Parking management method and apparatus using vehicle number pattern recognition

Abstract

The present invention may provide a parking management method including the steps of: acquiring a first license plate image for a first vehicle license plate of a first target vehicle entering and exiting a parking lot from a video captured in real time by a camera fixedly installed in the parking lot; extracting first license plate pattern features on the basis of the first license plate image using an artificial intelligence model and registering the extracted first license plate pattern features in a database; acquiring a second license plate image for a second vehicle license plate of a second target vehicle entering and exiting the parking lot from the real-time captured video, extracting second license plate pattern features on the basis of the second license plate image using the artificial intelligence model, and comparing the second license plate pattern features with the first license plate pattern features registered in the database to check whether the second target vehicle and the first target vehicle are the same vehicle.

Description

Parking management method and device using vehicle number pattern recognition

[0001] The present invention relates to a parking management method and device, and more particularly, to a parking management method for recognizing a target vehicle without recognizing a vehicle number.

[0002] License plate recognition technology automatically recognizes the numerical (and letter-based) information on license plates through video. This technology is used in a variety of fields, including speeding and illegal parking enforcement, road toll collection, and CCTV surveillance.

[0003] Typical license plate recognition technology involves an electronic device recognizing the location of a license plate through an image containing the number information, recognizing the position of the numbers within the license plate, and recognizing each digit. Here, the electronic device recognizes each digit by selecting one of ten digits, from 0 to 9.

[0004] In this regard, prior art document No. 10-2023-0018884 discloses "A device and method for recognizing vehicle license plates based on Faster-RCNN and a Korean handwriting recognition model." The prior art document describes training by labeling each digit on a vehicle license plate as one of ten characters, from 0 to 9.

[0005] The above-mentioned general license plate recognition technology can cause the following problems at entrances and exits of parking facilities such as parking lot entrances, department store entrances, and apartment entrances, where license plate recognition technology is utilized:

[0006] If the vehicle's driveway isn't straight and the license plate can't be captured directly, or if the license plate is contaminated, the license plate number may be misrecognized or not recognized. Furthermore, if any of the characters (numbers, letters, or symbols) on the license plate are interpreted as similar characters, the license plate number may be recognized as a different license plate number and the vehicle may be classified as a different vehicle. Furthermore, by directly recognizing each number on the license plate, personal information may be leaked.

[0007] [Prior Art Literature]

[0008] [Patent Document]

[0009] (Patent Document 1) Patent Publication No. 10-2023-0018884

[0010] In order to solve the above-described problem, the present invention aims to provide a parking management method and device that extracts license plate pattern features, recognizes a target vehicle through the license plate pattern features, extracts body pattern features, and increases the recognition accuracy of the target vehicle through the body pattern features.

[0011] In order to solve the above-described problem, the present invention aims to provide a parking management method and device that recognizes a target vehicle without recognizing the license plate number by extracting license plate pattern features and determining similarity based on the license plate pattern features.

[0012] In order to achieve the above-described object, an embodiment of the present invention provides a parking management method for recognizing a target vehicle without recognizing a license plate number, the method comprising: obtaining a first license plate image for a first license plate of a first target vehicle entering or exiting a parking lot from an image captured in real time by a camera fixedly installed in the parking lot; extracting first license plate pattern features based on the first license plate image using an artificial intelligence model trained on a plurality of image groups, and registering the first license plate pattern features in a database; wherein the plurality of image groups have multiple license plate images for different license plates, and the multiple license plate images in one image group have the same license plate number but are different images; and obtaining a second license plate image for a second vehicle license plate of a second target vehicle entering or exiting the parking lot from the real-time captured image, extracting second license plate pattern features based on the second license plate image using the artificial intelligence model, and comparing the second license plate pattern features with the first license plate pattern features registered in the database to determine whether the second target vehicle and the first target vehicle are the same vehicle.

[0013] An embodiment of the present invention can provide a parking management method in which a comparison between the second license plate pattern feature and the first license plate pattern feature is determined based on similarity.

[0014] An embodiment of the present invention may provide a parking management method further comprising the steps of: generating a virtual third license plate image based on a vehicle number input by a user; extracting a third license plate pattern feature based on the third license plate image using the artificial intelligence model; and comparing the extracted third license plate pattern feature with the first license plate pattern feature registered in the database to search for a first target vehicle having the input vehicle number.

[0015] An embodiment of the present invention provides a parking management device that recognizes a target vehicle without recognizing a license plate number, wherein a control unit of the parking management device obtains a first license plate image for a first vehicle license plate of a first target vehicle entering or exiting a parking lot from an image captured in real time by a camera fixedly installed in a parking lot, extracts a first license plate pattern feature based on the first license plate image using an artificial intelligence model trained in a plurality of image groups, and registers the first license plate pattern feature in a database, wherein - the plurality of image groups have a plurality of license plate images for different license plates, and the plurality of license plate images in one image group have the same license plate number but are different images -; obtains a second license plate image for a second vehicle license plate of a second target vehicle entering or exiting the parking lot from the real-time captured image, extracts a second license plate pattern feature based on the second license plate image using the artificial intelligence model, and compares the second license plate pattern feature with the first license plate pattern feature registered in the database to determine whether the second target vehicle and the first target vehicle are the same vehicle.

[0016] An embodiment of the present invention can provide a computer-readable recording medium having recorded thereon a program for performing the above-described parking management method.

[0017] In order to achieve the above-described object, a parking management method according to an embodiment of the present invention is a parking management method for recognizing a target vehicle without recognizing a license plate, the method comprising: a step of obtaining a body image and a license plate image of the target vehicle in real time by a camera installed at an entrance; a step of extracting a body pattern feature and a license plate pattern feature based on the body image and the license plate image using an artificial intelligence model; a step of determining a body similarity between the extracted body pattern feature and a registered body pattern feature in a database; a step of determining a license plate similarity between the extracted license plate pattern feature and a registered license plate pattern feature in the database; and a step of determining whether the target vehicle can enter or exit based on the body similarity and the license plate similarity.

[0018] In addition, in an embodiment of the present invention, the step of determining whether the target vehicle can enter or exit may include a step of calculating a comprehensive similarity based on a body weighted similarity obtained by applying a body weight to the body similarity and a license plate weighted similarity obtained by applying a license plate weight to the license plate similarity; and a step of determining whether the target vehicle can enter or exit based on the comprehensive similarity.

[0019] Additionally, in an embodiment of the present invention, the step of determining whether the target vehicle can enter or exit may allow entry or exit if the comprehensive similarity is greater than a threshold value.

[0020] In addition, the parking management method according to an embodiment of the present invention further includes a step of obtaining a license plate recognition level indicating the degree of exposure of the license plate of the target vehicle in the license plate image; and the step of determining whether the target vehicle can enter or exit can be performed when the license plate recognition level is greater than a license plate threshold value.

[0021] In addition, in the step of determining whether the target vehicle can enter or exit the parking lot in the embodiment of the present invention, a greater license plate weight can be applied to the license plate similarity as the license plate recognition degree increases.

[0022] In addition, in an embodiment of the present invention, the body image includes a first body image to an n-th body image, the license plate image includes a first license plate image to an n-th license plate image, the body pattern feature includes a first body pattern feature to an n-th body pattern feature, the license plate pattern feature includes a first license plate pattern feature to an n-th license plate pattern feature, the body similarity includes a first body similarity to an n-th body similarity, the license plate similarity includes a first license plate similarity to an n-th license plate similarity, the first license plate image to the n-th license plate image have different license plate recognition degrees indicating the degree of exposure of the license plate of the target vehicle, and all steps of the parking management method can be repeatedly performed n times for the body image and the license plate image captured at the same point in time according to the movement of the target vehicle.

[0023] In addition, in an embodiment of the present invention, the artificial intelligence model is trained with a plurality of different license plate images having the same number, and the plurality of license plate images may be a set of images to which at least one of color, shade, resolution, sharpness, horizontal ratio, vertical ratio, tilt, license plate contamination level, and license plate shooting angle is applied differently.

[0024] In addition, in an embodiment of the present invention, the artificial intelligence model is trained with a plurality of different body images having the same body, and the plurality of body images may be a set of images to which at least one of color, shade, resolution, sharpness, horizontal ratio, vertical ratio, tilt, body contamination, and body shooting angle is applied differently.

[0025] Additionally, in embodiments of the present invention, the AI ​​model can be trained using multiple image groups. The multiple image groups may contain multiple license plate images for different vehicle numbers. Multiple license plate images within a single image group may have the same vehicle number but be different images.

[0026] For example, the plurality of image groups may include a first image group having a plurality of first license plate images for a first vehicle number (e.g., '1234'), a second image group having a plurality of second license plate images for a second vehicle number (e.g., '2345'), a third image group having a plurality of third license plate images for a third vehicle number (e.g., '3456'), and so on.

[0027] Although multiple first license plate images within the first image group have the same first vehicle number, the images may be different images that are not identical to each other. For example, the images may differ in at least one or two or more combinations of size, ratio, damage, sharpness, blurriness, resolution, tilt, color, shading, resolution, shooting angle, contamination (degree of foreign matter) on the license plate, degree of partial cropping, saturation, brightness, degree of light reflection on the license plate during shooting, degree of light bleeding, etc.

[0028] In an embodiment of the present invention, the database may be composed of at least one body pattern feature and at least one license plate pattern feature.

[0029] Meanwhile, a parking management device according to an embodiment of the present invention is a parking management device that recognizes a target vehicle without recognizing a license plate, wherein a control unit of the parking management device obtains a body image and a license plate image of the target vehicle in real time by a camera installed at an entrance, extracts a body pattern feature and a license plate pattern feature based on the body image and the license plate image using an artificial intelligence model, determines a body similarity between the extracted body pattern feature and a registered body pattern feature in a database, determines a license plate similarity between the extracted license plate pattern feature and a registered license plate pattern feature in the database, and determines whether the target vehicle can enter or exit based on the body similarity and the license plate similarity.

[0030] Meanwhile, a computer-readable recording medium according to an embodiment of the present invention may record a program for performing a parking management method.

