KR20230099590A - Code recognition device, and character code recognition method using the same - Google Patents

Abstract

An embodiment of the present invention is a method of recognizing a character code, which is performed by a control unit included in a code recognition device, comprising: obtaining a code image including a character code; extracting feature maps of the code image using a machine learning model, and extracting the plurality of text labels and the number of text labels from the feature maps; Weight values corresponding to the feature maps are obtained from the machine learning model, and a combination of the feature maps and the weight values corresponding to the feature maps is used to determine the value of each character corresponding to the plurality of character labels of attention in the code image. Process of detecting code area; designating a position of each character in the character code using the code area of each character, the plurality of character labels, and the number of character labels; and outputting the character code included in the code image.

Description

Code recognition device and character code extraction method using the same {Code recognition device, and character code recognition method using the same}

The present invention relates to a code recognition device and a method for extracting a character code using the same, and in detail, a vehicle number can be extracted by recognizing a character code from a code image including character codes such as letters or numbers, such as a photograph of a license plate, It relates to a code recognition device and a character code extraction method using the same.

A code image means a digital image including a code having a standardized length and format. The code may include human-readable characters or non-human-readable symbols. Examples of characters include letters, numbers, and the like.

Hereinafter, a code composed of at least one letter, at least one number, and a combination of at least one letter and at least one number and having a standardized length and format is referred to as a character code.

Examples of character codes that can be seen in everyday life include resident registration numbers, phone numbers, barcode numbers, license plate numbers, and the like, and their digital images are called code images.

Recently, a technique for recognizing a character code from a code image using machine learning technology is rapidly developing. A general machine learning technique trains a machine learning model with a code image, classifies the code image into a class corresponding to a specific character code using the learned machine learning model, and recognizes the character code.

For example, in the problem of recognizing a code image consisting of one number, the machine learning model classifies the code image into a class corresponding to one of the numbers 0 to 9, and the number corresponding to the class is a number in the code image. Recognize. In this case, the machine learning model may be a 10 class classification model.

Expanding the above problem, in the problem of recognizing a code image composed of 4 numbers, the number of classes that can be expressed with 4 numbers is 10 * 10 * 10 * 10 = 10,000. That is, in order to recognize a code image composed of 4 numbers at once, a machine learning model capable of classifying 10,000 classes is required.

1 is a graph showing the classification accuracy of a general machine learning model according to the number of classes to be classified.

Referring to FIG. 1 , the classification performance of a general machine learning model is in inverse proportion to the number of classes to be classified. In other words, as the number of numbers in the code image increases, the number of classes increases, and the recognition performance of the machine learning model deteriorates.

The number of classes of ImageNet data used as a benchmark for the classification performance of the current deep learning model is 1,000, and the Top-1 Accuracy standard is 90.88% (CoAtNet-7). In light of this, the machine learning model that classifies 10,000 classes at the current technology level is still in the research stage, and there are problems that require a lot of time to raise performance to commercial level. Therefore, recognition of vehicle license plate numbers (7 digits: 10 million classes) or telephone numbers (11 digits: 100 billion classes) is a more difficult classification problem.

Because of the above problems, the OCR-based code recognition technology typically used in the prior art uses a method of recognizing each individual character by separating individual characters in a code image.

Figure 2 (a) shows a schematic flow chart for a conventional OCR-based code recognition technique. 2(b) illustrates a process of obtaining a source image by pre-processing a color image.

The conventional OCR-based code recognition technology is a method of reading a code image one character at a time from left to right or from top to bottom.

As shown in FIG. 2(a), the conventional OCR-based code recognition technique includes obtaining a preprocessed source image (SI) (S10), and rotating the source image to level the code in the source image. (S20), a process of extracting a character area (WA) including all of each word (eg, 'Hello', 'World') in the source image (S30), each individual character (eg, A process of extracting individual character areas (CA) of 'H', 'e', 'l', 'l', and 'o' (S40), each individual character (from each individual character area) using a deep learning algorithm For example, the process of recognizing 'H', 'e', 'l', 'l', 'o') (S50), each word (eg, 'Hello', 'World') by combining recognized individual letters A recognition process (S60) may be included.

