Computer, program, and information processing method

Abstract

To provide a computer, a program, and an information processing method that can accurately detect a prescribed motorcycle whose total displacement is equal to or less than a prescribed value.SOLUTION: A method includes: a reception unit 24a for receiving an image 120 of an object; an estimation unit 26a for estimating whether the image 120 contains a license plate of a prescribed motorcycle using an estimation model 110 for inputting the image of the object and outputting whether or not the image contains the license plate of the prescribed motorcycle, wherein the estimation model is generated by machine learning using teacher data 100 including an image containing the license plate of the prescribed motorcycle, an image containing license plates of vehicles other than the prescribed motorcycle and an image not containing license plates, and information whether or not the image contains the license plate of the prescribed motorcycle; and an output unit 28a for outputting a warning signal when the image 120 is estimated to contain the license plate of the prescribed motorcycle.SELECTED DRAWING: Figure 3

Description

[Technical field]

[0001] The present invention relates to a computer, a program, and an information processing method. [Background technology]

[0002] Systems that prevent erroneous entry of objects prohibited from entering expressways (e.g., pedestrians, bicycles, small motorcycles) and objects traveling the wrong way onto expressways based on at least one of person detection and license plate recognition have been known, such as those disclosed in Patent Document 1. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2016-021200 A Summary of the Invention [Problem to be solved by the invention]

[0004] Conventional systems disclosed in Patent Document 1 and the like had the problem of being unable to accurately detect certain motorcycles (specifically, for example, mopeds) whose total engine displacement is below a certain value (specifically, for example, 125cc) and which are prohibited from entering expressways.

[0005] The present disclosure has been made in consideration of these points, and aims to provide a computer, a program, and an information processing method that can accurately detect a specified bike whose total displacement is equal to or less than a specified value. [Means for solving the problem]

[0006] The computer of the present disclosure includes: A reception unit that receives an image of an object captured by a camera transmitted from a terminal including the camera; an estimation unit that estimates whether or not the image contains the license plate of a specific motorcycle from the image of the object accepted by the acceptance unit, using an estimation model that takes as input an image of the object and outputs whether or not the image contains the license plate of the specific motorcycle, the estimation model being generated by machine learning using teacher data including images including the license plate of a specific motorcycle whose total displacement is equal to or less than a specific value, images including the license plate of a vehicle other than the specific motorcycle, and images including no license plate, and information regarding whether or not the image contains the license plate of the specific motorcycle; an output unit that outputs a warning signal when the estimation unit estimates that the image includes the license plate of the specified motorcycle; The present invention is characterized by comprising:

[0007] The computer of the present disclosure includes: The present invention further includes a model generation unit that uses the training data including an image including the license plate of the specified motorcycle, an image including the license plate of a vehicle other than the specified motorcycle, and an image including no license plate, and information on whether the license plate of the specified motorcycle is included in the image, and generates an estimation model by machine learning, in which an image of an object is input and whether the image includes the license plate of the specified motorcycle is output, In the training data, images containing the license plate of the specified motorcycle may be images of license plates classified based on at least one of the shape of the license plate, the color of the license plate, the size of each number on the license plate, and the position of hiragana or alphabet characters on the license plate.

[0008] In the computer of the present disclosure, In the training data, images of license plates classified into at least several types including license plates of the specified motorcycle may be used, the types of images including license plates of the specified motorcycle being hexagonal, license plates that are not white or dark green, license plates where the numbers are not uniform in size, and license plates where hiragana or alphabet characters are positioned to the left of the numbers.

[0009] In the computer of the present disclosure, The estimation unit may use the estimation model to execute a first judgment step of judging whether or not the image of the object accepted by the reception unit includes a motorcycle license plate from the image of the object accepted by the reception unit, and a second judgment step of judging whether or not the motorcycle license plate included in the image of the object is the license plate of the specified motorcycle when it is judged by the first judgment step that the image of the object includes a motorcycle license plate.

[0010] In the computer of the present disclosure, In the estimation unit, when it is determined in the first determination step that the image of the object includes a motorcycle license plate, in the second determination step, it is determined whether or not the motorcycle license plates included in each of a plurality of images including the same object are the predetermined motorcycle license plate; The output unit may output a warning signal when it is determined in the second determination step that the motorcycle license plate included in each of the plurality of images is the predetermined motorcycle license plate.

