JP2020067669A - Information processing apparatus, and program - Google Patents

Abstract

To provide a technique for efficiently and accurately recognizing a number plate.SOLUTION: Provided is an information processing apparatus comprising: algorithm acquisition means for acquiring a predetermined algorithm obtained as a result of learning taken place by applying a method related to machine learning, on data related to an image for learning including the image including a plurality of number plates in a subject and the image created by processing the images; user information acquisition means for acquiring data on the image including characters, a color region, and a background in the subject in addition to the predetermined number plate; and algorithm applying means for outputting a result obtained by applying the predetermined algorithm acquired by the algorithm acquiring means to the data acquired by the user information acquisition means.SELECTED DRAWING: Figure 3

Description

  The present invention relates to an information processing device and a program.

2. Description of the Related Art In recent years, a device for recognizing an automobile license plate from an image has been used for purposes such as marketing and safety management. There are various recognition methods, but a method using machine learning has been generalized (for example, refer to Patent Document 1).
This patent document 1 describes a technique for recognizing a license plate by using deep learning which is a kind of machine learning.

JP, 2018-120445, A

  However, with only the conventional technology including Patent Document 1, it is necessary to prepare a large number of license plate images and the like for learning, and problems such as accuracy and cost have been pointed out.

  The present invention has been made in view of such circumstances, and an object of the present invention is to provide a technique for efficiently and highly accurately recognizing a license plate.

In order to achieve the above object, an information processing device of one embodiment of the present invention is
A predetermined value obtained as a result of learning by applying a method related to machine learning to data related to an image for learning including an image including a plurality of license plates as subjects and an image created by processing the images. Algorithm acquisition means for acquiring the algorithm,
In addition to the predetermined license plate, a user information acquisition unit that acquires data regarding an image including a character, a color region, and a background as a subject,
Algorithm applying means for outputting a result obtained by applying the predetermined algorithm acquired by the algorithm acquiring means to the data acquired by the user information acquiring means,
Equipped with.

  A program according to one aspect of the present invention is a program corresponding to the above-described information processing device according to one aspect of the present invention.

  According to the present invention, it is possible to provide a technology for efficiently and accurately recognizing a license plate.

It is a figure showing an example of an information processing system concerning one embodiment of the present invention. It is a block diagram which shows the hardware constitutions of the inference server in the information processing system of FIG. It is a functional block diagram which shows an example of a functional structure of an inference server, a user terminal, and a learning server in the information processing system of FIG. It is a flowchart explaining the flow of the license plate learning process performed by the learning server of FIG. 4 is a flowchart illustrating a flow of license plate identification processing executed by the inference server of FIG. 3. It is a figure which shows the specific example of the license plate character data utilized in the learning server of FIG. It is a figure which shows the specific example of the process of the license plate image performed in the learning server of FIG. It is a figure which shows the specific example of the image for learning which synthesize | combined the license plate image and the background image which were produced | generated in the learning server of FIG.

Embodiments of the present invention will be described below with reference to the drawings. Therefore, first, an outline of the information processing system according to the embodiment of the present invention will be described with reference to FIG.
Here, the license plate recognition service to which the information processing system according to the embodiment of the present invention is applied will be referred to as this service in the following description.

The information processing system shown in FIG. 1 is configured to include an inference server 1, a user terminal 2, and a learning server 3.
The inference server 1, the user terminal 2, and the learning server 3 are connected to each other via a network N such as the Internet.
The inference server 1 is managed by the administrator of this service. The inference server 1 applies a predetermined algorithm to an image (hereinafter, referred to as a “license plate image”) of the license plate captured from the user terminal 2 to recognize the license plate, for example. Used to do.
The user terminal 2 is used by a user who wants to provide this service. The user terminal 2 is used, for example, to transmit an image in which a license plate captured by a user or the like is captured to the inference server 1.
The learning server 3 is managed by the service manager or a vendor in charge of learning. The learning server 3 is used, for example, to perform learning based on the data of images in which a plurality of license plates are captured.

  FIG. 2 is a block diagram showing the hardware configuration of the inference server 1 in the information processing system of FIG.

