Evaluation method

Abstract

To evaluate a state of rust for each component.SOLUTION: An evaluation method for evaluating a state of rust in components on a rear side of a floor of a vehicle comprises: (A) a step of preparing a machine learning model generated by performing machine learning with the use of an aggregation of learning data sets which includes a learning image obtained by imaging the rear side of the floor of the vehicle and target data including information expressing a range occupied by the component that is at least one component associated with the learning image and arranged on the rear side of the floor in the learning image and that has a predetermined function; (B) a step of preparing an input image obtained by imaging the rear side of the floor of the vehicle being an evaluation object; (C) a step of detecting the range occupied by the component from the input image with the use of the machine learning model; and (D) a step of evaluating a degree of rust for each component based on a hue of the range in the input image.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present disclosure relates to an evaluation method. [Background technology]

[0002] Patent Document 1 describes a technique for determining the damage state of a vehicle using an image of the exterior of the vehicle. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent Publication No. 2021-167991 Summary of the Invention [Problem to be solved by the invention]

[0004] The underside of a vehicle's floor is prone to rust due to the adhesion of rain and mud from the road. If rust occurs on parts placed under the floor of a vehicle and deterioration due to corrosion progresses, the functionality of the parts will be impaired. For this reason, it is desirable to evaluate the rust condition of each part. [Means for solving the problem]

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

[0006] According to a first aspect of the present disclosure, there is provided an evaluation method for evaluating the state of rust on components under a vehicle floor, the evaluation method including the steps of: (A) preparing a machine learning model generated by performing machine learning using a set of training data sets including training images of the underfloor of the vehicle and target data associated with the training images and including information representing an area occupied by at least one component that is located under the underfloor in the training images and that performs a predetermined function; (B) preparing input images of the underfloor of a vehicle to be evaluated; (C) using the machine learning model to detect an area occupied by the component from the input image; and (D) evaluating the degree of rust for each component based on the color tone of the area in the input image. According to the above aspect, the area occupied by the parts placed under the floor is detected from the input image, which makes it easy to evaluate the degree of rust for each part. Furthermore, the degree of rust of the parts under the floor of the vehicle can be easily evaluated based on the color tone of the detected area.

[0007] In the evaluation method of the above form, in step (D), the degree of rust may be evaluated by comparing the color tone within the range with a reference color for each part that is set according to the age and region of the vehicle being evaluated. According to the above aspect, the degree of rust on the underfloor parts of the vehicle can be evaluated taking into account the driving environment. [Brief explanation of the drawings]

[0008] [Figure 1] 1 is a block diagram showing a schematic configuration of an evaluation system according to an embodiment of the present invention. [Figure 2] 10 is a flowchart showing a process executed by the evaluation device. [Figure 3] FIG. 10 is an explanatory diagram of evaluation data. DETAILED DESCRIPTION OF THE INVENTION

[0009] A. Implementation: 1 is a block diagram showing a schematic configuration of an evaluation system 10 according to this embodiment. The evaluation system 10 evaluates the state of rust on underfloor components of a vehicle to be evaluated. The evaluation system 10 includes a camera 50 and an evaluation device 100.

[0010] The camera 50 captures an image of the underside of the floor of the vehicle and supplies the captured image to the evaluation device 100. The camera 50 is capable of communicating with the evaluation device 100 via wireless communication or wired communication.

[0011] The evaluation device 100 is a computer that includes a memory 110 , an interface unit 120 , an input device 130 , a display device 140 , and a CPU 150 .

[0012] The memory 110 stores various programs and data used for various processes executed by the evaluation device 100. In the embodiment, the memory 110 stores data representing a learning model M1, a learning dataset LS, and evaluation data D1, which will be described later. The interface unit 120 is connected to a camera 50, an input device 130, and a display device 140. The input device 130 is, for example, a keyboard or a mouse. The display device 140 is, for example, a liquid crystal display or an organic EL (Electro Luminescence) display. The CPU 150 realizes various functions by executing the programs stored in the memory 110. The CPU 150 functions as a learning unit 210, a detection unit 220, and an evaluation unit 230 by executing the programs stored in the memory 110.

[0013] The learning unit 210 generates a learning model M1 through machine learning using the learning dataset LS. The learning model M1 is a machine learning model that detects a predetermined object from an input image. YOLO, R-CNN, etc. can be used as the learning model M1.

[0014] The detection unit 220 detects parts on the underside of the floor of the vehicle from the input image using the learning model M1. The input image is an image of the underside of the floor of the vehicle to be evaluated.

[0015] The evaluation unit 230 evaluates the degree of rust on the target part using the range of the part detected in the input image. Details of the processes performed by the learning unit 210, the detection unit 220, and the evaluation unit 230 will be described later.