[0011] The program of the present disclosure is A program for causing a computer to function as a receiving means, an estimating means, and an output means, The receiving means receives an image of an object captured by a camera transmitted from a terminal including the camera, the estimation means estimates whether or not the image contains the license plate of the specified motorcycle from the image of the object accepted by the acceptance means, using an estimation model that takes as input an image of the object and outputs as whether or not the image contains the license plate of the specified motorcycle, the estimation model being generated by machine learning using teacher data including images including images including the license plate of a specified motorcycle whose total displacement is equal to or less than a specified value, images including the license plate of a vehicle other than the specified motorcycle, and images including no license plate, and information as to whether or not the image contains the license plate of the specified motorcycle; The output means is characterized in that it outputs a warning signal when the estimation means estimates that the image contains the license plate of the specified motorcycle.

[0012] The information processing method of the present disclosure includes: An information processing method performed by a computer having a control unit, A step of receiving an image of an object captured by a camera by the control unit, the image being transmitted from a terminal including the camera; The control unit estimates whether the image contains the license plate of the specified motorcycle from the image of the object received using an estimation model that takes as input an image of the object and outputs as whether the image contains the license plate of the specified motorcycle, the estimation model being generated by machine learning using teacher data including images including images including the license plate of a specified motorcycle whose total displacement is equal to or less than a specified value, images including the license plate of a vehicle other than the specified motorcycle, and images including no license plate, and information regarding whether the image contains the license plate of the specified motorcycle; a step of outputting a warning signal when the control unit estimates that the image includes the license plate of the predetermined motorcycle; The present invention is characterized by comprising: Effect of the Invention

[0013] According to the computer, program, and information processing method disclosed herein, it is possible to accurately detect a specified bike whose total displacement is equal to or less than a specified value. [Brief description of the drawings]

[0014] [Figure 1] 1 is a schematic diagram of a prohibited bike detection system according to the present disclosure; [Diagram 2] FIG. 2 is a schematic diagram showing the configuration of the computer and the terminal shown in FIG. [Diagram 3] 2 is a diagram showing a flow of information processing performed by a control unit executing a program in the computer shown in FIG. 1. [Figure 4] FIG. 1 illustrates an example of classified motorcycle license plates and their features in one embodiment of the present disclosure. [Diagram 5] 1 is a flowchart showing an outline of the flow of information processing according to an information processing method according to the present disclosure. [Figure 6] 1A and 1B are diagrams showing examples of moped license plate types, other motorcycle license plates, and other images that were segmented to generate a learning model for image classification. [Figure 7] 11 is a flowchart illustrating an outline of a flow of another information processing according to the information processing method according to the present disclosure. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0015] There has long been a demand for an alert to be sent when motorbikes with an engine displacement of 125cc or less (hereafter referred to as mopeds) that are prohibited from entering highway interchanges (ICs) to deter them. Tens of thousands of vehicles pass through highway ICs every day, and each of these vehicles must be detected quickly with near 100% accuracy. If a moped is detected, a notification is sent to the highway toll booth attendant, but if there are many false positives, it will cause serious disruption to the operation of the toll booth. While AI is now capable of detecting people and bicycles, there have been no cases of detecting motorbikes with an engine displacement of 125cc or less. The reason for this is that even AI using deep learning cannot accurately determine whether a motorbike has an engine displacement of 125cc or less from its appearance / shape.

[0016] As a result of intensively studying the above problems, the present inventor focused on license plates to determine whether a motorcycle has a total engine displacement of 125cc or less. Although the format of license plates for four-wheeled vehicles and motorcycles with a displacement of 125cc or more is determined, the format of license plates for mopeds is determined by each local government, and there are a wide variety of shapes, sizes, and formats, and the arrangement of characters is also different. In addition, there are many so-called local numbers. Therefore, the present inventor came up with the idea of ​​generating an estimation model by machine learning using predetermined teacher data, in which the input is an image of an object and the output is whether or not the image contains a moped license plate, and using this estimation model (trained model) to estimate whether or not the image of the object contains a moped license plate, and completed the present invention.

[0017] Hereinafter, an embodiment of the present invention will be described. Fig. 1 is a schematic diagram of a prohibited motorcycle detection system according to the present disclosure. Fig. 2 is a schematic diagram showing the configuration of a computer and a terminal shown in Fig. 1. Fig. 3 is a diagram showing the flow of information processing performed by a control unit executing a program in the computer shown in Fig. 1.