[0031] In order to achieve the above-described object, a parking management method according to an embodiment of the present invention is a parking management method for recognizing a target vehicle without recognizing a license plate number, comprising: a step of obtaining a first license plate image of the target vehicle from an image captured in real time by a camera installed at an entrance; a step of extracting a first license plate pattern feature based on the first license plate image using an artificial intelligence model; a step of determining a similarity between the extracted first license plate pattern feature and a second license plate pattern feature registered in a database; and a step of determining whether the target vehicle can enter or exit based on the similarity; wherein the artificial intelligence model is trained with a plurality of different license plate images having the same number, and the plurality of license plate images may be a set of images to which at least one of a color, a shade, a resolution, a horizontal ratio, a vertical ratio, a slope, a license plate contamination level, and a license plate shooting angle is differently applied.

[0032] In addition, a parking management method according to an embodiment of the present invention may further include a step of obtaining a second license plate image of a vehicle to be registered; a step of extracting a second license plate pattern feature based on the second license plate image using the artificial intelligence model; and a step of registering the extracted second license plate pattern feature in the database.

[0033] Additionally, in an embodiment of the present invention, the vehicle to be registered may be a vehicle that is to enter through the entrance in real time, and the target vehicle may be a vehicle that the vehicle to be registered is to exit through the entrance.

[0034] Additionally, in an embodiment of the present invention, the vehicle to be registered may be a vehicle that repeatedly attempts to enter through the entrance, and the target vehicle may be a vehicle that the vehicle to be registered attempts to enter through the entrance.

[0035] In addition, the parking management method according to an embodiment of the present invention may further include a step of discarding the first license plate image obtained above.

[0036] In addition, the parking management method according to an embodiment of the present invention may further include a step of discarding the acquired second license plate image.

[0037] Additionally, in an embodiment of the present invention, the database may be composed of vector data of at least one second license plate pattern feature.

[0038] Meanwhile, a parking management method according to an embodiment of the present invention is a parking management method for recognizing a target vehicle without recognizing a license plate, the method comprising: a step of obtaining a license plate image of the target vehicle from an image captured in real time by a camera; a step of extracting a license plate pattern feature based on the license plate image of the target vehicle using an artificial intelligence model; a step of registering the extracted license plate pattern feature in a database; a step of generating an arbitrary virtual license plate image based on a vehicle number input by a user; a step of extracting a virtual license plate pattern feature based on the virtual license plate image using the artificial intelligence model; a step of determining a similarity between the extracted virtual license plate pattern feature and the license plate pattern feature registered in the database; and a step of calculating a fee to be charged to the target vehicle based on the similarity.

[0039] In addition, in an embodiment of the present invention, the step of settling the fee may include a step of providing a virtual license plate image of the target vehicle to an output unit; a step of providing information on the result of settling the fee of the target vehicle to the output unit; and a step of providing information on selecting at least one of a card, cash, and electronic payment method to the output unit.

[0040] Meanwhile, a parking management device according to an embodiment of the present invention is a parking management device that recognizes a target vehicle without recognizing a license plate number, wherein a control unit of the parking management device obtains a first license plate image of the target vehicle from an image captured in real time by a camera installed at an entrance, extracts a first license plate pattern feature based on the first license plate image using an artificial intelligence model, determines a similarity between the extracted first license plate pattern feature and a second license plate pattern feature registered in a database, and determines whether the target vehicle can enter or exit based on the similarity, wherein the artificial intelligence model is trained with a plurality of different license plate images having the same number, and the plurality of license plate images may be a set of images to which at least one of color, shade, resolution, clarity, horizontal ratio, vertical ratio, inclination, license plate contamination level, and license plate shooting angle is differently applied.

[0041] Meanwhile, a computer-readable recording medium according to an embodiment of the present invention may record a program for performing a parking management method.

[0042] The present invention recognizes a target vehicle without recognizing the license plate number, thereby enabling accurate recognition of the target vehicle even when the vehicle's access road is not straight or the license plate is contaminated, eliminating concerns that specific characters on the license plate are recognized as other similar characters, and protecting personal information by not directly recognizing each number on the license plate.

[0043] The present invention recognizes a target vehicle based on the license plate pattern features and body pattern features of the target vehicle, thereby increasing recognition accuracy compared to when the target vehicle is recognized using only the license plate.

[0044] FIG. 1 is a drawing showing a flowchart of a parking management method according to an embodiment of the present invention.

[0045] Figure 2a is a drawing showing a contaminated license plate.

[0046] Figure 2b is a drawing showing a license plate taken of a vehicle entering a non-straight driveway.

[0047] FIG. 3a and FIG. 3b are drawings showing a plurality of different license plate images having the same number in an embodiment of the present invention.

[0048] FIG. 4a and FIG. 4b are drawings showing a plurality of different body images having the same body in an embodiment of the present invention.

[0049] FIG. 5a is a drawing showing registration of body pattern features in an embodiment of the present invention.

[0050] FIG. 5b is a drawing showing registration of license plate pattern features in an embodiment of the present invention.

[0051] FIG. 6 is a flowchart showing a step of determining entry or exit of a target vehicle in an embodiment of the present invention.

[0052] FIG. 7a is a drawing showing an entire image including a first body image and a first license plate image in an embodiment of the present invention.

[0053] FIG. 7b is a drawing showing an entire image including a second body image and a second license plate image in an embodiment of the present invention.

[0054] FIG. 7c is a drawing showing an entire image including a third body image and a third license plate image in an embodiment of the present invention.

[0055] FIG. 7d is a drawing showing an entire image including a fourth body image and a fourth license plate image in an embodiment of the present invention.

[0056] FIG. 8 is a block diagram showing a parking management device according to an embodiment of the present invention.

[0057] Figure 9 is a drawing showing a flowchart of a parking management method according to an embodiment of the present invention.

[0058] Figure 10 is a drawing showing a flowchart of a parking management method according to an embodiment of the present invention.

[0059] Fig. 11 is a drawing showing registration of license plate pattern features in an embodiment of the present invention.

[0060] Figure 12 is a drawing showing a flowchart of a parking management method according to an embodiment of the present invention.

[0061] FIG. 13a is a diagram illustrating the extraction of virtual license plate pattern features based on the entire input vehicle number in an embodiment of the present invention.

[0062] FIG. 13b is a diagram illustrating the extraction of virtual license plate pattern features based on some input vehicle numbers in an embodiment of the present invention.

[0063] Those skilled in the art will be able to develop various devices that embody the principles of the invention and fall within the scope and spirit of the invention, even if not explicitly described or illustrated in this specification. Furthermore, all conditional terms and embodiments listed herein are expressly intended, in principle, to facilitate understanding of the invention's concept and should be understood as being in no way limiting to the specifically listed embodiments and conditions.

[0064] The above-described objects, features and advantages will become more apparent through the following detailed description of the invention in conjunction with the accompanying drawings, so that a person skilled in the art will be able to easily implement the technical idea of ​​the invention.

[0065] The embodiments described herein will be described with reference to cross-sectional and / or perspective views, which are ideal illustrations of the present invention. The dimensions of components depicted in these drawings may be exaggerated for the purpose of effectively explaining the technical content. The form of the illustrations may be altered due to manufacturing techniques and / or tolerances.

[0066] When describing various embodiments, components that perform the same function will be given the same names and reference numbers for convenience even if the embodiments are different. In addition, the expression "at least one of A, B, and C" means that it is composed of one, two, or three of A, B, and C. In addition, expressions such as "(name)", "first (name)", and "second (name)" are distinguished for convenience, and general features that are not mentioned as differences may be the same. Furthermore, configurations and operations already described in other embodiments will be omitted for convenience.

[0067] Below, the configuration of a parking management method and a parking management device (100) according to an embodiment of the present invention will be described.

[0068] First, the configuration of a parking management device (100) that performs a parking management method is described.

[0069] FIG. 8 is a block diagram showing a parking management device (100) according to an embodiment of the present invention.

[0070] Referring to FIG. 8, a parking management device (100) according to an embodiment of the present invention can perform a parking management method. The parking management device (100) can be configured as an electronic device. The parking management device (100) and the electronic device can include a control unit (110), a memory (120), an input unit (130), and an output unit (140). Not all of these components are essential components of the parking management device (100), and the parking management device (100) can be implemented with more components or with fewer components.

[0071] The control unit (110) can process and process signals and control each component. The control unit (110) can process and process data input from the input unit (130), data output from the output unit (140), and data stored in the memory (120). The control unit (110) can function as a processor. The control unit (110) can instruct or command the parking management device (100) to perform a parking management method.

[0072] The memory (120) can store programs for processing and controlling the control unit (110). In addition, the memory (120) can also temporarily store data input from the input unit (130), data output from the output unit (140), etc. One or more memories (120) may be provided.

[0073] The memory (120) may include at least one storage medium among flash memory, HDD, SDD, RAM, and ROM. The type of the memory (120) is not limited to those described above, and any means capable of storing data may be selected.

[0074] Meanwhile, a recording medium according to an embodiment of the present invention can store a program that performs a parking management method according to an embodiment of the present invention. The recording medium can include at least one of the memories (120) described above.

[0075] The input unit (130) may include at least one of a keyboard, a mouse, a microphone, a button, and a touch screen as a means for inputting content to be processed and processed in the control unit (110) or stored in the memory (120). However, the type of the input unit (130) is not limited to those described above, and any means that can transmit signals and / or data to an electronic device by a user may be selected.

[0076] The output unit (140) may include at least one of a display and a speaker as a means for outputting content processed and processed in the control unit (110) or stored in the memory (120). However, the type of the output unit (140) is not limited to the above-described ones, and any means that can transmit signals and / or data to a user via an electronic device may be selected.

[0077] The parking management device (100) according to an embodiment of the present invention is not limited to being installed in one space, but may be installed separately in different spaces.

[0078] Next, a parking management method according to an embodiment of the present invention will be described.

[0079] FIG. 1 is a flowchart illustrating a parking management method according to an embodiment of the present invention. FIG. 2a is a drawing illustrating a contaminated license plate. FIG. 2b is a drawing illustrating a license plate photographed from a vehicle entering a non-straight driveway. FIGS. 3a and 3b are drawings illustrating multiple different license plate images (200) having the same number in an embodiment of the present invention. FIGS. 4a and 4b are drawings illustrating multiple different body images (300) having the same body in an embodiment of the present invention. FIG. 5a is a drawing illustrating registration of a body pattern feature (350) in an embodiment of the present invention. FIG. 5b is a drawing illustrating registration of a license plate pattern feature (250) in an embodiment of the present invention. FIG. 6 is a drawing illustrating a flowchart of a step for determining entry or exit of a target vehicle (10) in an embodiment of the present invention.