Referring to FIG. 2(b), a pre-processed source image is binarization in which a color image is first converted to grayscale, and pixel values are analyzed to increase brightness and contrast, dividing pixel values into two ranges and classifying them as 0 and 1. It refers to a code image consisting only of black text codes on a white background after working (refer to the direction of the arrow).

Conventional OCR-based code recognition technology has the advantage of being able to use a 10-class machine learning model with high accuracy because it sequentially recognizes individual characters by separating characters from a code image into individual character areas one by one.

However, the conventional OCR-based code recognition technology has a disadvantage in that the recognition rate is high only when the distinction between individual characters and characters is clear. In general camera images, there are cases in which the distinction between individual characters is not clear, or images with motion blur caused by filming while moving, or images in which characters are not clear because the camera itself shakes, reflects light, and is out of focus. Since most of them, the conventional OCR-based code recognition technology has significantly poor recognition performance for general camera images.

In order to compensate for this, there are correction technologies to make the image clear by removing noise, aligning processing to horizontally align oblique strings, and edge detection processing to clarify the edge area. It causes a problem that slows down the recognition speed.

Registered Patent No. 10-2179552 (OCR-based evidence collection device and method)

In order to solve the above problems, an embodiment of the present invention is to provide a code recognition device, a code recognition method, and a license plate extraction system using the same, which recognizes a character code without separating individual character areas on a code image. can

Embodiments of the present invention do not require preprocessing correction technology to clearly distinguish between individual characters, and have a character code recognition rate regardless of the image quality of a code image, a code recognition device, a code recognition method, and a vehicle using the same A number extraction system can be provided.

Embodiments of the present invention can provide a code recognition device, a code recognition method, and a license plate extraction system using the same, which have a faster recognition speed than conventional OCR-based code recognition technology.

A method of extracting a character code, performed by a control unit included in a code recognition device according to an embodiment of the present invention, comprising: obtaining a code image including a character code; extracting feature maps of the code image using a machine learning model, and extracting the plurality of text labels and the number of text labels from the feature maps; Weight values corresponding to the feature maps are obtained from the machine learning model, and a combination of the feature maps and the weight values corresponding to the feature maps is used to determine the value of each character corresponding to the plurality of character labels of attention in the code image. Process of detecting code area; designating a position of each character in the character code using the code area of each character, the plurality of character labels, and the number of character labels; and outputting the character code included in the code image.

A code recognition device, a method for recognizing a code, and a vehicle number extraction system using the same according to an embodiment of the present invention recognize character codes without separating individual character areas on a code image, and occur when extracting individual character areas. has the effect of eliminating errors that

A code recognition device, a code recognition method, and a vehicle number extraction system using the same according to an embodiment of the present invention do not require a process of extracting individual character areas, so a preprocessing correction technology for clearly distinguishing between individual characters. There is no need for this, and it has the effect of clearly recognizing the character code even if the distinction is unclear due to the interference pattern between individual characters.

A code recognition device, a code recognition method, and a vehicle number extraction system using the same according to an embodiment of the present invention have an effect of having a faster recognition speed than conventional OCR-based code recognition technology.

A code recognition device, a code recognition method, and a vehicle number extraction system using the same according to an embodiment of the present invention, such as a video taken by a CCTV for a general method or a black box for a vehicle, have severe motion blur or low resolution by long-distance shooting. , it has the effect of having a high character code recognition rate even in poor image quality due to noise such as low illumination and backlighting.