[0018] [Prohibited Bike Detection System 1] FIG. 1 shows a prohibited motorcycle detection system 1 according to the present disclosure. The prohibited motorcycle detection system 1 detects motorcycles (license plates) with a total engine displacement of 125cc or less, and can be installed near interchanges and toll booths on expressways. As shown in FIGS. 1 and 2, the prohibited motorcycle detection system 1 has a computer 2 and a terminal 3 including a camera 11. The computer 2 and the terminal 3 are connected so as to be able to communicate with each other. The computer 2, which will be described later, estimates whether or not a specific motorcycle with a total engine displacement of a specific value or less is included in an image 120 or a video 130 taken by the camera 11. In this specification, the term "motorcycle" refers to so-called two-wheeled vehicles, single vehicles, motorcycles, etc., and includes various types. If the total engine displacement of a motorcycle is 125cc or less, it is a motorized bicycle, including a first-class moped and a second-class moped. In addition to motorcycles, the term "vehicle" includes vehicles that are required to have a license plate attached, such as ordinary cars, small cars, light cars, and large special-purpose cars.

[0019] <Computer 2> The configuration of the computer 2 of the present disclosure will be described with reference to Fig. 2. Fig. 2 is a schematic diagram illustrating the configuration of the computer 2 of the present disclosure. The computer 2 of this embodiment is composed of an industrial computer, a tablet terminal, or the like, and includes a control unit 20, a storage unit 30, a communication unit 40, a display unit 50, and an operation unit 60, as shown in Fig. 2.

[0020] The control unit 20 is composed of a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), an AI inference device, etc., and controls the operation of the computer 2. Specifically, the control unit 20 executes programs stored in a storage unit 30, which will be described later, thereby functioning as a model generating means 22, a receiving means 24, an estimating means 26, and an output means 28. Each of these means will be described later.

[0021] The storage unit 30 is configured with, for example, a hard disk drive (HDD), a random access memory (RAM), a read only memory (ROM), a solid state drive (SSD), etc. Moreover, the storage unit 30 is not limited to being built into the computer 2, and may be a storage medium (for example, a USB memory) that is detachably attached to the computer 2. In this embodiment, the storage unit 30 is configured to store the program executed by the control unit 20, the estimation model 110, etc.

[0022] The communication unit 40 includes a communication interface for transmitting and receiving signals to and from an external device wirelessly or via a wired connection. The communication unit 40 allows the control unit 20 to transmit and receive signals to and from the terminal 3.

[0023] The display unit 50 is, for example, a monitor, and displays various screens by receiving a display command signal from the control unit 20. The operation unit 60 is, for example, a keyboard, and is capable of giving various commands to the control unit 20. Note that in this embodiment, a display operation unit such as a touch panel in which the display unit 50 and the operation unit 60 are integrated may be used.

[0024] (Details of control unit 20) (Model generation means 22) The model generation means 22 uses teacher data 100 including images including a license plate of a specific motorcycle whose total displacement is equal to or less than a specific value, images including license plates of vehicles other than the specific motorcycle, and images including no license plate, and information on whether or not the license plate of the specific motorcycle is included in the image, and generates an estimation model 110 by machine learning, in which the input is an image of an object and the output is whether or not the image includes the license plate of the specific motorcycle. Various types of machine learning such as deep learning can be used.

[0025] Among the teacher data 100, "images including the license plate of a specific motorcycle (whose total displacement is equal to or less than a specific value)" include, for example, images including the license plate of a moped with a total displacement of 125cc or less, such as an image including a license plate as shown in Fig. 4(a). Fig. 4 shows examples of motorcycle license plate shapes and character arrangements classified by the inventor after examining types of motorcycles over 125cc / under 125cc through nationwide network collection and photography in order to determine the license plates of motorcycles (mopeds) of 125cc or less.

[0026] Furthermore, as shown in Fig. 4, as the "image including a specific motorcycle license plate", an image of a license plate classified based on at least one of the following (features of the license plate): the shape of the license plate, the color of the license plate, the size of each number on the license plate, and the position of the hiragana or alphabet on the license plate. More specifically, as the type of "image including a specific motorcycle license plate", an image of a license plate classified into at least a plurality of types of license plates, for example, hexagonal license plates, license plates that are not white or dark green, license plates with non-uniform sizes of each number, license plates with hiragana or alphabets to the left of the numbers, license plates with dark blue characters, and license plates with a white / yellow / pink / light blue background, may be used.

[0027] Among the teacher data 100, examples of "images including license plates of vehicles other than a specific motorcycle" include, in the above example, images including license plates of motorcycles with total engine displacement of more than 125cc that is not 125cc or less, images including license plates of automobiles, etc. Also, examples of "images including images that do not include license plates" include images of bicycles or people, and images of objects similar in shape to license plates (for example, see FIG. 6(f)).