[0080] Referring to FIG. 1, a parking management method according to an embodiment of the present invention may include the steps of: acquiring a body image (300) and a license plate image (200) of a target vehicle (10) in real time by a camera installed at an entrance; extracting a body pattern feature (350) and a license plate pattern feature (250) based on the body image (300) and the license plate image (200) using an artificial intelligence model (150); determining a body similarity between the extracted body pattern feature (350) and a registered body pattern feature (350') in a database; determining a license plate similarity between the extracted license plate pattern feature (250) and a registered license plate pattern feature (250') in a database; and determining whether the target vehicle (10) can enter or exit based on the body similarity and the license plate similarity.

[0081] First, a step of acquiring a body image (300) and a license plate image (200) of a target vehicle (10) in real time by a camera installed at the entrance can be performed by the control unit (110).

[0082] The target vehicle (10) may refer to a vehicle attempting to pass through an entrance. The target vehicle (10) may refer to a vehicle for which entry / exit is determined by a parking management device (100). The target vehicle (10) may refer to at least one of a two-wheeled vehicle, a three-wheeled vehicle, and a four-wheeled vehicle, but is not limited thereto, and may be a concept that includes all means of transportation that are operated with a license plate attached. The target vehicle (10) may be divided into a license plate and a body excluding the license plate.

[0083] An entrance / exit may be a location where a parking management device (100) can be installed to manage the entry and exit of a target vehicle (10). For example, the entrance / exit may refer to at least one entrance / exit of a parking lot, department store, supermarket, or apartment, but is not necessarily limited thereto. Furthermore, the entrance / exit may be a location where a barrier, toll payment machine, or the like controlled by the parking management device (100) is installed.

[0084] The camera can capture images of the body of the target vehicle (10) and the license plate of the target vehicle (10). The camera can capture images of the target vehicle (10) in real time. Specifically, the camera can capture images of the body of the target vehicle (10) in real time and capture images of the body (300) and the license plate (200). The camera can capture dynamic images and / or static images. The camera can capture images of the body and license plate of the target vehicle (10) moving to pass through an entrance.

[0085] Cameras are installed at the entrances and exits of the parking lot and record vehicles entering and exiting in real time.

[0086] At least one camera may be provided. The camera may be positioned to face the vehicle before entering the parking lot to capture the target vehicle (10) entering the parking lot. The camera may be positioned to face the vehicle before exiting the parking lot to capture the target vehicle (10) exiting the parking lot. The camera may capture the target vehicle (10) from various angles.

[0087] The body image (300) may refer to an image of the body of the target vehicle (10). The body image (300) may be an image of the body of the target vehicle (10) taken from various angles.

[0088] The license plate image (200) may refer to an image of the license plate of the target vehicle (10). The license plate image (200) may include at least one of numbers, letters, and symbols.

[0089] The video of the target vehicle (10) captured in real time by a fixed camera installed at the entrance may contain various angles of the target vehicle (10) moving according to the shape of the road before and after the entrance. The video of the target vehicle (10) captured in real time may include an image of the body of the target vehicle (10) and an image of the license plate captured at the same time, and the image of the body of the target vehicle (10) and the image of the license plate may be divided into multiple images at different points in the real-time video.

[0090] Next, a step of extracting body pattern features (350) and license plate pattern features (250) based on the body image (300) and license plate image (200) using an artificial intelligence model (150) can be performed by the control unit (110).

[0091] The artificial intelligence model (150) may be a model trained using a body image (300) and a license plate image (200) as input. The artificial intelligence model (150) may be a deep learning model, a machine learning model, or the like. The artificial intelligence model may be implemented as a model such as CNN, YOLO, or Faster RCNN, but is not limited thereto. A number may be a concept including at least one of letters, numbers, and symbols.

[0092] The artificial intelligence model (150) may be a model trained using the body image (300) and the license plate image (200) as a single image. Alternatively, the artificial intelligence model (150) may be a model trained using the body image (300) as a single image, or a model trained using the license plate image (200) as a single image.

[0093] The artificial intelligence model (150) may include a first artificial intelligence model (151) and a second artificial intelligence model (152). The first artificial intelligence model (151) may be trained with a body image (300), and the second artificial intelligence model (151) may be trained with a license plate image (200).

[0094] That is, the artificial intelligence model (150) may be a model learned using the body image (300) and the license plate image (200) as a single image, and may be divided into a first artificial intelligence model (151) and a second artificial intelligence model (152) and may be models learned using the body image (300) and the license plate image (200), respectively.

[0095] The artificial intelligence model (150) can be trained with body images (300) having various bodies. The artificial intelligence model (150) can be trained with multiple different body images (300) having the same body. That is, the artificial intelligence model (150) can be trained with various bodies, and can be trained with multiple body images (300) for each body.

[0096] A plurality of body images (300) may refer to a set of body images (300) to which body parameters (at least one of color, shade, resolution, sharpness, horizontal ratio, vertical ratio, tilt, body contamination, and body shooting angle) are applied differently. However, the body parameters are not limited to the types described above.

[0097] For example, the artificial intelligence model (150) can be trained with various body images (300) having different body shooting angles applied to the body of the 'Company A Model B', for the body of the 'Company A Model B'. The same applies to color, shade, resolution, sharpness, horizontal ratio, vertical ratio, inclination, and body contamination level.

[0098] The learning data for the body may be composed of multiple body images (300) and label values ​​for the body. For example, the label value may be the model name of the target vehicle. The learning data may include multiple groups of identical body images grouped by body images (300) of the same body, and the multiple groups of identical body images may be composed of different vehicles. The same body image group may include multiple body images (300) having the same vehicle body but composed of different images.

[0099] The artificial intelligence model (150) can extract a body pattern feature (350) based on the body image (300). The body pattern feature (350) may be data representing a unique feature of the body image (300). The body pattern feature (350) may be a vector (value) representing a unique feature of the body image (300). The artificial intelligence model can extract a vector, which is a unique feature, from the body image (300).

[0100] The artificial intelligence model (150) can be trained with license plate images (200) having various numbers. The artificial intelligence model (150) can be trained with multiple different license plate images (200) having the same number. That is, the artificial intelligence model (150) can be trained with various numbers, and can be trained with multiple license plate images (200) for each number.

[0101] The training data for the license plate of the artificial intelligence model (150) may be composed of multiple license plate images (200) and numbers on the license plate images. The training data may include multiple groups of identical license plate images, each grouped by license plate images with the same number, and the multiple groups of identical license plate images may have different numbers. The groups of identical license plate images may include multiple license plate images with the same number but composed of different images.

[0102] A plurality of license plate images (200) may refer to a set of license plate images (200) to which license plate parameters (at least one of color, shade, resolution, sharpness, horizontal ratio, vertical ratio, tilt, license plate contamination level, and license plate shooting angle) are applied differently. However, the license plate parameters are not limited to the types described above. In other words, a plurality of license plate images within the same license plate image group may be different images generated by differently setting at least one of the plurality of parameters.

[0103] For example, the artificial intelligence model (150) can be trained with license plate images (200) having different license plate shooting angles for the license plate '12허5051'. The same applies to shade, resolution, clarity, horizontal ratio, vertical ratio, inclination, license plate contamination, and license plate shooting angle (see FIG. 3a).

[0104] The artificial intelligence model (150) can extract license plate pattern features (250) based on the license plate image (200). The license plate pattern features (250) may be data representing unique features of the license plate image (200). The license plate pattern features (250) may be vectors (values) representing unique features of the license plate image (200). The artificial intelligence model (150) can extract vectors, which are unique features, from the license plate image (200).

[0105] Referring to Fig. 2a, when a camera captures a contaminated license plate, the color of a portion of the license plate may be distorted, and a portion of the number may appear unclear. The artificial intelligence model (150) is trained with license plate images (200) to which license plate parameters (color and / or license plate contamination) are applied differently, and can extract license plate pattern features (250) from license plate images (200) to which license plate parameters (color and / or license plate contamination) are applied differently.

[0106] Accordingly, the parking management device (100) and the parking management method can accurately recognize the target vehicle (10) even if the license plate is contaminated. Since the parking management device (100) recognizes the distorted license plate image (200) itself (with different parameters applied) and extracts the license plate pattern features (250), it can recognize the target vehicle (10) without recognizing the license plate number and determine whether it is the same vehicle.

[0107] In addition, when the camera captures the body of a contaminated target vehicle (10), the color of a part of the body may be distorted, and a part of the body may appear unclear. The artificial intelligence model (150) is trained with body images (300) to which body parameters (body contamination levels) are applied differently, and can extract body pattern features (350) from body images (300) to which body parameters (body contamination levels) are applied differently.

[0108] Accordingly, the parking management device (100) and the parking management method can accurately recognize the target vehicle (10) even if the body is contaminated. Since the parking management device (100) recognizes the distorted body image (300) itself (with different parameters applied) and extracts the body pattern features (350), it can perform an auxiliary role in recognizing the target vehicle (10) without recognizing the vehicle number, thereby determining whether it is the same vehicle.

[0109] Referring to FIGS. 2B, 3A, and 3B, when a camera fixedly installed at an entrance captures a license plate of a vehicle entering a non-straight driveway (or when the camera captures a license plate from a direction other than the front), the license plate of the vehicle may be distorted. The artificial intelligence model (150) is trained with license plate images (200) to which license plate parameters (at least one of the horizontal ratio, vertical ratio, inclination, and license plate shooting angle) are differently applied, and can extract license plate pattern features (250) from license plate images (200) to which license plate parameters (at least one of the horizontal ratio, vertical ratio, inclination, and license plate shooting angle) are differently applied.

[0110] Accordingly, the parking management device (100) and the parking management method can accurately recognize the target vehicle (10) without being limited by the angle, distance, etc. at which the license plate is photographed. Since the parking management device (100) recognizes the distorted license plate image (200) itself (with different license plate parameters applied) and extracts the license plate pattern features (250), it can recognize the target vehicle (10) without recognizing the license plate number and determine whether it is the same vehicle.