  The inference server 1 includes a CPU (Central Processing Unit) 11, a ROM (Read Only Memory) 12, a RAM (Random Access Memory) 13, a bus 14, an input / output interface 15, an output unit 16, and an input unit 17. The storage unit 18, the communication unit 19, and the drive 20 are provided.

The CPU 11 executes various processes according to the program recorded in the ROM 12 or the program loaded from the storage unit 18 into the RAM 13.
The RAM 13 also stores data necessary for the CPU 11 to execute various processes.

  The CPU 11, ROM 12, and RAM 13 are connected to each other via a bus 14. An input / output interface 15 is also connected to the bus 14. An output unit 16, an input unit 17, a storage unit 18, a communication unit 19, and a drive 20 are connected to the input / output interface 15.

The output unit 16 is composed of a display, a speaker, and the like, and outputs various kinds of information as images and sounds.
The input unit 17 is composed of a keyboard, a mouse and the like, and inputs various information.

The storage unit 18 includes a hard disk, a DRAM (Dynamic Random Access Memory), or the like, and stores various data.
The communication unit 19 communicates with other devices (the user terminal 2 and the learning server 3 in the example of FIG. 1) via the network N including the Internet.

A removable medium 21 including a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, or the like is appropriately mounted on the drive 20. The program read from the removable medium 21 by the drive 20 may be installed in the storage unit 18 as needed.
Further, the removable medium 21 can also store various data stored in the storage unit 18 in the same manner as the storage unit 18.

  Although not shown, the hardware configurations of the user terminal 2 and the learning server 3 in the information processing system of FIG. 1 can be basically the same as the hardware configuration of the inference server 1. Then, the description is omitted.

Next, the functional configurations of the inference server 1, the user terminal 2, and the learning server 3 having such a hardware configuration will be described with reference to FIG.
FIG. 3 is a functional block diagram showing an example of the functional configuration of the inference server 1, the user terminal 2, and the learning server 3 in the information processing system of FIG.
Here, various processes related to this service are roughly classified into a learning process and a license plate identification process which will be described later. Therefore, first, the learning process executed mainly in the inference server 1 and the user terminal 2 will be described.

As shown in FIG. 3, in the CPU 31 of the learning server 3, the data acquisition unit 61, the data processing unit 62, the learning image generation unit 63, the position coordinate data generation unit 64, and the learning unit 65 function. .
In the data processing unit 62, the license plate image generation unit 71, the bump processing unit 72, the noise processing unit 73, the shadow processing unit 74, and the distortion processing unit 75 function.
A license plate DB 100 and a learning process DB 200 are provided in one area of the storage unit 38.

The data acquisition unit 61 of the learning server 3 acquires data of an image of a plurality of license plates (hereinafter referred to as “license plate data”) from a database or the like (not shown), and obtains the information from the license plate DB 100. To store.
In addition to the license plate data, the license plate DB 100 also stores character areas, color areas, and background data, which are components of the license plate.

  The data processing unit 62 appropriately extracts the license plate data stored in the license plate DB 100 and processes the data. The specific processing is as follows.

The license plate image generator 71 processes characters and colors to generate a new learning image.
That is, the license plate image includes four types of character areas: area name, classification number, hiragana, and number. The license plate image generation unit 71 determines the character region portion included in the license plate image by arbitrarily selecting and combining the contents of the character region included in the license plate image stored in the license plate DB 100, for example. , New license plate data for learning (hereinafter referred to as “processed license plate data”) can be generated.

Here, the method of determining the character area portion included in the license plate image is not particularly limited, but for example, for the area name, one area name is randomly selected from all areas, and one character is also used for hiragana. At random, select either 2 digits or 3 digits for the classification number at random, randomly select one character from 0 to 9 for each number of digits, and the number is an integer from 0 to 9999. The rule is to select one and insert a hyphen between the 2nd and 3rd digits if there are 4 digits, and pad to the right with "・" if the number of digits is less than 4 digits. You can decide.
Further, the license plate image generation unit 71 may change the color of the plate in the license plate image to generate the processed license plate data.
That is, the license plate image generation unit 71 has four types, that is, a green character on a white background, a white character on a green background, a black character on a yellow background, and a yellow character on a black background, depending on whether the vehicle is used or not. The license plate image generation unit 71 may randomly select and combine the above-described four types of colors with respect to the determined character portion of the license plate image to generate processed license plate data.