[0016] Fig. 2 is a flowchart showing the processing executed by the evaluation device 100. The processing shown in Fig. 2 is executed by the CPU 150 functioning as the learning unit 210, the detection unit 220, and the evaluation unit 230. For example, when the user instructs the start of processing via the input device 130, the CPU 150 starts the processing shown in Fig. 2.

[0017] In step S101, the CPU 150 generates a learning model M1 by machine learning using the learning dataset LS.

[0018] The training dataset LS includes a set of pairs of training images and target data associated with the training images. The training images are images of the underside of a vehicle floor. The target data includes coordinate value information representing the area occupied by a predetermined component in the training image and the name of the component. The predetermined component is a component located on the underside of the vehicle floor and is to be evaluated for its degree of rust. In the embodiment, a component that performs a predetermined function is set as the evaluation target. Examples of components located on the underside of the vehicle and to be evaluated include a front suspension member, a rear suspension member, a propeller shaft, and a spare tire housing. For example, if the area occupied by a certain component in the training image is rectangular, the target data includes coordinate values ​​of four points representing that area. For example, a user with some knowledge of vehicle undersides uses an annotation tool to set coordinate value information representing the area occupied by a predetermined component. Furthermore, even for the same type of component, the shape, size, location, etc. of the component vary depending on the vehicle model. For this reason, images of the undersides of vehicles of various types are used as learning images.

[0019] In step S102, CPU 150 uses learning model M1 to detect the area occupied by the vehicle underfloor parts from the input image. The input image is stored in memory 110 in advance. The input image is an image captured by camera 50 of the underfloor part of the vehicle whose rust level is to be evaluated. CPU 150 outputs, as detection results, the names of the detected parts, coordinate value information indicating the area occupied by the detected parts in the input image, and a reliability score indicating the reliability of the detection result. The reliability score is expressed as a value indicating the probability that the detection result is reliable. CPU 150 displays, for example, an image on display device 140 in which a frame indicating the area occupied by the detected parts is superimposed on the input image.

[0020] In step S103, CPU 150 evaluates the degree of rust of the detected under-floor parts by comparing the area occupied by the under-floor parts detected in step S102 with evaluation data D1. Evaluation data D1 is stored in memory 110 in advance.

[0021] FIG. 3 is an explanatory diagram of the evaluation data D1. The evaluation data D1 includes a plurality of sample images IM and a table TB in which a value representing the degree of rust for each sample image IM is defined. A plurality of sample images IM used to evaluate the degree of rust are prepared. The plurality of sample images IM include images of parts with different degrees of rust. The sample images IM include images of two or more identical parts with different degrees of rust. The table TB includes information representing the correspondence between the part names of the underfloor parts, the image ID which is identification information for the sample image IM, and the rust level which is a value representing the degree of rust.

[0022] As shown in FIG. 2, CPU 150 compares the color tones of the range detected in step S102 with those of sample image IM and selects a sample image IM having a color tone similar to that of the detected range. Color tone refers to differences in color tone due to the relationship between lightness, saturation, and hue. The color tone of sample image IM is also called a reference color. CPU 150 references table TB and stores the rust level value associated with the selected sample image IM in memory 110 as the evaluation result of the degree of rust of the target part. For example, if the color tones of 60% of the range detected in step S102 are similar to the color tone of sample image IM, CPU 150 uses the rust level associated with that sample image IM as the evaluation result of the degree of rust of the target part.

[0023] In step S104, CPU 150 determines whether the rust degree has been evaluated for all parts detected in the input image. If the rust degree of all detected parts has been evaluated (step S104; YES), the process shown in Fig. 2 is terminated. If the rust degree of all detected parts has not been evaluated (step S104; NO), CPU 150 executes the process of step S103 again.

[0024] As described above, in the evaluation method according to the embodiment, the area occupied by the components placed under the floor is detected from the input image, making it easy to evaluate the degree of rust for each component. Furthermore, the degree of rust for the components under the floor of the vehicle can be easily evaluated based on the color tone of the detected area. Furthermore, the presence or absence of defects for each component can be detected based on the results of the evaluation of the degree of rust.

[0025] B. Other Embodiments: B1. Alternative Embodiment 1: The evaluation device 100 may also evaluate the degree of rust taking into account the age of the vehicle being evaluated. In this case, evaluation data is prepared in advance for each predetermined age category. The age category may be, for example, 0 to 5 years, 6 to 10 years, 10 to 15 years, or 15 years or more. When evaluating the degree of rust, the CPU 150 receives input of the age of the vehicle being evaluated from the user. The CPU 150 evaluates the degree of rust of the detected parts using evaluation data corresponding to the category corresponding to the specified age. Thus, the degree of rust of the underfloor parts of the vehicle can be evaluated taking into account the age of the vehicle. For example, data evaluating the degree of rust according to the age of the vehicle is collected. The collected data can be used to evaluate the durability of the underfloor parts, and the evaluation results can be used in vehicle product development.