[0111] Referring to FIGS. 2B, 4A, and 4B, when a camera fixedly installed at an entrance captures the body of a vehicle entering a non-straight driveway (or when the camera captures the body from a direction other than the front), the body of the vehicle may be distorted. The artificial intelligence model (150) is trained with body images (300) to which body parameters (at least one of the horizontal ratio, vertical ratio, inclination, and body shooting angle) are differently applied, and can extract body pattern features (350) from body images (300) to which body parameters (at least one of the horizontal ratio, vertical ratio, inclination, and body shooting angle) are differently applied.

[0112] Accordingly, the parking management device (100) and the parking management method can accurately recognize the target vehicle (10) without limitations on the angle, distance, etc., at which the body is photographed. Since the parking management device (100) recognizes the distorted body image (300) itself (with different body parameters applied) and extracts the body pattern features (350), it can perform an auxiliary role in recognizing the target vehicle (10) without recognizing the vehicle number, thereby determining whether it is the same vehicle.

[0113] Next, a step of determining the body similarity between the extracted body pattern feature (350) and the registered body pattern feature (350') in the database can be performed by the control unit (110).

[0114] The body pattern feature (350) may be data and / or vectors representing unique features of the body image (300). The body pattern feature (350) may be extracted by an artificial intelligence model (150) (according to a command from the control unit (110)) based on the body image (300).

[0115] Referring to FIG. 5A, the registered body pattern feature (350') may be data and / or a vector representing a unique feature of the body image (300). The registered body pattern feature (350') may be registered in a database and / or stored in a memory (120). The registered body pattern feature (350') may be registered in advance. The registered body pattern feature (350') may be acquired in advance from the body image (300) of the target vehicle (10). The target vehicle (10) from which the registered body pattern feature (350') and the body pattern feature (350) are extracted may be the same, and the body image (300) from which the extraction is made may be different.

[0116] Both the body pattern feature (350) and the registered body pattern feature (350') may be data and / or vectors based on the same target vehicle (10). The body pattern feature (350) and the registered body pattern feature (350') may be data unrelated to the vehicle number and may be data from which the vehicle number cannot be inferred.

[0117] A database (DB) may refer to a collection of at least one body pattern feature (350) and at least one license plate pattern feature (250). The database may include registered body pattern features (350'). The database may be composed of vector data of registered body pattern features (350'). The control unit (110) may access the database to read or write registered body pattern features (350').

[0118] Similarity can refer to the degree of similarity between two vectors. The greater the similarity, the closer it is to determining that the target vehicle (10) and the registered vehicle are the same vehicle. The smaller the similarity, the closer it is to determining that the target vehicle (10) and the registered vehicle are not the same vehicle.

[0119] Similarity can be calculated using at least one of the following methods: Cosine Similarity, Jaccard Similarity, Pearson Similarity, Euclidean Distance, Manhattan Distance, and Minkowski Distance. The methods for determining similarity are not limited to those described above.

[0120] The control unit (110) can determine or calculate the body similarity between the body pattern feature (350) and the registered body pattern feature (350'). The control unit (110) can determine or calculate the body similarity between the vector according to the body pattern feature (350) and the vector according to the registered body pattern feature (350'). The body similarity can mean the degree of similarity between the body pattern feature (350) (vector) and the registered body pattern feature (350') (vector).

[0121] Next, a step of determining the license plate similarity between the extracted license plate pattern feature (250) and the registered license plate pattern feature (250') in the database can be performed by the control unit (110).

[0122] The license plate pattern feature (250) may be data and / or vectors representing unique features of the license plate image (200). The license plate pattern feature (250) may be extracted by an artificial intelligence model (150) (according to a command from the control unit (110)) based on the license plate image (200).

[0123] Referring to FIG. 5b, the registration plate pattern feature (250') may be data and / or a vector representing a unique feature of the license plate image (200). The registration plate pattern feature (250') may be registered in a database and / or stored in a memory (120). The registration plate pattern feature (250') may be registered in advance. The registration plate pattern feature (250') may be obtained in advance from the license plate image (200) of the target vehicle (10). The registration plate pattern feature (250') and the target vehicle (10) from which the license plate pattern feature (250) is extracted may be the same, and the license plate image (200) from which the license plate pattern feature (250) is extracted may be different.

[0124] Both the license plate pattern feature (250) and the registration license plate pattern feature (250') may be data and / or vectors based on the same target vehicle (10). The license plate pattern feature (250) and the registration license plate pattern feature (250') may be data unrelated to the vehicle number and may be data from which the vehicle number cannot be inferred.

[0125] The database may be composed of vector data of at least one body pattern feature (350) and vector data of at least one license plate pattern feature (250). The at least body pattern feature (350) and the license plate pattern feature (250) may become registered body pattern features (350') and registered license plate pattern features (250'), respectively, after registration.

[0126] Since the body pattern feature (350), registered body pattern feature (350'), license plate pattern feature (250) and registered license plate pattern feature (250') are data that are not related to the vehicle number and cannot infer the vehicle number, the personal information of the user of the target vehicle (10) and the registered vehicle can be protected.

[0127] The database may include registration plate pattern features (250'). The database may be composed of vector data of the registration plate pattern features (250'). The control unit (110) may access the database to read or write the registration plate pattern features (250').

[0128] The control unit (110) can determine or calculate the license plate similarity between the license plate pattern feature (250) and the registration license plate pattern feature (250'). The control unit (110) can determine or calculate the license plate similarity between a vector according to the license plate pattern feature (250) and a vector according to the registration license plate pattern feature (250'). The license plate similarity can mean the degree of similarity between the license plate pattern feature (250) (vector) and the registration license plate pattern feature (250') (vector).

[0129] Next, a step of determining whether to allow the target vehicle (10) to enter or exit based on the body similarity and license plate similarity can be performed by the control unit (110).

[0130] In this case, the body image (300) and the license plate image (200) that serve as the basis for the body similarity and license plate similarity may be images taken of the target vehicle (10) at the same time.

[0131] Referring to FIG. 6, the step of determining whether the target vehicle (10) can enter or exit may include a step of calculating a comprehensive similarity based on a body weighted similarity that applies a body weight to the body similarity and a license plate weighted similarity that applies a license plate weight to the license plate similarity; and a step of determining whether the target vehicle (10) can enter or exit based on the comprehensive similarity.

[0132] The control unit (110) can determine whether to enter or exit based on the body similarity and license plate similarity.

[0133] The control unit (110) can calculate all body similarities by comparing the body pattern features (350) of the target vehicle (10) with all registered body pattern features (350') of the database. The control unit (110) can calculate all license plate similarities by comparing the license plate pattern features (250) of the target vehicle (10) with all registered license plate pattern features (250') of the database.

[0134] The control unit (110) can calculate a body weighted similarity by applying a body weight to the body similarity. The control unit (110) can calculate a license plate weighted similarity by applying a license plate weight to the license plate similarity. The control unit (110) can calculate a comprehensive similarity based on the body weighted similarity and the license plate weighted similarity. The control unit (110) can compare the body pattern features (350) and license plate pattern features (250) of the target vehicle (10) with all registered body pattern features (350') and registered license plate pattern features (250') of the database, and calculate all comprehensive similarities.

[0135] For example, the comprehensive similarity, body similarity, and license plate similarity can form the following relationship. The relationship can be composed of 'comprehensive similarity = body weighted similarity + license plate weighted similarity', 'body weighted similarity = body similarity * body weight', and 'license plate weighted similarity = license plate similarity * license plate weight'. However, the relationship is not limited to the above.

[0136] The control unit (110) can determine whether to allow the target vehicle (10) to enter or exit based on the comprehensive similarity.

[0137] For example, if the comprehensive similarity is greater than the threshold value, the control unit (110) can recognize that the target vehicle (10) entering or exiting the parking lot and the registered vehicle are the same vehicle. Furthermore, if there are multiple registered vehicles with a comprehensive similarity greater than the threshold value, the registered vehicle with the greatest comprehensive similarity among the multiple comprehensive similarities can be recognized as the same vehicle as the target vehicle (10) entering or exiting the parking lot. If the control unit (110) recognizes that the target vehicle (10) entering or exiting the parking lot is identical to the registered vehicle, the control unit (110) can make a decision that entry or exit is possible. That is, the step of determining whether the target vehicle (10) can enter or exit the parking lot can allow entry or exit if the comprehensive similarity is greater than the threshold value.

[0138] If the similarity is less than the threshold value, the control unit (110) can recognize that the target vehicle (10) entering or exiting the parking lot and the registered vehicle are different vehicles. If the control unit (110) recognizes that the target vehicle (10) entering or exiting the parking lot is not identical to the registered vehicle, the control unit (110) can decide that entry or exit is not possible. In other words, the step of determining whether the target vehicle (10) can enter or exit the parking lot can disallow entry or exit if the comprehensive similarity is less than the threshold value.

[0139] The parking management device (100) and parking management method according to an embodiment of the present invention can accurately recognize the target vehicle (10) and protect personal information by determining the similarity and determining whether to enter or exit based on the license plate pattern feature (250) without recognizing the vehicle number.

[0140] In addition, the parking management device (100) and parking management method according to an embodiment of the present invention can significantly reduce the false recognition rate for the target vehicle (10) because they recognize the pattern characteristics of the license plate without recognizing the vehicle number even if the license plate is contaminated, the entire license plate is not photographed, or is photographed at an angle.

[0141] In addition, the parking management device (100) and parking management method according to the embodiment of the present invention can further reduce the false recognition rate for the target vehicle (10) because they recognize the body pattern characteristics of the target vehicle in addition to the license plate pattern characteristics of the target vehicle.

[0142] Next, the body recognition and license plate recognition that vary depending on the exposure level of the target vehicle (10) in an embodiment of the present invention will be described.

[0143] FIG. 7a is a diagram showing an entire image including a first body image and a first license plate image in an embodiment of the present invention. FIG. 7b is a diagram showing an entire image including a second body image and a second license plate image in an embodiment of the present invention. FIG. 7c is a diagram showing an entire image including a third body image and a third license plate image in an embodiment of the present invention. FIG. 7d is a diagram showing an entire image including a fourth body image and a fourth license plate image in an embodiment of the present invention.

[0144] A parking management method according to an embodiment of the present invention may include a step of obtaining a license plate recognition degree, which indicates the degree of exposure of the license plate of a target vehicle (10) from a license plate image (200). In addition, the parking management method according to an embodiment of the present invention may further include a step of obtaining a body recognition degree, which indicates the degree of exposure of the body of the target vehicle (10) from a body image (300).