1 is a graph showing the classification accuracy of a general machine learning model according to the number of classes to be classified.
Figure 2 (a) shows a schematic flow chart for a conventional OCR-based code recognition technique.
2(b) illustrates a process of obtaining a source image by pre-processing a color image.
3 shows a configuration diagram of a code recognition device according to an embodiment of the present invention.
4 is a flowchart of a code recognition method according to an embodiment of the present invention.
5 is an overall conceptual diagram of a code recognition device according to an embodiment of the present invention.
6 is a detailed conceptual diagram of a code recognition device according to an embodiment of the present invention.
7(a) is a picture of a car license plate as an example of a code image, and FIG. 7(b) shows an example of learning data.
8 is an example showing a result recognized according to a code recognition method according to an embodiment of the present invention.
9 is a conceptual diagram of a vehicle number extraction system using a code recognition device according to an embodiment of the present invention.
10 is a flowchart of a control method of a vehicle number extraction system using a code recognition device according to an embodiment of the present invention.

Specific structural or functional descriptions of the embodiments according to the concept of the present invention disclosed in this specification are only illustrated for the purpose of explaining the embodiments according to the concept of the present invention, and the embodiments according to the concept of the present invention It can be embodied in various forms and is not limited to the embodiments described herein.

Embodiments according to the concept of the present invention can apply various changes and have various forms, so the embodiments are illustrated in the drawings and described in detail in this specification. However, this is not intended to limit the embodiments according to the concept of the present invention to specific disclosure forms, and includes all changes, equivalents, or substitutes included in the spirit and technical scope of the present invention.

Technical terms used in this specification are used only to describe specific embodiments, and are not intended to limit the present invention. Singular expressions include plural expressions unless the context clearly dictates otherwise. In this specification, terms such as "comprise" or "having" are intended to indicate that there is a feature, number, step, operation, component, part, or combination thereof described in this specification, but one or more other features It should be understood that it does not preclude the possibility of the presence or addition of numbers, steps, operations, components, parts, or combinations thereof.

The suffixes "module" and "unit" for the components used in this specification are given or used interchangeably in consideration of only the ease of writing the specification, and do not have a meaning or role that distinguishes each other by themselves, and this It may mean a functional or structural combination of hardware for performing a method according to an embodiment of the invention or software capable of driving the hardware.

Hereinafter, a code recognition device according to an embodiment of the present invention, a code recognition method, and a license plate extraction system using the same will be described.

3 shows a configuration diagram of a code recognition device according to an embodiment of the present invention.

Referring to FIG. 3 , a code recognition apparatus 1000 according to an embodiment of the present invention may include a code information recognition unit 1420, a code area detection unit 1430, and a code location designation unit 1440.

When the code recognition device 1000 according to an example is a server, it may include a memory 1100, a communication unit 1300, and a control unit 1400, and the control unit 1400 calculates the received information and stores the result values in memory. It can be stored in (1100) or transmitted to another device through the communication unit (1300). In this case, the code information recognition unit 1420, the code area detection unit 1430, and the code location designation unit 1440 act as the control unit 1400, and may execute previously stored source codes or software.

When the code recognition device 1000 according to an example is a recording medium, it may be configured with a memory 110 that stores information or provides requested information. In this case, the code information recognition unit 1420, the code area detection unit 1430, and the code location designation unit 1440 may be source codes stored in the memory 1100 or software itself.

When the code recognition device 1000 according to an example is a user terminal such as a smart phone or a personal computer, it may be composed of a memory 1100, an output unit 1200, a communication unit 1300, and a control unit 1400, and the control unit 1400 may store the result of calculating the received information in the memory 1100, transmit it to another device through the communication unit 1300, or output it to the user through the output unit 1200. In this case, the code information recognition unit 1420, the code area detection unit 1430, and the code location designation unit 1440 act as the control unit 1400, and may execute previously stored source codes or software.

The memory 1100 may store source codes or programs for processing and control of the controller 1400, store information input or output to the code recognition device 1000, and read stored information.