The bump processing unit 72 generates processed license plate data by subjecting the character area portion of the license plate image to bump processing in which the character portion is raised three-dimensionally.
Specifically, for example, the bump processing unit 72 performs bump processing by randomly determining the shadow direction between the directions from 4 o'clock to 8 o'clock when the position directly above the license plate image is 12 o'clock.
By this bump processing, the character portion becomes three-dimensional, and highlights and shadows occur around the character portion. By this bump processing, the bump processing unit 72 can generate a plurality of processed license plate data in which the contour of the character portion of the license plate has different shades.

The noise processing unit 73 generates processed license plate data by performing noise processing such as dirt and shine on the character area portion of the license plate image.
Due to the noise such as dirt and shininess, the actual license plate may affect the outline of characters. By this noise processing, the noise processing unit 73 can generate a plurality of processed license plate data in which the contour of the character portion of the license plate image is different.
Depending on the storage system power of the image, noise may naturally occur due to the characteristics of the storage format (for example, JPEG block noise or the like).

The shading processing unit 74 generates shading license plate data by performing shading processing on the license plate image by adding shading due to the influence of the sunlight and the shape of the body of the automobile.
The shading processing unit 74 performs shading so that shadows randomly fall from above between the directions from 3 o'clock to 9 o'clock when the top of the license plate image is 12 o'clock. Since there are few situations in which shadows enter from the bottom to the top, the width of the angle in the shadow direction may be narrower than in bump processing.
By this shading processing, the shading processing unit 74 can generate a plurality of processed license plate data in which the shade of the contour of the character portion of the license plate is different.

The distortion processing unit 75 generates processed license plate data by performing distortion processing that adds distortion when the license plate image is moved in the three-dimensional direction.
Specifically, for example, the distortion processing unit 75 randomly inclines up to ± 30 degrees with respect to each of the X-axis, Y-axis, and Z-axis of the license plate image from the state in which the license plate image faces directly in front.
By this distortion processing, the distortion processing unit 75 can generate a plurality of processed license plate data in which the appearance of the character area portion of the license plate is different.

  The learning image generating unit 63 synthesizes the background image stored in the license plate DB 100 with the license plate data stored in the license plate DB 100 or the above-mentioned processed license plate data, thereby learning for learning. Generate an image. The background image used for composition may be randomly determined, for example. The background image referred to here may be any image as long as it does not include the license plate.

The position coordinate data generation unit 64 generates coordinate data of the positions of the four corners of the license plate portion in the learning image. The position coordinate data is data that is learned together with the learning image during the processing in the learning unit described later.
By generating the coordinate data of the four corner positions of the license plate in this way, the four corners of the license plate are identified from the external information of the license plate, such as the positional relationship between the license plate and the surrounding environment where the license plate is installed. It is possible to prevent the algorithm from being generated.

The learning unit 65 is based on the license plate data stored in the license plate DB 100 and the learning image generated by the learning image generation unit 63, characters and colors of the license plate, position coordinate data of four corners of the license plate, and the like. Then, a method related to machine learning is applied and learned to output a license plate identification algorithm (hereinafter, referred to as “identification algorithm”).
Here, the method regarding machine learning adopted by the learning unit 65 is not particularly limited and is arbitrary.
Further, the learning unit 65 stores the output identification algorithm in the learning processing DB 200 and transmits it to the inference server 1 via the communication unit 39.
The various functional configurations that can function in the learning process of the learning process and the license plate identification process related to the service have been briefly described above. Next, various functional configurations that can function in the license plate identification process will be briefly described.

As shown in FIG. 3, in the CPU 11 of the inference server 1, the data management unit 51, the algorithm acquisition unit 52, the identification unit 53, and the result output unit 54 function.
In the license plate identification process described later, image data including a license plate that the user desires to identify from the user terminal 2 or the like to the inference server 1 (hereinafter, referred to as “identification desired image data”). Has been sent.

The data management unit 51 of the inference server 1 acquires the identification desired image data transmitted from the user terminal 2 or the like via the communication unit 19.
Here, the above-mentioned identification desired image data also includes various data regarding a character area, a color area, and a background, which are constituent elements of the license plate.

