO PCT/JP2022/034916 patent/WO2023058445A1/ja not_active Ceased
- 2022-09-20 CN CN202280062758.0A patent/CN117957433A/zh active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2009102009A (ja) * | 2009-01-23 | 2009-05-14 | Yokohama Rubber Co Ltd:The | タイヤ偏摩耗管理方法 |
| US20170254727A1 (en) * | 2016-03-04 | 2017-09-07 | TIREAUDIT.COM, Inc. | Mesh Registration System and Method for Diagnosing Tread Wear |
| JP2018020752A (ja) * | 2016-07-22 | 2018-02-08 | 横浜ゴム株式会社 | メンテナンス時期提示装置、タイヤ販売管理システム、メンテナンス時期提示方法、タイヤ販売管理方法、メンテナンス時期提示プログラムおよびタイヤ販売管理プログラム |
| JP2019035626A (ja) * | 2017-08-10 | 2019-03-07 | 株式会社ブリヂストン | タイヤ画像の認識方法及びタイヤ画像の認識装置 |
| JP2019184405A (ja) * | 2018-04-09 | 2019-10-24 | 株式会社豊田自動織機 | タイヤ管理システム及びタイヤ管理方法 |
| JP2021524040A (ja) | 2018-06-28 | 2021-09-09 | ニッサン ノース アメリカ,インク | ハイブリッド機械学習システム及び方法を用いたタイヤ摩耗推定 |
| JP2021063703A (ja) * | 2019-10-11 | 2021-04-22 | 横浜ゴム株式会社 | タイヤ摩耗度推定装置、タイヤ摩耗度学習装置、タイヤ摩耗度推定方法、学習済モデルの生成方法及びプログラム |
| CN111257019A (zh) * | 2020-01-17 | 2020-06-09 | 杭州中车数字科技有限公司 | 一种跨座式单轨列车部件磨耗检测设备及其检测方法 |
| JP2021165915A (ja) * | 2020-04-06 | 2021-10-14 | 横浜ゴム株式会社 | 消耗状態学習装置、学習済モデルの生成方法及びプログラム |
Non-Patent Citations (1)
| Title |
|---|
| See also references of EP4390361A4 |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2024254667A1 (pt) * | 2023-06-14 | 2024-12-19 | Robert Bosch Limitada | Método e sistema para determinação de uma condição de pneu, equipamento, método para treinar uma rede neural para estimar a profundidade de um sulco de pneu e memórias lidas por computador correspondentes |
| WO2025074765A1 (ja) * | 2023-10-02 | 2025-04-10 | 株式会社ブリヂストン | 非接触測定装置、非接触測定方法、及びプログラム |
Also Published As
| Publication number | Publication date |
|---|---|
| US12602800B2 (en) | 2026-04-14 |
| CN117957433A (zh) | 2024-04-30 |
| JP2023055602A (ja) | 2023-04-18 |
| US20250238942A1 (en) | 2025-07-24 |
| EP4390361A1 (en) | 2024-06-26 |
| EP4390361A4 (en) | 2024-12-25 |
| JP7020581B1 (ja) | 2022-02-16 |
| EP4390361B1 (en) | 2025-11-05 |
| JP2023055522A (ja) | 2023-04-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2023058445A1 (ja) | タイヤの状態の推定方法 | |
| US20200342576A1 (en) | Digital Image Completion by Learning Generation and Patch Matching Jointly | |
| KR102707468B1 (ko) | 신경망, 신경망 학습 방법 및 이미지 신호 처리 튜닝 시스템 | |
| US11921777B2 (en) | Machine learning for digital image selection across object variations | |
| US11068754B1 (en) | Systems and methods regarding image distification and prediction models | |
| US20170185871A1 (en) | Method and apparatus of neural network based image signal processor | |
| US11176414B1 (en) | Systems and methods for 3D image distification | |
| KR20190104282A (ko) | 영상 기반으로 정보를 제공하는 방법 및 이를 위한 이동 단말 | |
| CN112949629B (zh) | 目标检测方法、装置、计算机设备及可读存储介质 | |
| CN109801275B (zh) | 基于图像识别的马铃薯病害检测方法和系统 | |
| Rahman et al. | Deep learning-based left ventricular ejection fraction estimation from echocardiographic videos | |
| CN119850930B (zh) | 一种电子价签位置识别方法及系统 | |
| JP7495596B2 (ja) | 交換支援システム、交換支援方法およびプログラム | |
| JP7428870B2 (ja) | タイヤ摩耗度推定装置、タイヤ摩耗度学習装置、タイヤ摩耗度推定方法、学習済モデルの生成方法及びプログラム | |
| CN114549556B (zh) | 一种图像分割的方法、相关装置、设备以及存储介质 | |
| CN111274946A (zh) | 一种人脸识别方法和系统及设备 | |
| CN114926754B (zh) | 图像检测方法、存储介质及处理器 | |
| US20250316075A1 (en) | Machine Learning for Computation of Visual Attention Center | |
| CN114627069B (zh) | 车轮触地点的获取方法、存储介质和电子设备 | |
| JP2021165915A (ja) | 消耗状態学習装置、学習済モデルの生成方法及びプログラム | |
| KR20250163182A (ko) | 의미론적 편향이 감소된 이미지 복원 모델을 이용하여 저해상도 이미지를 복원하기 위한 전자 장치, 방법, 및 비-일시적 컴퓨터 판독 가능 기록 매체 | |
| JP2015125541A (ja) | 視線予測システム、視線予測方法、および視線予測プログラム | |
| CN116897363A (zh) | 方法、计算机和系统 | |
| CN117150369B (zh) | 超重预测模型的训练方法及电子设备 | |
| US12412425B2 (en) | Estimation system, estimation method, and recording medium recording estimation program |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 22878316 Country of ref document: EP Kind code of ref document: A1 |
|
| WWE | Wipo information: entry into national phase |
Ref document number: 202280062758.0 Country of ref document: CN |
|
| WWE | Wipo information: entry into national phase |
Ref document number: 2022878316 Country of ref document: EP |
|
| ENP | Entry into the national phase |
Ref document number: 2022878316 Country of ref document: EP Effective date: 20240319 |
|
| WWE | Wipo information: entry into national phase |
Ref document number: 18698381 Country of ref document: US |
|
| WWE | Wipo information: entry into national phase |
Ref document number: 202417032538 Country of ref document: IN |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |
|
| WWP | Wipo information: published in national office |
Ref document number: 18698381 Country of ref document: US |
|
| WWG | Wipo information: grant in national office |
Ref document number: 2022878316 Country of ref document: EP |
|
| WWG | Wipo information: grant in national office |
Ref document number: 18698381 Country of ref document: US |