The output unit 1200 may output an audio signal, a video signal, an image signal, and the like. The output unit 1200 may include a display unit that displays and outputs information processed by the code recognition device 1000 or a sound output unit that outputs audio data stored in the memory 1100 .

The communication unit 1300 may include at least one component that allows the code recognition device 1000 to communicate with another device (not shown) or a server (not shown). Another device (not shown) may be a computing device or a sensing device, but is not limited thereto.

The controller 1400 controls overall operations of the code recognition device 1000 in general. The control unit 1400 is electrically connected to the memory 1100, the output unit 1200, and the communication unit 1300, and can read information stored in the memory 1100 or store information in the memory 110, and output Information may be provided to a user through the unit 1200 or information may be received or transmitted through the communication unit 1300 .

4 is a flowchart of a code recognition method according to an embodiment of the present invention. 5 is an overall conceptual diagram of a code recognition device according to an embodiment of the present invention. 6 is a detailed conceptual diagram of a code recognition device according to an embodiment of the present invention. 7(a) is a picture of a car license plate as an example of a code image, and FIG. 7(b) shows an example of learning data. 8 is an example showing a result recognized according to a code recognition method according to an embodiment of the present invention.

As described above, the code recognition method according to an embodiment of the present invention to be described below may be performed by the controller 1400 of the code recognition device 1000 or stored in a recording medium such as the memory 1100. can For convenience of description, the control unit 1400 of the code recognition device 1000 will be described from the viewpoint of performing the code recognition method, but the code recognition method described below is coded through a programming language and stored as source code on a recording medium. Or it can mean as software.

Referring to FIG. 4, the code recognition method according to an embodiment of the present invention includes a process of recognizing code information in a code image (S120), a process of detecting a code area in a code image (S130), and specifying a code location. It may include a process (S140) of doing. Furthermore, the code recognition method according to an embodiment of the present invention may further include a step of acquiring a code image (S110) before the step of recognizing code information (S120), and also a step of designating a code location ( After S140), a process of outputting a code (S150) may be further included.

Hereinafter, steps S110 to S150 will be described in detail.

In step S110, the code recognition device 1000 may obtain a code image. The code recognition device 1000 may obtain a code image that is photographed through a camera (not shown), photographed through a CCTV, received through the communication unit 1300, or stored in the memory 1100.

The code image refers to a digital image including a character code. A character code is a combination of at least one letter (eg, uppercase letters A to Z, lowercase letters a to z, etc.), at least one number (eg, Arabic numerals 0 to 9), or at least one letter and at least one number. It is a combination of characters, and generally means a code with a standardized length and format.

As shown in FIG. 7 (a), an example of a code image may be a license plate photo. If the car license plate picture is taken with a CCTV camera for general method, it is of poor quality as shown on the left, and it may be difficult to visually check the character code corresponding to the car number. You can visually check the code.

In step S120, the code recognition device 1000 may recognize code information from the code image.

Referring to FIG. 5 , the code recognition device 1000 may extract code information of a character code from a code image using a code information recognition unit 1420 .

The code information of the character code may include character labels constituting the character code and the number of character labels. The character label refers to the type of characters or numbers present in the character code. For example, if the character code is “DE008”, the character labels may be D, E, 0, and 8. Code information of “DE008” includes character label D and number 1, character label E and number 1, character label 0 and number 2, character label 8 and number 1, {(D:1), (E:1) , (0:2), (8:1)}.

For this reason, the code recognition apparatus 1000 can obtain the same code information from code images having different character codes. For example, the first code image of the first character code “4423” and the second code image of the second character code “4243” have different character codes, but the code recognition device 1000 has the first code image. and obtains the same code information from the second code image. The code information of the first code image and the second code image includes character label 2 and number 1, character label 3 and number 1, character label 4 and number 2, {(2,1), (3,1), (4,2)}.

Referring to FIG. 6 , the code information recognizing unit 1420 may be included in a machine learning model M that receives a code image and outputs code information.