  The algorithm acquisition unit 52 acquires the license plate identification algorithm from the learning server 3 via the communication unit 19.

The identification unit 53 applies the identification algorithm of the license plate acquired from the learning server 3 to the identification desired image data acquired by the data management unit 51 to identify the identification desired image data.
Here, the identification of the identification desired image data may include character recognition of the license plate image included in the identification desired image data, or may be correct or incorrect of the character information of the license plate image.
The result output unit 54 transmits the result identified by the identification unit 53 to the user terminal 2 via the communication unit 19.

Next, the license plate learning process executed by the learning server 3 having the functional configuration of FIG. 2 will be described with reference to FIG.
FIG. 4 is a flowchart illustrating the flow of the license plate learning process executed by the learning server 3.

As shown in FIG. 4, the learning server 3 executes the following series of processes.
In step S1, the license plate data stored in the license plate DB 100 is extracted as appropriate and processed.

  In step S2, the license plate image generation unit 71 processes the data regarding the character area and the color area, which are the constituent elements of the license plate, stored in the license plate DB 100 to generate a license plate image.

  In step S3, the bump processing unit 72 performs bump processing for three-dimensionally raising the character portion of the license plate image to generate processed license plate data.

  In step S4, the noise processing unit 73 performs noise processing such as dirt and shine on the character portion of the license plate image to generate processed license plate data.

  In step S5, the shading processing unit 74 performs shading processing to add shading due to the influence of the sunlight and the shape of the body of the automobile, and generates processed license plate data.

  In step S6, the distortion processing unit 75 performs distortion processing that adds distortion when the license plate image is moved in the three-dimensional direction, and generates processed license plate data.

  In step S7, the learning image generation unit 63 synthesizes the background image stored in the license plate DB 100 with the license plate data stored in the license plate DB 100 or the above-described processed license plate data to perform learning. Generate a learning image for.

  In step S8, the position coordinate data generation unit 64 generates coordinate data of the positions of the four corners of the license plate portion in the learning image.

  In step S9, the learning unit 65 causes the license plate data acquired by the data acquisition unit 61 and the learning image generated by the learning image generation unit 63, the license plate character data, the color data, and the position coordinates of the four corners of the license plate. A method related to machine learning is applied to the data to generate a license plate identification algorithm.

In step S10, the learning unit 65 stores the output identification algorithm in the learning processing DB 200 and transmits it to the inference server 1 via the communication unit 39.
The license plate learning process ends through the above processes.

  In this way, by processing license plates, a large number of license plates can be learned and a highly accurate license plate identification algorithm can be generated.

Next, the license plate identification processing executed by the inference server 1 having the functional configuration of FIG. 2 will be described with reference to FIG.
FIG. 5 is a flowchart for explaining the flow of license plate identification processing executed by the inference server 1.

As shown in FIG. 5, the inference server 1 executes the following series of processes.
In step S21, the data management unit 51 acquires the identification desired image data from the user terminal 2 via the communication unit 19.

In step S22, the algorithm acquisition unit 52 acquires the identification algorithm sent from the learning server 3.
In step S23, the identification unit 53 applies an identification algorithm to the identification desired image data acquired by the data management unit 51 to identify the identification desired image data.

In step S24, the result output unit 54 transmits the identification result of the identification desired image data to the user terminal 2 via the communication unit 19.
In step S25, the data management unit 51 confirms whether new identification desired image data has been transmitted. As a result of the confirmation, if new desired identification image data is acquired, the process returns to step SS21 and the license plate identification process is performed again. Here, if the new identification desired image data has not been acquired, the license plate identification processing ends.

  In this way, by performing the license plate identification process each time new desired identification image data is acquired, the license plate can always be identified with high accuracy by the latest algorithm.

Next, the characters included in the license plate image in the learning server 3 will be briefly described with reference to FIG.
FIG. 6 is a diagram showing an example of characters used in the license plate in the learning server 3.