[0028] In the training data 100, "information as to whether or not the image contains the license plate of a specific motorcycle" refers in the above example to the presence or absence of a moped in the image.

[0029] (Means of reception 24) The receiving means 24 receives an image 120 of an object captured by a camera, which is transmitted from the terminal 3 including a camera. The receiving means 24 may also receive a video of the object captured by the camera, and may convert this video into the image 120 by a decoding unit (not shown) of the computer 2 (video decoding).

[0030] (Estimation means 26) The estimation means 26 estimates whether or not image 120 contains the license plate of a specified motorcycle from image 120 of an object accepted by the acceptance means 24 using an estimation model 110 which takes as input an image of an object and outputs whether or not the image contains the license plate of a specified motorcycle, the estimation model 110 being generated by machine learning using training data 100 which includes images including images including the license plate of a specified motorcycle, images including the license plate of a vehicle other than the specified motorcycle, and images including no license plate, as well as information as to whether or not the image contains the license plate of the specified motorcycle. As described later, the estimation means 26 may use the estimation model 110 to execute a first determination step of determining whether or not the image 120 of the object received by the receiving means 24 includes a motorcycle license plate from the image 120 of the object received by the receiving means 24, and a second determination step of determining whether or not the motorcycle license plate included in the image 120 of the object is a license plate of a predetermined motorcycle when it is determined in the first determination step that the image 120 of the object includes a motorcycle license plate. In this case, when it is determined in the first determination step that the image 120 of the object includes a motorcycle license plate, the estimation means 26 may determine in the second determination step whether or not the motorcycle license plate included in each of the multiple images 120 including the same object is a license plate of a predetermined motorcycle. The estimation model 110 used is not limited to the one generated by the model generation means 22, and an estimation model 110 generated externally may be used.

[0031] (output means 28) The output means 28 outputs a warning signal when the estimation means 26 estimates that the image 120 includes the license plate of a specific motorcycle. The information output from the output means 28 that the image 120 includes the license plate of a specific motorcycle, i.e., that the specific motorcycle has been detected, may be displayed on the display unit 50 or a terminal such as a highway toll booth. In addition, the warning signal may be output by outputting a sound or blinking a light, or may be outputted simultaneously with a sound or a light. As shown in FIG. 1, a warning may be displayed on a display board of the no-entry motorcycle detection system 1.

[0032] 3, in this embodiment, the control unit 20 executes a program stored in the storage unit 30 to function as the model generation means 22, thereby forming a model generation unit 22a. The control unit 20 executes a program stored in the storage unit 30 to function as the reception means 24, thereby forming a reception unit 24a. The control unit 20 executes a program stored in the storage unit 30 to function as the estimation means 26, thereby forming an estimation unit 26a. The control unit 20 executes a program stored in the storage unit 30 to function as the output means 28, thereby forming an output unit 32a.

[0033] (Terminal 3) Next, a description will be given of the configuration of the terminal 3. The terminal 3, for example, monitors vehicles passing through an interchange on a highway, and includes a camera 11 and a communication unit 12 as shown in FIG.

[0034] The camera 11 captures video of vehicles including motorcycles. The camera 11 included in the terminal 3 includes an object detection camera that captures images of targets such as vehicles and a license plate detection camera. Any known camera can be used as the camera 11. Still images 120 of each frame are acquired from the video of the vehicle captured by the camera 11. Note that it is preferable to also use an infrared camera as the camera 11 in order to reduce halation that occurs when capturing images at night or on license plates.

[0035] The communication unit 12 includes a communication interface for transmitting and receiving signals to and from an external device wirelessly or via a wire. The video and still images 120 captured by the camera 11 are transmitted to the computer 2 by the communication unit 12.

[0036] [Information processing method 1] Next, an information processing method by the computer 2 (the intrusion prohibited bike detection system 1) will be described with reference to Fig. 5. The following processing is performed by the control unit 20 executing a program stored in the storage unit 30.

[0037] First, the control unit 20 uses training data 100 including images including a specified motorcycle license plate, images including license plates of vehicles other than the specified motorcycle, and images including no license plate, as well as information as to whether or not the image includes the specified motorcycle license plate, to generate an estimation model 110 through machine learning, whose input is an image of an object and whose output is whether or not the image includes the specified motorcycle license plate (Figure 5, step S10).

[0038] Next, the control unit 20 receives the image 120 of the object captured by the camera 11, which is transmitted from the terminal 3 including the camera 11 (FIG. 5, step S20).

[0039] Next, the control unit 20 uses the estimation model 110 generated in step S10 to estimate whether or not the image 120 received in step S20 includes a license plate of a predetermined motorcycle (FIG. 5, step S30).