The machine learning model (M) may be composed of a neural network, and code information ( For example, text labels and the number of text labels) can be output.

In detail, the machine learning model M may include a first network M1 that extracts a plurality of feature maps representing features of the code image from the code image, and the first network M1 ) may be Convolutional Neural Networks (CNNs) that are mainly used when extracting features of a code image. In this case, the plurality of feature maps may be expressed as an intermediate recognition result value of the machine learning model (M), and the final recognition result value may mean a character label corresponding to code information and the number of character labels.

In addition, the machine learning model (M) may include a second network (M2) for extracting code information from a plurality of feature maps, for example, the second network (M2) is a flattened layer (flattened layer), FC layer (Fully-Connected layer), an activation function, and a softmax function may be sequentially connected.

As another example, the machine learning model M may be a model learned with training data composed of code images including character codes and code information labeled on each code image. As shown in FIG. 7 (b), for example, the learning data may include a code image (X) including a character code consisting of 4 digits and code information (Y) labeled on the code image. Therefore, the learned code information recognizing unit 1420 may output text labels and the number of corresponding text labels as input of code images.

Furthermore, the process of recognizing code information in a code image (S120) includes the process of extracting a character area of a character code in which individual characters are aggregated in a code image (S121), and character using the code information recognition unit 1420 in the character area A process of obtaining code information of the code (S122) may be included. That is, the code recognition device 1000 does not directly extract code information from the code image, but first extracts the text area from the code image, excludes the background part of the code image that is not the text area, and extracts the text area from the image of the extracted text area. By extracting the code information of the character code, the code recognition speed can be improved.

The character area of a character code is a part containing a character code having a standardized length and format within a code image. For example, in the license plate of FIG. 7(a), the character area of the license plate is '2SN 4414 It may mean a minimum area including '.

In step S130, the code recognition apparatus 1000 may detect a code region in the code image.

Referring to FIG. 5 , the code recognition apparatus 1000 may detect a code region of attention from a code image by a code region detection unit 1430 .

The code area detection unit 1430 may detect a code area of attention on the code image based on code information extracted from the code image. The code area means a location where each character of the character code is predicted to be located in the code image. In other words, the code area refers to the area that the code area detection unit 1430 focuses attention on in the code image to recognize each character corresponding to the character label of the code information.

Referring to FIG. 6 , the code region detection unit 1430 may be included in a machine learning model M that outputs a code region from a code image based on code information extracted from a code image.

The machine learning model (M) may be composed of a neural network, and the code domain through backward propagation that calculates and stores variables of the neural network in order from the output layer of the neural network outputting character labels to the input layer. can detect

In detail, the machine learning model (M) may include a third network (M3) in which an input layer is connected to a plurality of feature maps, and the third network (M3) is a character label in an output layer of the third network (M3). To output each individual character, it is learned by backward propagation that calculates and stores the variables of the third network M3 from the output layer of the third network M3 to the input layer of the third network M3.

The machine learning model M may extract weight values for each of a plurality of feature maps for each individual character, and extract an attention map of each individual character by combining a plurality of feature maps and corresponding weight values.

As an example of extracting the attention map of each individual character by combining a plurality of feature maps and corresponding weight values, the average value of one feature map among the plurality of feature maps is multiplied by the corresponding weight value to apply the weight value of the feature map. , and the attention map of each individual character can be obtained by summing the weighted values for all feature maps. However, this is only one example of obtaining the attention map of each individual character, and the present invention is not limited thereto.

In the above description, the second network M2 and the third network M3 are separated and configured as individual networks, but the present invention is not limited thereto, and the second network M2 and the third network M3 are integrated into one integrated network. can be configured with In this case, the integrated network may output code information, and weight values corresponding to a plurality of feature maps may be extracted by calculating variables of the integrated network using the output code information.