In the example of FIG. 6A, the characters "Naniwa" are displayed as an example of the area name M1. In the example of FIG. 6B, the character “530” is displayed as an example of the classification number M2. In the example of FIG. 6C, the character "A" is displayed as an example of the hiragana M3. In the example of FIG. 6D, the character “00-00” is described as an example of the number M4. In reality, there is no license plate starting from 0 in Japan, but here, for convenience of explanation, "00-00" is assumed to exist.
Here, the “region name”, “classification number”, “hiragana”, and “number” in this embodiment will be described. The "region name", "classification number", "hiragana", and "number" are display formats unique to Japanese license plates.
The "region name" indicates the region name in which the displayed license plate is registered. That is, in the example of FIG. 6A, for example, the registration area of the license plate P1 on which the characters "Naniwa" are displayed is Naniwa (Naniwa automobile inspection registration office).
The “classification number” is a number used to classify the type and purpose of use of the automobile on which the license plate is installed. Specifically, for example, in the example of FIG. 6B, the license plate P2 on which the characters of the classification number “530” are displayed indicates that the license plate P2 is, for example, “small passenger”.
The “hiragana” is used to classify the application of the automobile in which the license plate is installed, etc., like the “classification number” described above. However, the contents of classification are slightly different.
The "number" is a number for uniquely identifying the displayed license plate.
As described above, the characters displayed on the license plate may be determined according to a predetermined rule defined in advance by the Road Traffic Law or the like.

That is, since there are predetermined rules based on the Road Traffic Act and the like for the license plate classification number and the like, the learning server 3 may improve the accuracy of learning by performing learning including these rules. is there.
On the other hand, since these rules have a high possibility of being changed, the above-mentioned license plate image generating unit 71 randomly selects a character and intentionally changes these rules as if the processed license plate data was generated. It does not have to be included in learning.
In other words, the learning server 3 may adopt only an arbitrary rule out of various rules regarding the license plate and use it for learning.

Further, the learning unit 65 of the learning server 3 individually applies the machine learning method to the images including the constituent elements shown in (a) to (d) of FIG. 6 among the learning images. You can also study.
Further, the learning unit 65 can generate not only the identification algorithm for the entire license plate but also the identification algorithm for each component of the license plate as the identification algorithm for the license plate.

In this way, learning each of the constituent elements separately is particularly useful when a new learning image is added and relearning is performed.
That is, the structure of the license plate may change due to changes in laws and regulations. As described above, if machine learning is performed for each component of the license plate, even if there is a change in the structure of the license plate, re-learning only at the changed portion will provide a sufficient identification system. Expected to be able to keep.

Next, processing of the license plate image executed in the learning server 3 will be described with reference to FIG. 7.
FIG. 7 is a diagram showing a specific example of the processing of the license plate image executed in the learning server 3.

FIG. 7A shows the license plate K1 after processing in which a shadow is formed as a result of bump processing by three-dimensionally raising the character "NA". FIG.7 (b) has shown the license plate K2 after the process which processed the noise process by making the upper part of the license plate dirty. FIG. 7 (c) shows a license plate K3 after being processed by distortion by moving the license plate as viewed from the right side, and a plate after processing by moving the license plate as viewed from above and being subjected to distortion processing. The license plate K4 is shown.
As described above, the data processing unit 62 can generate a large number of license plate images that are the basis of the learning image by performing various processes on the license plate.

Next, a learning image in which the license plate image and the background image are combined in the learning server 3 will be described with reference to FIG.
FIG. 8 is a diagram showing a specific example of a learning image in which the license plate image and the background image generated are combined in the learning server 3.

FIG. 8 shows a learning image G in which the license plate P5 is combined with the background image.
As described above, the learning image generation unit 63 can generate a large number of learning images by synthesizing the randomly selected background image with the license plate image generated and variously processed. .

  Although the embodiments of the present invention have been described above, the present invention is not limited to the above-described embodiments, and modifications, improvements, etc. within the scope of achieving the object of the present invention are included in the present invention. Is. In addition, the effects described in the present embodiment are merely enumeration of the most suitable effects resulting from the present invention, and the effects according to the present invention are not limited to those described in the present embodiment.

Further, as the license plate character data, other than the one acquired from the inference server, one generated by the learning server 3 by a predetermined algorithm may be used. When the learning server 3 generates the characters to be displayed on the license plate, the learning server 3 may be randomly generated, or in consideration of the above rule, the characters to be displayed on the license plate are generated under certain restrictions. May be.
Further, in generating the license plate image, the data processing unit 62 may be subjected to various kinds of processing other than those described in this specification. For example, frame processing for combining a frame around the license plate may be performed.