[0026] Alternatively, the evaluation device 100 may evaluate the degree of rust by taking into account the age and region of the vehicle to be evaluated. In this case, evaluation data is prepared in advance according to a predetermined category for age and a predetermined category for region. The region may be, for example, North America, Central and South America, Europe, Africa, Asia, or the Middle East. For example, evaluation data is prepared for a vehicle to be evaluated whose region is North America and whose age is 0 to 5 years. Evaluation data is also prepared for a vehicle to be evaluated whose region is North America and whose age is 6 to 10 years. When evaluating the degree of rust, the CPU 150 receives input from the user of the region and age of the vehicle to be evaluated. The CPU 150 evaluates the degree of rust of the detected parts using evaluation data corresponding to the specified region and age. Therefore, the degree of rust of parts under the vehicle floor can be evaluated by taking into account the vehicle's driving environment. For example, data evaluating the degree of rust according to the vehicle's region and age is collected. The collected data can be used to evaluate the durability of underfloor components, and the evaluation results can be used in vehicle product development.

[0027] B2. Alternative Embodiment 2: The learning unit 210 may also generate a learning model M1 through machine learning using the learning dataset LS, and then verify the accuracy of the generated learning model M1. The verification is performed using a verification dataset different from the learning dataset LS. The verification dataset includes a set of pairs of verification images and target data associated with the verification images. The verification images are images of the underside of a vehicle floor. The target data includes coordinate value information representing the area occupied by a predetermined component in the verification image and the name of the component. The learning unit 210 uses the learning model M1 to detect the area occupied by the underside of the vehicle floor from the verification image. During verification, the detection error by the learning model M1 is calculated by comparing the detected area with the target data. Furthermore, the detection accuracy is calculated, for example, based on the error resulting from detecting the area indicated by the underside component from multiple detection images. If the detection accuracy does not meet a predetermined standard, the learning unit 210 can perform additional learning to update the learning model M1.

[0028] B3. Alternative Embodiment 3: Furthermore, the evaluation device 100 may evaluate the degree of rust only for parts designated by the user among the underfloor parts that can be detected using the learning model M1. For example, the CPU 150 displays a list of underfloor parts that can be detected using the learning model M1 on the display device 140. The user designates the parts to be evaluated via the input device 130. The CPU 150 evaluates the degree of rust only for the designated parts. In this case, evaluation is performed only for the parts that require evaluation, thereby reducing the processing load on the evaluation device 100.

[0029] B4. Alternative Embodiment 4: Furthermore, the training images included in the training dataset LS may be processed images of the underside of a vehicle floor. For example, images of the underside of a vehicle floor that are flipped horizontally may be used as training images. Furthermore, enlarged images of the underside of a vehicle floor may be used as training images.

[0030] B5. Alternative Embodiment 5: 2, an example has been described in which the process of generating the learning model M1 (see step S101 in FIG. 2) is followed by the process of evaluating the degree of rust (see steps S102 to S104 in FIG. 2). However, the process of generating the learning model M1 and the process of evaluating the degree of rust do not necessarily have to be performed consecutively.

[0031] The present disclosure is not limited to the above-described embodiments and can be realized in various configurations without departing from the spirit thereof. For example, the technical features in the embodiments corresponding to the technical features in each aspect described in the Summary of the Invention section can be appropriately replaced or combined to solve some or all of the above-described problems or achieve some or all of the above-described effects. Furthermore, if a technical feature is not described as essential in this specification, it can be deleted as appropriate.

[0032] The present disclosure may be realized in various forms other than those described above, for example, in the form of a non-transitory storage medium on which a computer program is recorded. [Explanation of symbols]

[0033] 10...Evaluation system, 50...Camera, 100...Evaluation device, 110...Memory, 120...Interface unit, 130...Input device, 140...Display device, 150...CPU, 210...Learning unit, 220...Detection unit, 230...Evaluation unit, D1...Evaluation data, IM...Sample image, LS...Learning dataset, M1...Learning model, TB...Table

Claims

1. 1. A method for evaluating the state of rust on underfloor components of a vehicle, comprising: (A) preparing a machine learning model generated by performing machine learning using a set of learning datasets including: learning images of the underfloor of the vehicle; and target data associated with the learning images and including information representing an area occupied by at least one component that is located on the underfloor in the learning images and that performs a predetermined function; (B) preparing an input image of the underside of the vehicle to be evaluated; (C) detecting an area occupied by the component from the input image using the machine learning model; (D) evaluating the degree of rust for each of the components based on the range of color tones in the input image; Evaluation methods including.

2. The evaluation method according to claim 1, In step (D), the degree of rust is evaluated by comparing the color tone within the range with a reference color for each of the parts that is set according to the age and region of the vehicle to be evaluated. Evaluation method.

Images