Meanwhile, the attention map of each individual character includes digitized values for each horizontal and vertical position of the code image, and the code region of each individual character can be detected through the position of the peak value in the horizontal and vertical directions. Visualizing the attention map of each individual character shows a bright area on the code image, which is the code area of each individual character (see FIG. 6).

6 shows an example in which the attention maps of individual letters D, E, 0, and 8 are visualized as images, respectively, and the horizontal and vertical peak values based on the lower left of each image are ) at point (0.35, 0.50), character 0 at point (0.52, 0.52) and point (0.77, 049), and character 8 at point (0.87, 0.50). That is, in the code image of FIG. 6, the letter D, the letter E, and the letter 8 each have one code area, and the character 0 has two code areas.

As described above, the code area detection unit 1430 determines the code area for each individual character of the character code (or for each character corresponding to the character label) based on the recognized character label and the extracted feature map. code area) can be detected.

For example, when the type of character label is a number, the code area detection unit 1430 sequentially uses each character (for example, from 0 to 9) in all cases that the corresponding character label may have, and codes for each character. area can be detected. As another example, the code area detection unit 1430 may sequentially use only the characters recognized as character labels to detect the code area for each character.

In step S140, the code recognition device 1000 may designate a code location.

Referring to FIG. 5 , the code recognition apparatus 1000 may designate the location of each individual character constituting the character code in the code image based on the code area detected by the code location designation unit 1440 .

The code location specifying unit 1440 may obtain the center position of the code area using the code area of each character, and designate each character corresponding to the code area of each character at the center position.

The code location specifying unit 1440 may count the number of bright areas in the code area of each character. For example, the code location specifying unit 1440 may define as a code region a region in which the horizontal and vertical numerical values of the attention map on the code image are greater than or equal to preset horizontal and vertical threshold values, and the number of code regions can be counted

When the number of character labels is plural, the code location designation unit 1440 may designate a character corresponding to the corresponding character label to the code area of the character counted as a plurality of characters.

In addition, the code location specifying unit 1440 can express the code area of each character on one code image, and can designate each character corresponding to a character label at a position corresponding to the code area of each character (FIG. 5 reference).

In step S150, the code recognition device 1000 may output a character code.

The code recognizing apparatus 1000 may output character codes after designating positions of all characters on the code image. The code recognition device 1000 may output a character code through the output unit 1200, and output the character code through an output unit (monitor) of another device (eg, a client PC) through the communication unit 1300. can

Referring to FIG. 8 , the code recognizing apparatus 1000 of the present invention can provide a result of accurately recognizing a character code even in a code image of poor quality. GT is the actual character code value in the image where the code is hardly visible due to poor image quality, and predicted is the value of the character code predicted (recognized, extracted) through the code recognition device 1000, and it can be seen that both exactly match. .

9 is a conceptual diagram of a vehicle number extraction system using a code recognition device according to an embodiment of the present invention. 10 is a flowchart of a control method of a vehicle number extraction system using a code recognition device according to an embodiment of the present invention.

The vehicle number extraction system using the code image recognition device according to an embodiment of the present invention analyzes existing CCTV images for various purposes such as crime prevention or speed prevention, even if a separate vehicle number recognition dedicated camera is not additionally installed Thus, it is possible to provide a system capable of extracting a license plate number.

Referring to FIG. 9, the vehicle number extraction system using the code image recognition device according to an embodiment of the present invention includes a code recognition device 1000, a client PC 2000, and a video management system 3000 (VMS). , arrears/arranged management server 4000 may be included.

The client PC 2000 may be connected to the code recognition device 1000 and provide the user with information on the found problem vehicle through a screen of an output unit (monitor). The video management system 3000 (VMS, Video Management System) is a server that receives and stores images captured by CCTVs 3100 previously installed in roads or city facilities through a network. The delinquent/arbitrage management server 4000 is a server that stores and manages license plates of vehicles for which taxes have not been paid or license plates of vehicles involved in crime.