[0145] The license plate recognition degree may indicate the degree of exposure of the license plate appearing in the license plate image (200). The body recognition degree may indicate the degree of exposure of the body appearing in the body image (300). Here, the degree of exposure of the license plate may be related to the angle at which the license plate was photographed. For example, if the license plate recognition degree is 100, it indicates a case where the license plate was photographed from the front (the angle between the normal to the surface forming the license plate and the shooting direction is 0 or 180 degrees), and if the license plate recognition degree is 0, it indicates a case where the license plate was photographed from the side (the angle between the normal to the surface forming the license plate and the shooting direction is perpendicular) or a case where the license plate was not photographed.

[0146] Referring to Fig. 7a, the entire image of the target vehicle (10) exposes the license plate by 0(%) and the body by 50(%). The body image (300) of the target vehicle (10) acquired from the entire image includes 50(%) of the shape of the body of the target vehicle (10), and thus the body recognition rate may be 50. The license plate image (200) of the target vehicle (10) acquired from the entire image includes 0(%) of the shape of the license plate of the target vehicle (10), and thus the license plate recognition rate may be 0.

[0147] Since the body recognition rate is 50 and the license plate recognition rate is 0, the control unit (110) can calculate the comprehensive similarity based only on the body similarity.

[0148] Referring to Fig. 7b, the entire image of the target vehicle (10) exposes the license plate at 0(%) and the body at 100(%). The body image (300) of the target vehicle (10) acquired from the entire image includes 100(%) of the shape of the target vehicle (10), and thus the body recognition rate may be 100. The license plate image (200) of the target vehicle (10) acquired from the entire image includes 0(%) of the shape of the license plate, and thus the license plate recognition rate may be 0.

[0149] Since the body recognition rate is 100 and the license plate recognition rate is 0, the control unit (110) can calculate the comprehensive similarity based only on the body similarity.

[0150] Referring to Fig. 7c, the entire image of the target vehicle (10) exposes the license plate by 50% (approximately 45 degrees) and the body by 100%. The body image (300) of the target vehicle (10) acquired from the entire image includes 100% of the shape of the target vehicle (10), and thus the body recognition rate may be 100. The license plate image (200) of the target vehicle (10) acquired from the entire image includes 50% of the shape of the license plate, and thus the license plate recognition rate may be 50.

[0151] Since the body recognition rate is 100 and the license plate recognition rate is 50, the control unit (110) can calculate the comprehensive similarity based on the body similarity and the license plate similarity. At this time, since the body recognition rate is greater than the license plate recognition rate, the body weight may be greater than the license plate weight.

[0152] Referring to FIG. 7d, the entire image of the target vehicle (10) exposes the license plate at 90% (approximately 5 to 10 degrees) and the body at 100%. The body image (300) of the target vehicle (10) acquired from the entire image includes 100% of the shape of the target vehicle (10), and thus the body recognition rate can be 100. The license plate image (200) of the target vehicle (10) acquired from the entire image includes 90% of the shape of the license plate, and thus the license plate recognition rate can be 90.

[0153] Since the body recognition rate is 100 and the license plate recognition rate is 90, the control unit (110) can calculate the comprehensive similarity based on the body similarity and the license plate similarity. At this time, since the body recognition rate is greater than the license plate recognition rate, the body weight may be greater than the license plate weight. However, the body recognition rate and the license plate recognition rate are not limited to matching the body exposure information (%) and the license plate exposure information (%), respectively, as in the example described above.

[0154] As described with reference to FIGS. 7A to 7D, the control unit (110) can obtain license plate recognition and / or body recognition. The license plate recognition and body recognition may be associated with license plate weights and body weights, respectively.

[0155] The relative sizes of the body weight and the license plate weight can be determined based on the relative sizes of the body recognition rate and the license plate recognition rate. If the body recognition rate is greater than the license plate recognition rate, the body weight may be greater than the license plate weight. If the body recognition rate is less than the license plate recognition rate, the body weight may be less than the license plate weight. If the body recognition rate and the license plate recognition rate are the same, the body weight and the license plate weight may be the same.

[0156] For example, the body weight may be determined as the ratio of the body recognition rate to the sum of the body recognition rate and the license plate recognition rate, and the license plate weight may be determined as the ratio of the license plate recognition rate to the sum of the body recognition rate and the license plate recognition rate. In this case, the sum of the body weight and the license plate weight may be set to 1.

[0157] License plate weights can be applied to license plate similarity. The control unit (110) can apply a greater license plate weight to license plate similarity as the license plate recognition rate increases. The greater the license plate recognition rate, the greater the license plate weight.

[0158] As the exposure of the license plate increases or the angle at which the license plate is photographed becomes closer to the front, the license plate weight increases, so the control unit (110) can further increase the weight of the license plate image (200) when calculating the comprehensive similarity.

[0159] Body weights can be applied to body similarity. The control unit (110) can apply a greater body weight to body similarity as the body recognition degree increases. The greater the body recognition degree, the greater the body weight can be.

[0160] As the exposure of the body increases, the body weight increases, so the control unit (110) can further increase the weight of the body image (300) when calculating the comprehensive similarity.

[0161] The parking management device (100) and parking management method according to the embodiment of the present invention calculate the comprehensive similarity by varying the license plate weight and the body weight according to the degree of exposure of the license plate and the body, so that the identity of the target vehicle (10) and the registered vehicle can be determined with higher accuracy.

[0162] In an embodiment of the present invention, the parking management method can determine whether the target vehicle (10) can enter or exit the parking lot by photographing the target vehicle (10) in real time. All steps of the parking management method can be repeatedly performed n times according to the movement of the target vehicle (10). Here, n can be a natural number greater than 1. That is, the parking management method can repeatedly perform each step according to the movement of the target vehicle (10) and repeatedly determine whether to enter or exit the parking lot according to the degree of exposure of the body and license plate of the target vehicle (10).

[0163] The body image (300) may include a first body image to an n-th body image. The license plate image (200) may include a first license plate image to an n-th license plate image. The body pattern feature (350) may include a first body pattern feature to an n-th body pattern feature. The license plate pattern feature (250) may include a first license plate pattern feature to an n-th license plate pattern feature. The body similarity may include a first body similarity to an n-th body similarity. The license plate similarity may include a first license plate similarity to an n-th license plate similarity. The comprehensive similarity may include a first comprehensive similarity to an n-th comprehensive similarity.

[0164] The 'first' expression can correspond to the case where each step of the parking management method is performed once, and the 'n' expression can correspond to the case where each step of the parking management method is performed n times.

[0165] The control unit (110) can repeatedly obtain the first license plate image and / or the first body image to the n-th license plate image and / or the n-th body image from the camera and repeatedly determine whether the target vehicle (10) can enter or exit the parking lot according to the movement of the target vehicle (10).

[0166] At this time, the first license plate recognition degree to the n-th license plate recognition degree corresponding to the first license plate image to the n-th license plate image may be different from each other, and the first body recognition degree to the n-th body recognition degree corresponding to the first body image to the n-th body image may be different from each other.

[0167] In an embodiment of the present invention, the step of determining whether the target vehicle (10) can enter or exit can be performed when the license plate recognition level is greater than the license plate threshold value (first condition). The license plate threshold value can be set in advance.

[0168] Taking FIGS. 7a and 7b as examples, since the license plate recognition is 0, when checking whether the target vehicle (10) is the same vehicle only with the body image (300) of the target vehicle, all target vehicles (10) that are the same brand and type as any one of the registered vehicles can be recognized as the same vehicle, so the step of determining whether the target vehicle (10) can enter or exit can be performed only when the license plate recognition is greater than a license plate threshold value set in advance.

[0169] For example, assuming that the target vehicle (10) moves in the order of FIGS. 7a to 7d, n is 4, and the license plate threshold is 50, when it reaches FIG. 7c, that is, when k = 3, the third license plate recognition degree becomes 50, so it is possible to determine whether the target vehicle (10) can enter or exit based on the third license plate image and the third body image.

[0170] In addition, the step of determining whether the target vehicle (10) can enter or exit can be performed when the body recognition level is greater than the body threshold value (second condition). In addition, the step of determining whether the target vehicle (10) can enter or exit can be performed when the first condition and / or the second condition are satisfied.

[0171] The parking management device (100) and parking management method according to an embodiment of the present invention can accurately recognize the target vehicle (10) by recognizing the target vehicle (10) without recognizing the vehicle number, and can protect the user's personal information.

[0172] The parking management device (100) and parking management method according to an embodiment of the present invention can increase recognition accuracy compared to a case where a vehicle is recognized only by a license plate by recognizing the target vehicle (10) based on the license plate pattern feature (250) and the body pattern feature (350).

[0173] The parking management device (100) and parking management method according to an embodiment of the present invention can increase the recognition accuracy of the target vehicle (10) by recognizing the target vehicle (10) by taking into account the license plate recognition and / or the body recognition.

[0174] The parking management device (100) and parking management method according to an embodiment of the present invention can increase the recognition accuracy of the target vehicle (10) by recognizing the target vehicle (10) by changing the ratio of the license plate weight and the body weight.

[0175] Next, a parking management method according to an embodiment of the present invention will be described.

[0176] Decision on entry or exit

[0177] FIG. 9 is a flowchart illustrating a parking management method according to an embodiment of the present invention. FIG. 2a is a diagram illustrating a contaminated license plate. FIG. 2b is a diagram illustrating a license plate photographed from a vehicle entering a non-straight driveway. FIG. 3a is a diagram illustrating multiple different license plate images (200) having the same number according to an embodiment of the present invention. FIG. 3b is a diagram illustrating multiple different license plate images (200) having the same number according to an embodiment of the present invention.

[0178] Referring to FIG. 9, a parking management method according to an embodiment of the present invention may include a step (S110) of obtaining a first license plate image (200) of a target vehicle (10) from an image captured in real time by a camera installed at an entrance; a step (S120) of extracting a first license plate pattern feature (250) based on the first license plate image (200) using an artificial intelligence model (150); a step (S130) of determining a similarity between the extracted first license plate pattern feature (250) and a second license plate pattern feature (250) registered in a database; and a step (S140) of determining whether the target vehicle (10) can enter or exit based on the similarity.

[0179] Here, the artificial intelligence model can be trained with multiple different license plate images (200) having the same number. The multiple license plate images (200) may be a set of images to which at least one of color, shade, resolution, horizontal ratio, vertical ratio, tilt, license plate contamination level, and license plate shooting angle is applied differently.