Further, for example, in the above-described embodiment, the image that the learning unit 65 performs machine learning is not particularly limited, and the image acquired by the data acquisition unit 61 may be directly learned. The learning image does not have to be completely processed in the data processing unit, and may be partially processed or may not be processed at all.
Furthermore, when the learning unit 65 executes machine learning, the position coordinate data of the four corners of the license plate is not always necessary, and data without this data can be learned.

Further, for example, in the above-described embodiment, when various processes are executed by the inference server 1 or the learning server 3, only necessary parts may be cut off, unnecessary part image quality may be reduced, or the like.
As a result, the inference server 1 or the learning server 3 can reduce the amount of calculation required for various processes and can perform various processes in a short time.

Further, for example, when a series of processes is executed by software, a program forming the software is installed in a computer or the like from a network or a recording medium.
The computer may be a computer embedded in dedicated hardware. Further, the computer may be a computer capable of executing various functions by installing various programs, for example, a general-purpose personal computer.

  Further, for example, a recording medium including such a program is not only configured by a removable medium that is distributed separately from the device body in order to provide the program to the user, but is also installed in the device body in advance to the user. It is composed of the provided recording medium and the like. The removable medium is composed of, for example, a magnetic disk (including a floppy disk), an optical disk, a magneto-optical disk, or the like. The optical disk is configured by, for example, a CD-ROM (Compact Disk-Read Only Memory), a DVD (Digital Versatile Disk), or the like. The magneto-optical disk is composed of MD (Mini-Disk) or the like. In addition, the recording medium provided to the user in a state of being pre-installed in the apparatus main body is constituted by, for example, a ROM in which a program is recorded, a hard disk included in a storage unit, or the like.

In the present specification, the steps of writing the program recorded on the recording medium and the sales steps are not limited to the processing performed in time series according to the order, but are not necessarily performed in time series, but may be performed in parallel. Alternatively, it also includes processing that is executed individually.
In addition, in the present specification, the term “system” means an overall device including a plurality of devices and a plurality of means.

  Further, for example, the series of processes described above can be executed by hardware or software.

  Further, for example, each hardware configuration shown in FIG. 3 is merely an example for achieving the object of the present invention, and is not particularly limited.

Further, for example, the functional block diagram shown in FIG. 3 and the flowchart shown in FIG. 4 and FIG. 5 are merely examples, and are not particularly limited. That is, it is sufficient if the information processing system has a function capable of executing the above-described series of processes as a whole, and what kind of functional block is used to realize this function is not particularly limited to the example of FIG. .
Further, the location of the functional block is not particularly limited to that shown in FIG. 3, and may be arbitrary.
Further, one functional block may be configured by hardware alone, software alone, or a combination thereof.
Furthermore, these functional blocks do not have to exist in each of the inference server 1, the user terminal 2 and the learning server 3 described above, and are optional.
Specifically, for example, the user terminal 2 and the learning server 3 may be configured to include the function of the inference server 1 and vice versa.

In summary, the information processing apparatus to which the present invention is applied may have various configurations as long as it has the following configuration.
That is, the information processing apparatus to which the present invention is applied (for example, the inference server 1 in FIG. 3) is
A predetermined value obtained as a result of learning by applying a method related to machine learning to data related to an image for learning including an image including a plurality of license plates as subjects and an image created by processing the images. An algorithm acquisition unit (for example, the algorithm acquisition unit 52 in FIG. 3) that acquires an algorithm,
In addition to a predetermined license plate, a user information acquisition unit (for example, the data management unit 51 in FIG. 3) that acquires data regarding an image including a character, a color region, and a background as a subject,
Algorithm applying means for outputting a result obtained by applying the predetermined algorithm acquired by the algorithm acquiring means to the data acquired by the user information acquiring means (for example, the identifying unit 53 in FIG. 3). When,
Equipped with.
As a result, a large amount of learning data regarding license plates can be prepared for highly accurate license plate recognition.