The code recognition device 1000 may be a server including a Poor Image License Plate Recognition (PILR) system. The code recognition device 1000 may be a device that performs the code recognition method according to the embodiment of the present invention of FIG. 4 described above.

The code recognition device 1000 receives a CCTV image from the video management system 3000, obtains a frame-by-frame code image from the CCTV image, recognizes code information in the code image, detects a code area in the code image, The code position is specified using code information and code area, the entire text code is output based on the code position, vehicle number information of the problem vehicle that is in arrears or wanted is received from the arrears/arranged management server (4000), and the code When a matching vehicle number is found by comparing the character code output from the image with the vehicle number, a notification of the discovery of the problem vehicle may be provided to the client PC 2000 .

Referring to FIG. 10, the control method of the vehicle number extraction system using the code recognition device according to an embodiment of the present invention includes a process of acquiring a CCTV image (S200), a process of acquiring a code image (S210), and a code The process of recognizing code information in an image (S220), the process of detecting a code area in a code image (S230), the process of designating a code location (S240), the process of outputting a character code (S250), the vehicle number in question It may include a process of acquiring (S260) and a process of providing a notification about the problem vehicle (S270).

As described above, the code recognition device 1000 according to an embodiment of the present invention includes at least one of a memory 1100, an output unit 1200, a communication unit 1300, and a control unit 1400, and steps S200 to S270 The process may be performed by the control unit 1400 or may be a source code or software stored in the memory 1100.

In step S200, the code recognizing device 1000 may obtain a CCTV image captured by the CCTV 3100 and CCTV identification information capturing the CCTV image from the VMS 3000. The CCTV image may be a low-quality image captured by a general camera.

In step S210, the code recognition device 1000 may obtain a code image by dividing the CCTV image into frame units and performing a black-and-white binarization process on the color image in frame units. Alternatively, the code recognition apparatus 1000 may obtain a code image binarized in black and white by processing in VMS.

In steps S220 to S250, the code recognition apparatus 1000 may equally perform steps S120 to S150 of FIG. 4, and duplicate descriptions will be omitted.

In step S260, the code recognition device 1000 may receive and store the license plate number of the vehicle in question from the arrears/arrangement management server 4000. Alternatively, if the character code output in step S250 is the vehicle number, the code recognition device 1000 transmits the output vehicle number to the arrears/arrangement management server 4000 to request information on whether or not it is the vehicle number, and Information on whether the vehicle number transmitted from the arrangement management server 4000 is the vehicle number of the vehicle in question may be received.

In step S270, the code recognition device 1000 compares the vehicle number received from the arrears/arrangement management server 4000 in real time when the character code output in step S250 is a vehicle number and finds the same number as the client PC ( 2000) can be notified of the discovery of the problem vehicle and license plate number and the identification information of the CCTV that captured the problem vehicle.

Although the present invention has been described with reference to the embodiments shown in the drawings, this is only exemplary, and those skilled in the art will understand that various modifications and equivalent other embodiments are possible therefrom. Therefore, the true technical protection scope of the present invention should be determined by the technical spirit of the attached claims.

1000: code recognition device
1100: memory
1200: output unit
1300: Ministry of Communication
1400: control unit

Claims (1)

In the character code recognition method, which is performed by a control unit included in a code recognition device,
obtaining a code image including a character code;
extracting feature maps of the code image using a machine learning model, and extracting the plurality of text labels and the number of text labels from the feature maps;
Weight values corresponding to the feature maps are obtained from the machine learning model, and a combination of the feature maps and the weight values corresponding to the feature maps is used to determine the value of each character corresponding to the plurality of character labels of attention in the code image. Process of detecting code area;
designating a position of each character in the character code using the code area of each character, the plurality of character labels, and the number of character labels; and
Character code recognition method comprising the step of outputting the character code included in the code image.

Images