[0180] First, a step (S110) of obtaining a first license plate image (200) of a target vehicle (10) from a video captured in real time by a camera installed at an entrance can be performed by the control unit (110).

[0181] The target vehicle (10) may refer to a vehicle attempting to pass through an entrance. The target vehicle (10) may refer to a vehicle for which entry / exit is determined by the parking management device (100). The target vehicle (10) may refer to at least one of a two-wheeled vehicle, a three-wheeled vehicle, and a four-wheeled vehicle, but is not limited thereto. The target vehicle (10) may be a concept encompassing all means of transportation that are operated with a license plate attached.

[0182] An entrance / exit may be a location where a parking management device (100) can be installed to manage the entry and exit of a target vehicle (10). For example, the entrance / exit may refer to at least one entrance / exit of a parking lot, department store, supermarket, or apartment, but is not necessarily limited thereto. Furthermore, the entrance / exit may be a location where a barrier, toll payment machine, or the like controlled by the parking management device (100) is installed.

[0183] The camera can capture a target vehicle (10) and / or the license plate of the target vehicle (10). The camera can capture a video (or license plate image (200)) of the target vehicle (10) in real time. The camera can capture dynamic and / or static images. The camera can capture the license plate of the target vehicle (10) moving to pass through the entrance.

[0184] At least one camera may be provided. The camera may be positioned to face a vehicle before entering the parking lot to capture the target vehicle (10) entering the parking lot. The camera may be positioned to face a vehicle exiting the parking lot to capture the target vehicle (10) exiting the parking lot. The camera may capture the target vehicle (10) from various angles.

[0185] The license plate image (200) may refer to an image of the license plate of the target vehicle (10). The license plate image (200) may include at least one of numbers, letters, and symbols. The first license plate image (200) may be captured by a camera.

[0186] Next, a step (S120) of extracting a first license plate pattern feature (250) based on a first license plate image (200) using an artificial intelligence model (150) can be performed by the control unit (110).

[0187] The artificial intelligence model (150) may be a model trained using a license plate image (200) as input. The artificial intelligence model (150) may be a deep learning model, a machine learning model, or the like. The artificial intelligence model (150) may be implemented as a model such as CNN, YOLO, or Faster RCNN, but is not limited thereto. A number may be a concept including at least one of letters, numbers, and symbols.

[0188] The artificial intelligence model (150) can be trained with license plate images (200) having various numbers. The artificial intelligence model (150) can be trained with multiple different license plate images (200) having the same number. In other words, the artificial intelligence model (150) can be trained with multiple license plate images (200) having various numbers.

[0189] The training data of the artificial intelligence model (150) may be composed of multiple license plate images (200) and numbers on the license plate images. The training data may include multiple groups of identical number images, each grouped with license plate images (200) having the same number, and the multiple groups of identical number images may have different numbers. The groups of identical number images may include multiple license plate images (200) having the same number but composed of different images.

[0190] A plurality of license plate images (200) may refer to a set of license plate images (200) to which parameters (at least one of color, shade, resolution, sharpness, horizontal ratio, vertical ratio, tilt, license plate contamination level, and license plate shooting angle) are applied differently. In other words, a plurality of license plate images within the same license plate image group may be different images generated by differently setting at least one of the plurality of parameters. However, the parameters are not limited to the types described above.

[0191] For example, the artificial intelligence model (150) can be trained with license plate images (200) having different license plate shooting angles for license plates with the number '12허5051'. The same applies to shade, resolution, clarity, horizontal ratio, vertical ratio, inclination, license plate contamination, and license plate shooting angle (see FIG. 3a).

[0192] The artificial intelligence model (150) can extract a license plate pattern feature (250) (or a first license plate pattern feature (250)) based on the license plate image (200) (or a first license plate image (200)). The license plate pattern feature (250) can be data representing a unique feature of the license plate image (200). The license plate pattern feature (250) can be a vector (value) representing a unique feature of the license plate image (200). The artificial intelligence model (150) can extract a vector, which is a unique feature, from the license plate image (200).

[0193] Referring to Fig. 2a, when a camera captures a contaminated license plate, the color of a portion of the license plate may be distorted, and a portion of the number may appear unclear. The artificial intelligence model (150) is trained with license plate images (200) to which parameters (color and / or license plate contamination) are applied differently, and can extract license plate pattern features (250) from license plate images (200) to which parameters (color and / or license plate contamination) are applied differently.

[0194] Accordingly, the parking management device (100) and the parking management method can accurately recognize the target vehicle (10) even if the license plate is contaminated. Since the parking management device (100) recognizes the distorted license plate image (200) itself (with different parameters applied) and extracts the license plate image (200) pattern, it can recognize the target vehicle (10) without recognizing the license plate number and determine whether it is the same vehicle.

[0195] Referring to FIGS. 2B, 3A, and 3B, when the camera captures the license plate of a vehicle entering a non-straight driveway (or when the camera captures the license plate from a direction other than the front), the license plate may be distorted. The artificial intelligence model (150) is trained with license plate images (200) to which parameters (at least one of the horizontal ratio, vertical ratio, inclination, and license plate shooting angle) are applied differently, and can extract license plate pattern features (250) from license plate images (200) to which parameters (at least one of the horizontal ratio, vertical ratio, inclination, and license plate shooting angle) are applied differently.

[0196] Accordingly, the parking management device (100) and the parking management method can accurately recognize the target vehicle (10) without limitations on the angle, distance, etc. at which the license plate is photographed. Since the parking management device (100) recognizes the distorted license plate image (200) itself (with different parameters applied) and extracts the license plate image (200) pattern, it can recognize the target vehicle (10) without recognizing the vehicle number and determine whether it is the same vehicle.

[0197] Next, a step (S130) of determining the similarity between the extracted first license plate pattern feature (250) and the second license plate pattern feature (250) registered in the database can be performed by the control unit (110).

[0198] The first license plate pattern feature (250) may be data and / or vectors representing unique features of the license plate image (200). The first license plate pattern feature (250) may be extracted by an artificial intelligence model (150) (according to a command from the control unit (110)) based on the first license plate image (200).

[0199] The second license plate pattern feature (250) may be data and / or vectors representing unique features of the license plate image (200). The second license plate pattern feature (250) may be registered in a database and / or stored in memory. The second license plate pattern feature (250) may be extracted by an artificial intelligence model (150) based on the second license plate image (200). The extraction of the second license plate pattern feature (250) will be described later.

[0200] Both the first license plate pattern feature (250) and the second license plate pattern feature (250) may be data (or vectors) based on the same target vehicle (10). The license plate pattern feature (250), the first license plate pattern feature (250), and the second license plate pattern feature (250) may be data unrelated to the vehicle number, and may be data from which the vehicle number cannot be inferred.

[0201] A database (DB) may refer to a collection of at least one license plate pattern feature (250). The database may include a second license plate pattern feature (250). The database may be composed of vector data of the second license plate pattern feature (250). The control unit (110) may access the database to read or write the license plate pattern feature (250).

[0202] Similarity can refer to the degree of similarity between two vectors. The greater the similarity, the closer it is to determining that the target vehicle (10) and the registered vehicle are the same vehicle. The smaller the similarity, the closer it is to determining that the target vehicle (10) and the registered vehicle are not the same vehicle.

[0203] Similarity can be calculated using at least one of the following methods: Cosine Similarity, Jaccard Similarity, Pearson Similarity, Euclidean Distance, Manhattan Distance, and Minkowski Distance. The methods for determining similarity are not limited to those described above.

[0204] The control unit (110) can determine or calculate the similarity between the first license plate pattern feature (250) and the second license plate pattern feature (250). The control unit (110) can determine or calculate the similarity between a vector according to the first license plate pattern feature (250) and a vector according to the second license plate pattern feature (250).

[0205] Next, a step (S140) of determining whether to allow the target vehicle (10) to enter or exit based on the similarity may be performed by the control unit (110).

[0206] The control unit (110) can determine whether to allow entry or exit based on the similarity. The control unit (110) can compare the first license plate pattern feature (250) of the target vehicle (10) with all license plate pattern features (250) (or second license plate pattern features (250)) of the database to determine or calculate all similarities.

[0207] For example, if the similarity is greater than the threshold, the control unit (110) can recognize the identity of the first license plate pattern feature (250) and the second license plate pattern feature (250), and can recognize that the target vehicle (10) entering or exiting the parking lot and the registered vehicle are the same vehicle. Furthermore, if there are multiple second license plate pattern features (250) whose similarity with the first license plate pattern feature (250) is greater than the threshold, the registered vehicle having the greatest similarity among the multiple second license plate pattern features (250) can be recognized as the same vehicle as the target vehicle (10) entering or exiting the parking lot. If the control unit (110) recognizes that the target vehicle (10) entering or exiting the parking lot is identical to the registered vehicle, the control unit (110) can make a decision that entry or exit is possible.

[0208] If the similarity is less than the threshold, the control unit (110) can acknowledge the non-identity between the first license plate pattern feature (250) and the second license plate pattern feature (250), and can acknowledge that the target vehicle (10) entering or exiting the parking lot is a different vehicle from the registered vehicle. If the control unit (110) acknowledges that the target vehicle (10) entering or exiting the parking lot is not identical to the registered vehicle, the control unit (110) can make a decision that entry or exit is not possible.

[0209] The parking management device (100) and the parking management method can determine whether the target vehicle (10) can enter or exit the parking lot by comparing the second license plate pattern feature (250) registered in advance with the first license plate pattern feature (250) of the target vehicle (10) entering or exiting the parking lot (S110, S120, S130, S140).

[0210] The parking management device (100) and parking management method according to an embodiment of the present invention can accurately recognize the target vehicle (10) and protect personal information by determining the similarity and determining whether to enter or exit based on the license plate pattern feature (250) without recognizing the vehicle number.

[0211] In addition, the parking management device (100) and parking management method according to the embodiment of the present invention can significantly reduce the false recognition rate for the target vehicle (10) because it recognizes the pattern characteristics of the license plate without recognizing the vehicle number even if the license plate is contaminated, the entire license plate is not photographed, or is photographed at an angle.

[0212] Vehicle registration

[0213] Fig. 10 is a flowchart illustrating a parking management method according to an embodiment of the present invention. Fig. 11 is a diagram illustrating registration of license plate pattern features (250) according to an embodiment of the present invention.