Further, the algorithm acquisition means is a bump processing means (for example, the bump processing part 72 in FIG. 3) that three-dimensionally raises the character portion of the license plate image, and noise such as dirt and shine on the character portion of the license plate image. A noise processing means (for example, the noise processing section 73 in FIG. 3) to be added, and a shadow processing means (for example, shadow in FIG. 3) for adding a shadow to the character portion of the license plate image due to the influence of the sunlight or the shape of the body of the automobile. Processing unit 74), and at least one processing unit selected from the group consisting of a distortion processing unit (for example, the distortion processing unit 75 of FIG. 3) that adds distortion when the license plate image is moved in the three-dimensional direction. The predetermined algorithm determined using the data about the learning image, including at least the image created by processing, It can be Tokusuru.
As a result, a large amount of learning data regarding license plates can be prepared, and highly accurate license plate recognition can be performed.

Further, the information processing device (for example, the inference server 1 in FIG. 3) is
All or part of the data relating to the plurality of license plate images and the learning image can be acquired based on the data of the position coordinates of the four corners of the license plate.
With this, by generating coordinate data of the positions of the four corners of the license plate, the four corners of the license plate can be identified from external information of the license plate, such as some positional relationship between the license plate and the surrounding environment. It is possible to prevent the algorithm from being generated.

Further, the information processing device (for example, the inference server 1 in FIG. 3) is
The algorithm acquisition means, with respect to the data related to the learning image, a region name portion of the license plate, a classification number portion of the license plate, a hiragana portion of the license plate, and a number portion of the license plate. It is possible to separately apply a method related to machine learning and obtain a determined predetermined algorithm.
Thereby, even if the configuration of the license plate is changed, only the changed portion can be relearned.

  1 ... inference server, 2 ... user terminal, 3 ... learning server, 11 ... CPU, 12 ... ROM, 13 ... RAM, 14 ... bus, 15 ... Output interface, 16 ... Display unit, 17 ... Input unit, 18 ... Storage unit, 19 ... Communication unit, 20 ... Drive, 21 ... Removable media, 31 ... CPU, 38 ... storage unit, 39 ... communication unit, 51 ... data management unit, 52 ... algorithm acquisition unit, 53 ... identification unit, 54 ... result output unit, 61 ... data Acquisition unit, 62 ... Data processing unit, 63 ... Learning image generation unit, 64 ... Position coordinate data generation unit, 65 ... Learning unit, 71 ... License plate image generation unit, 72 ... ..Bump processing section, 73 ... Noise processing section, 74 ... Shading processing , 75 ... distortion processing unit, 100 ... license plate DB, 200 ··· learning process DB

Claims (5)

A predetermined value obtained as a result of learning by applying a method related to machine learning to data related to an image for learning including an image including a plurality of license plates as subjects and an image created by processing the images. Algorithm acquisition means for acquiring the algorithm,
In addition to the predetermined license plate, a user information acquisition unit that acquires data regarding an image including a character, a color region, and a background as a subject,
Algorithm applying means for outputting a result obtained by applying the predetermined algorithm acquired by the algorithm acquiring means to the data acquired by the user information acquiring means,
An information processing apparatus including. The algorithm acquisition means is at least one of bump processing means for raising the character, noise processing means for adding noise to the character, shading processing means for shading the character, and distortion processing means for distorting the character. Acquiring the predetermined algorithm obtained by using the learning image including the image processed by the processing means,
The information processing apparatus according to claim 1. The plurality of license plate images, in whole or in part, is obtained based on the data of one coordinate of the four corners of the license plate,
The information processing apparatus according to claim 1. The algorithm acquisition means, at least the area related to the area name of the license plate, the area related to the classification number of the license plate, the area related to the hiragana of the license plate, the area related to the number of the license plate Divided into four regions, and a method related to machine learning is applied to each region to obtain the determined predetermined algorithm,
The information processing apparatus according to any one of claims 1 to 3. A program executed by the information processing device,
A predetermined value obtained as a result of learning by applying a method related to machine learning to data related to an image for learning including an image including a plurality of license plates as subjects and an image created by processing the images. An algorithm acquisition step to acquire the algorithm,
In addition to a predetermined license plate, a user information acquisition step of acquiring data regarding an image including a character, a color region, and a background as a subject,
An algorithm applying step of outputting a result obtained by applying the predetermined algorithm acquired in the algorithm acquiring step to the data acquired in the user information acquiring step,
A program to function as.

Images