[0214] Referring to FIGS. 10 and 11, a parking management method according to an embodiment of the present invention may further include a step (S101) of obtaining a second license plate image (200) of a vehicle to be registered; a step (S102) of extracting a second license plate pattern feature (250) based on the second license plate image (200) using an artificial intelligence model (150); and a step (S103) of registering the extracted second license plate pattern feature (250) in a database.

[0215] First, a step (S101) of obtaining a second license plate image (200) of a vehicle to be registered can be performed by the control unit (110).

[0216] The control unit (110) can compare the license plate pattern characteristics (250) of the target vehicle (10) with the license plate pattern characteristics (250) of the registered vehicle. The registered vehicle may refer to a vehicle registered in the database, and the vehicle to be registered may refer to a vehicle to be registered in the database.

[0217] The control unit (110) can acquire a second license plate image (200) of a vehicle to be registered. The second license plate image (200) may be an image of the vehicle to be registered captured in real time by a camera installed at the entrance. Alternatively, the second license plate image (200) may be an image of the vehicle to be registered captured in advance in non-real time by another capturing device. For example, it may be an image captured and transmitted by a camera such as a smartphone.

[0218] Next, a step (S102) of extracting a second license plate pattern feature (250) based on a second license plate image (200) using an artificial intelligence model (150) can be performed.

[0219] The artificial intelligence model (150) can extract license plate pattern features (250) (or second license plate pattern features (250)) based on the license plate image (200) (or second license plate image (200)). The second license plate pattern features (250) may be data and / or vectors representing unique features of the second license plate image (200). Since the artificial intelligence model (150) is as described above, a detailed description thereof will be omitted.

[0220] Next, a step (S103) of registering the extracted second license plate pattern feature (250) in a database can be performed by the control unit (110).

[0221] The control unit (110) can register the second license plate pattern feature (250) in a database. The database does not include a license plate image (200) and a vehicle number, but may include the license plate pattern feature (250).

[0222] Since the license plate pattern feature (250) is data that is not related to the vehicle number and cannot infer the vehicle number, the personal information of the vehicle user can be protected by registering the license plate pattern feature (250) in the database.

[0223] Meanwhile, the step (S102) of extracting second license plate pattern features (250) based on the second license plate image (200) using the artificial intelligence model (150) is a step of generating a plurality of similar second license plate images that are identical to the number of the second license plate but can be recognized as different images based on the second license plate image (200) in order to increase the recognition rate of the second license plate of the artificial intelligence model (150), and extracting a plurality of similar second license plate pattern features from each of the plurality of similar second license plate images using the artificial intelligence model (150).

[0224] A plurality of similar second license plate images can be generated by applying different parameters (at least one of color, shading, resolution, sharpness, aspect ratio, height ratio, tilt, license plate contamination, and license plate shooting angle) to the second license plate image.

[0225] In this case, the step (S103) of registering the extracted second license plate pattern feature (250) described above in the database may store not only the second license plate pattern feature (250) but also a plurality of similar second license plate pattern features in the database. In the database, the second license plate pattern feature (250) and a plurality of similar second license plate pattern features may be stored as one vehicle to be registered.

[0226] A parking management device (100) and a parking management method register a vehicle to be registered (S101, S102, S103), and compare the registered second license plate pattern feature (250) (or a plurality of similar second license plate pattern features) with the first license plate pattern feature (250) of a target vehicle (10) entering or exiting the parking lot to determine whether the target vehicle (10) can enter or exit the parking lot.

[0227] The parking management method according to an embodiment of the present invention can perform each step differently depending on whether the vehicle is irregular or regular. Irregular vehicles may refer to unpredictable, unspecified vehicles, such as taxis, while regular vehicles may refer to specific, predictable vehicles, such as resident vehicles or delivery vehicles.

[0228] In an embodiment of the present invention, the vehicle to be registered may be a vehicle that wishes to enter irregularly through the entrance, and the target vehicle (10) may be a vehicle that the vehicle to be registered wishes to exit through the entrance.

[0229] An irregular vehicle may refer to a vehicle that is not normally registered in the database but must pass through an entrance or exit. A parking management method may register vehicles (vehicles scheduled to be registered) that wish to enter through an irregular entrance or exit in real time in a database, and determine whether or not to allow such vehicles to exit by using the registered vehicles as target vehicles (10) that wish to exit through the entrance or exit.

[0230] In an embodiment of the present invention, the vehicle to be registered may be a vehicle that regularly enters through an entrance, and the target vehicle (10) may be a vehicle that the vehicle to be registered enters through an entrance.

[0231] A regular vehicle can refer to a vehicle registered in the database and required to pass through an entrance or exit. A parking management method can register vehicles that regularly attempt to enter or exit through an entrance or exit (vehicles to be registered) in the database in real time, and determine whether or not to allow entry by using the registered vehicles as target vehicles attempting to re-enter through the entrance or exit.

[0232] Accordingly, the parking management device (100) and parking management method according to the embodiment of the present invention can make a decision that an irregular vehicle that has entered can leave the parking lot when it wants to leave, and can make a decision that a regular vehicle can enter the parking lot when it wants to re-enter.

[0233] Discard license plate image (200)

[0234] The parking management method according to an embodiment of the present invention may further include a step of discarding the acquired first license plate image (200).

[0235] The first license plate image (200) can be obtained from a target vehicle (10) that is to enter or exit the parking lot. The control unit (110) can extract the first license plate pattern feature (250) based on the first license plate image (200) and discard the first license plate image (200). Since the first license plate image (200) may be stored (or temporarily stored) in the memory (120) when the first license plate pattern feature (250) is extracted, the control unit (110) can discard the first license plate image (200) to protect the user's personal information.

[0236] The parking management method according to an embodiment of the present invention may further include a step of discarding the acquired second license plate image (200).

[0237] The second license plate image (200) can be obtained from a vehicle to be registered. The control unit (110) extracts a second license plate pattern feature (250) based on the second license plate image (200), registers the second license plate pattern feature (250) in a database, and discards the second license plate image (200). Since the second license plate image (200) may be stored (or temporarily stored) in the memory (120) when the second license plate pattern feature (250) is extracted, the control unit (110) discards the second license plate image (200), thereby discarding the second license plate image (200) including the vehicle number, thereby protecting the user's personal information.

[0238] Fee settlement

[0239] Figure 12 is a flowchart illustrating a parking management method according to an embodiment of the present invention. Figure 13a is a diagram illustrating the extraction of virtual license plate pattern features (350a) based on the entire input vehicle number in an embodiment of the present invention. Figure 13b is a diagram illustrating the extraction of virtual license plate pattern features (350a) based on a portion of the input vehicle number in an embodiment of the present invention.

[0240] Referring to FIGS. 12, 13a and 13b, a parking management method according to an embodiment of the present invention includes a step (S210) of obtaining a license plate image (200) of a target vehicle (10) from an image captured in real time by a camera; a step (S220) of extracting a license plate pattern feature (250) based on the license plate image (200) of the target vehicle (10) using an artificial intelligence model (150); a step (S230) of registering the extracted license plate pattern feature (250) in a database; a step (S240) of generating an arbitrary virtual license plate image (300a) (200) based on a vehicle number input by a user; a step (S250) of extracting a virtual license plate pattern feature (350a) based on the virtual license plate image (300a) (200) using an artificial intelligence model (150); It may include a step (S260) of determining the similarity between the extracted virtual license plate pattern feature (350a) and the license plate pattern feature (250) registered in the database; and a step (S270) of calculating the fee to be charged to the target vehicle (10) based on the similarity.

[0241] First, a step (S210) of obtaining a license plate image (200) of a target vehicle (10) from a video captured in real time by a camera can be performed by the control unit (110).

[0242] The control unit (110) can obtain a license plate image (200) of the target vehicle (10). The license plate image (200) may be an image captured in real time by a camera of the target vehicle (10) to be entered. The step of the control unit (110) obtaining the license plate image (200) of the target vehicle (10) is performed in the same manner as described above, so a detailed description thereof will be omitted.

[0243] Next, a step (S220) of extracting a license plate pattern feature (250) based on a license plate image (200) of a target vehicle (10) using an artificial intelligence model (150) can be performed by the control unit (110).

[0244] The control unit (110) can extract license plate pattern features (250) based on the license plate image (200). The license plate pattern features (250) may be data and / or vectors representing unique features of the license plate image (200). The step of the control unit (110) extracting the license plate pattern features (250) is performed in the same manner as described above, so a detailed description thereof will be omitted.

[0245] Next, a step (S230) of registering the extracted license plate pattern features (250) in a database can be performed by the control unit (110).

[0246] The control unit (110) can register license plate pattern features (250) in a database. The database does not include a license plate image (200) and a vehicle number, but may include license plate pattern features (250). The step of registering the extracted license plate pattern features (250) in the database by the control unit (110) is performed in the same manner as described above, so a detailed description thereof will be omitted.

[0247] As another example, for user convenience, when storing the license plate pattern feature (250) in the database, the control unit may store at least one of the license plate image (200) and the entire target vehicle image by matching it to the license plate pattern feature (250). Here, the license plate image (200) and the entire target vehicle image may be images of the target vehicle captured by a camera at the entrance.

[0248] A step (S240) of generating an arbitrary virtual license plate image (300a) based on a vehicle number input by a user can be performed by the control unit (110).

[0249] A user can input a vehicle number through the input unit (130). The control unit (110) can obtain the vehicle number entered by the user. The control unit (110) can generate an arbitrary virtual license plate image (300a) based on the vehicle number entered by the user. The user can input all or part of the license plate information.

[0250] Taking Fig. 13a as an example, a license plate (or vehicle number) can be configured to include a vehicle type symbol ('06'), a use symbol ('you'), and a serial number ('4347'). A user can input all of the vehicle type symbol, use symbol, and serial number of the license plate (see Fig. 13a), or the user can input only the serial number of the license plate (see Fig. 13b).

[0251] The virtual license plate image (300a) may be a concept that contrasts with an image obtained by actually photographing the target vehicle (10). The virtual license plate image (300a) may be an image of a license plate that does not actually exist. The virtual license plate image (300a) may be a temporarily generated image. The virtual license plate image (300a) may be generated in any form. The virtual license plate image (300a) may be an image including a vehicle number obtained from a user.

[0252] Continuing with reference to FIG. 13a, the control unit (110) can receive the entire license plate from the user and obtain it as a vehicle number. The control unit (110) can obtain the vehicle number '06너4347'. Based on the obtained vehicle number, the control unit (110) can generate a virtual license plate image (300a) titled '06너4347'.

[0253] Continuing with reference to FIG. 13b, the control unit (110) can receive a portion of a license plate from a user and obtain it as a vehicle number. The control unit (110) can obtain the vehicle number '5146'. Based on the obtained vehicle number, the control unit (110) can generate a virtual license plate image (300a) of '5146'.

[0254] Next, a step (S250) of extracting a virtual license plate pattern feature (350a) based on a virtual license plate image (300a) using an artificial intelligence model (150) can be performed by the control unit (110).

[0255] The control unit (110) can extract virtual license plate pattern features (350a) based on the virtual license plate image (300a). The virtual license plate pattern features (350a) may be data and / or vectors representing unique features of the virtual license plate image (300a). Specifically, the virtual license plate pattern features (350a) may be extracted by an artificial intelligence model (150) (according to a command from the control unit (110)) based on the virtual license plate pattern image. The step of the control unit (110) extracting the virtual license plate pattern features (350a) is performed in the same manner as the extraction of the license plate pattern features (250) described above, and thus a detailed description thereof will be omitted.

[0256] Next, a step (S260) of determining the similarity between the extracted virtual license plate pattern feature (350a) and the license plate pattern feature (250) registered in the database can be performed by the control unit (110).

[0257] The control unit (110) can determine or calculate the similarity between the virtual license plate pattern feature (350a) and the license plate pattern feature (250).

[0258] Next, a step (S270) of calculating a fee to be charged to the target vehicle (10) based on the similarity may be performed. At this time, the step of calculating a fee may include a step of providing a virtual license plate image (300a) of the target vehicle (10) to the output unit (140); a step of providing fee calculation result information of the target vehicle (10) to the output unit (140); and a step of providing information on selecting at least one of a card, cash, and electronic payment method to the output unit (140).

[0259] The control unit (110) can calculate a fee to be charged to the target vehicle (10) based on the similarity. The fee may be a parking fee.

[0260] For example, if the similarity is greater than the threshold, the control unit (110) can recognize the identity of the virtual license plate pattern feature (350a) and the license plate pattern feature (250), and can perform toll settlement for the target vehicle (10). Furthermore, if there are multiple license plate pattern features (250) having a similarity with the virtual license plate pattern feature (350a) greater than the threshold, toll settlement for the target vehicle (10) can be performed by matching the license plate pattern feature (250) having the greatest similarity among the multiple license plate pattern features (250).

[0261] If the similarity is less than the threshold, the control unit (110) can acknowledge the non-identity between the virtual license plate pattern feature (350a) and the license plate pattern feature (250), and can re-perform the step of generating an arbitrary virtual license plate image (300a) based on the vehicle number input by the user.

[0262] Since the virtual license plate pattern feature (350a) and the license plate pattern feature (250) are data that are not related to the vehicle number and cannot infer the vehicle number, the user's personal information can be protected by not directly handling the vehicle number.

[0263] The control unit (110) can provide a virtual license plate image (300a) to the output unit (140). By providing the virtual license plate image (300a) to the output unit (140), the user can confirm that the target vehicle (10) for which the fee is to be settled is his / her own vehicle.

[0264] As another example, for the convenience of the user, the control unit (110) can provide the user with at least one of a license plate image (200) and an entire image of the target vehicle that matches the license plate pattern feature (250) stored in the database through the output unit (140), so that the user can confirm that the target vehicle (10) for which the fee is to be settled is his / her actual vehicle.

[0265] The control unit (110) can provide the toll settlement result information of the target vehicle (10) to the output unit (140). By providing the toll settlement result information to the output unit (140), the user can confirm the toll to be paid.

[0266] The control unit (110) can provide information on selecting at least one settlement method among card, cash, and electronic payment to the output unit (140). By providing the settlement method selection information to the output unit (140), the user can be guided to select a settlement method through the input unit (130).

[0267] The step of discarding the acquired license plate image (200) and the generated virtual license plate image (300a) can be performed by the control unit (110).

[0268] The license plate image (200) can be obtained from the target vehicle (10), and the virtual license plate image (300a) can be obtained from the vehicle number input by the user. The control unit (110) can extract license plate pattern features (250) based on the license plate image (200) and discard the license plate pattern features (250). The control unit (110) can extract virtual license plate pattern features (350a) based on the virtual license plate image (300a), provide the virtual license plate image (300a) to the output unit (140), and discard the virtual license plate image (300a). By discarding the license plate image (200) and the virtual license plate image (300a) by the control unit (110), the user's personal information can be protected.

[0269] The parking management device (100) and parking management method according to an embodiment of the present invention can recognize a target vehicle (10) without recognizing the vehicle number by determining similarity based on the license plate pattern features (250, 350a).

[0270] The parking management device (100) and parking management method according to an embodiment of the present invention can accurately recognize a target vehicle (10) even when the vehicle's entrance road is not straight or the license plate is contaminated, eliminate concerns that specific characters on the license plate are recognized as other similar characters, and protect personal information because each number on the license plate is not directly recognized.

[0271] In addition, a parking management method for recognizing a target vehicle without recognizing a vehicle number according to an embodiment of the present invention can configure a new parking management method by combining the steps (processes) described above with respect to FIGS. 1, 6, 9, 10, and 12.

[0272] In various embodiments, a parking management method for recognizing a target vehicle without recognizing a license plate number according to an embodiment of the present invention may perform a step of obtaining a first license plate image for a first license plate number of a first target vehicle entering and exiting a parking lot from an image captured in real time by a camera fixedly installed in a parking lot, with reference to FIGS. 1, 10, and 12.

[0273] In various embodiments, a parking management method for recognizing a target vehicle without recognizing a license plate number according to an embodiment of the present invention may perform a step of extracting a first license plate pattern feature based on the first license plate image using an artificial intelligence model learned as a group of multiple images, and registering the extracted first license plate pattern feature in a database, with reference to FIGS. 1, 10, and 12.

[0274] In this case, the multiple image groups have multiple license plate images for different vehicle numbers, and the multiple license plate images within one image group may have the same vehicle number but may be different images.

[0275] In various embodiments, a parking management method for recognizing a target vehicle without recognizing a license plate according to an embodiment of the present invention may perform a step of obtaining a second license plate image for a second vehicle license plate of a second target vehicle entering or exiting the parking lot from the real-time captured video with reference to FIGS. 1 and 9, extracting a second license plate pattern feature based on the second license plate image using the artificial intelligence model, and comparing the second license plate pattern feature with the first license plate pattern feature registered in the database to determine whether the second target vehicle and the first target vehicle are the same vehicle.

[0276] In this case, the comparison between the second license plate pattern features and the first license plate pattern features can be judged based on similarity.

[0277] In various embodiments, a parking management method for recognizing a target vehicle without recognizing a license plate number according to an embodiment of the present invention may perform a step of generating a virtual third license plate image based on a license plate number input by a user, extracting a third license plate pattern feature based on the third license plate image using the artificial intelligence model, and searching for a first target vehicle having the input license plate number by comparing the extracted third license plate pattern feature with the first license plate pattern feature registered in the database, with reference to FIG. 12.

[0278] As described above, the present invention has been described with reference to preferred embodiments thereof, but it will be apparent to those skilled in the art that various modifications or variations may be made to the present invention without departing from the spirit and scope of the present invention as set forth in the following claims.

[0279] [Explanation of symbols]

[0280] 10: Target vehicle

[0281] 100: Parking management device

[0282] 110: Control unit

[0283] 120: Memory

[0284] 130: Input section

[0285] 140: Output section

[0286] 150: Artificial Intelligence Model

[0287] 200: License plate image

[0288] 250: License Plate Pattern Features

[0289] 300: Body image

[0290] 350: Body pattern features

[0291] 300a: Virtual license plate image

[0292] 350a: Virtual License Plate Pattern Features

Claims

1. In a parking management method for recognizing a target vehicle without recognizing the license plate number, A step of obtaining a first license plate image of a first vehicle license plate of a first target vehicle entering or exiting a parking lot from a video captured in real time by a camera fixedly installed in the parking lot; A step of extracting first license plate pattern features based on the first license plate image using an artificial intelligence model learned from multiple image groups, and registering the extracted first license plate pattern features in a database, wherein the multiple image groups have multiple license plate images for different vehicle numbers, and the multiple license plate images within one image group have the same vehicle number but are different images; A parking management method comprising the step of obtaining a second license plate image for a second vehicle license plate of a second target vehicle entering or exiting the parking lot from the real-time captured video, extracting a second license plate pattern feature based on the second license plate image using the artificial intelligence model, and comparing the extracted second license plate pattern feature with the first license plate pattern feature registered in the database to determine whether the second target vehicle and the first target vehicle are the same vehicle.

2. In paragraph 1, A parking management method wherein the comparison between the above second license plate pattern features and the above first license plate pattern features is judged based on similarity.

3. In paragraph 1, A parking management method comprising the steps of: generating a virtual third license plate image based on a vehicle number input by a user, extracting third license plate pattern features based on the third license plate image using the artificial intelligence model, and comparing the extracted third license plate pattern features with the first license plate pattern features registered in the database to search for a first target vehicle having the input vehicle number.

4. In a parking management device that recognizes a target vehicle without recognizing the license plate number, The control unit of the above parking management device is: Obtain a first license plate image of the first vehicle license plate of the first target vehicle entering and exiting the parking lot from a video captured in real time by a camera fixedly installed in the parking lot, Using an artificial intelligence model trained with multiple image groups, a first license plate pattern feature is extracted based on the first license plate image, and registered in a database, wherein - the multiple image groups have multiple license plate images for different vehicle numbers, and the multiple license plate images within one image group have the same vehicle number but are different images -; A parking management device that obtains a second license plate image for a second vehicle license plate of a second target vehicle entering or exiting the parking lot from the real-time captured video, extracts second license plate pattern features based on the second license plate image using the artificial intelligence model, and compares the extracted second license plate pattern features with the first license plate pattern features registered in the database to determine whether the second target vehicle and the first target vehicle are the same vehicle.

5. A computer-readable recording medium having recorded thereon a program for performing a parking management method according to any one of paragraphs 1 to 3.

Images