[0040] Next, when the control unit 20 estimates that the image 120 includes the license plate of a predetermined motorcycle (FIG. 5, step S40: YES), it outputs a warning signal (FIG. 5, step S50).

[0041] On the other hand, if the control unit 20 estimates that the image 120 does not include the license plate of the predetermined motorcycle (FIG. 5, step S40: NO), the warning signal is not output and the process returns to step S20.

[0042] [Information processing method 2] Logically, it is possible to read the characters on the photographed license plate and determine whether it is a specific motorcycle or not based on the type of license plate as described above. However, motorcycle license plates are generally small, making them difficult to identify using AI. Motorcycles move quickly at ICs, so subject blurring is likely to occur when reading license plates. Subject blurring of a moving motorcycle and glare from taillights at night can make the characters unclear, making it difficult for deep learning object detection to distinguish small objects. In particular, the hiragana portion of the license plate is small and difficult to read accurately. Therefore, rule-based judgments such as those described above can be difficult.

[0043] Therefore, we decided to adopt a method of image classification using deep learning instead of characters as a method of comprehensively judging license plates. However, with simple two-class classification, it is difficult to control which features the neural network learns, and there is a concern that if a sequence of numbers such as "1234" is included as training data 100, a license plate of a motorcycle with a total engine displacement of 126 or more containing "1234" may be falsely detected as a moped license plate (overlearning).

[0044] Therefore, the present inventor discovered that by subdividing the classification of license plates into subclasses each having a common feature, finer feature amounts can be extracted, and mopeds can be detected with higher accuracy. Hereinafter, such an information processing method 2 according to the present disclosure will be described.

[0045] Specifically, for example, subclasses of hexagonal license plates (FIG. 6(a)), license plates with small numbers on the left and large numbers on the right (FIG. 6(b)), and license plates other than these (FIG. 6(c)) can be set, and images including each of these license plates can be used as training data 100 to generate an estimation model 110 by machine learning, in which the input is an image of the object and the output is whether or not the image includes a license plate of a moped. In addition, in relation to plates such as those in FIGS. 6(a) to 6(c), images with subject blur, unclear images, and images taken from various angles can be additionally learned. In addition to the above, false detections can be reduced by additionally learning images other than moped license plates such as those in FIGS. 6(d) to 6(f).

[0046] In addition, license plates having characteristics of motorcycles of 125cc or less and motorcycles of over 125cc may be additionally trained as shown in Fig. 4. For example, as types of images including moped license plates, images of license plates classified into at least two of the following types may be used: license plates that are hexagonal, license plates that are not white or dark green, license plates with numbers that are not uniform in size, license plates with hiragana or alphabet characters to the left of the numbers, license plates with dark blue characters, and license plates with a white / yellow / pink / light blue background.

[0047] As described above, according to information processing method 2, it is possible to make the robot learn that, for example, hiragana characters are written on the left side of the license plate of a moped (Fig. 4(a), Fig. 6(c)), and hiragana characters are written on the top right side of the license plate of a motorcycle over 125cc (Fig. 4(b), Fig. 6(d)). In addition, the robot can learn that the feature of a hexagon is more important than the sequence of numbers, and that the feature of a sequence of characters being smaller on the left and larger on the right also has a high priority.

[0048] [Information processing method 3] In the information processing method according to the present disclosure, inference in two stages can be performed in order to suppress erroneous detection and erroneous determination. This information processing method 3 will be described below. FIG. 7 shows the processing flow of information processing method 3.

[0049] First, a plurality of images 120 are acquired from the video 130 transmitted by the terminal 3 (video decoding; FIG. 7, step S1).

[0050] Next, object detection (primary inference, first judgment) is performed on these images 120 to judge the bodies of the car and motorcycle and the license plates in these bodies (FIG. 7, step S2). Here, the control unit 20 (estimation unit 26a) uses the estimation model 110 to judge whether or not the image 120 of the object accepted by the acceptance unit 24a includes the license plate of the motorcycle from the image 120 of the object accepted by the acceptance unit 24a (first judgment step). In this way, the (license plate of the) car is excluded, and only the license plate of the motorcycle is extracted, and only the license plate of the motorcycle is subject to judgment in the following step. In addition, when something other than the license plate (for example, FIG. 6(f) etc.) is erroneously detected, it is possible to suppress erroneous judgment as a specific motorcycle (for example, a moped). The estimation model 110 may be generated by the information processing method 1 or 2. YOLO, SSD, etc. can be used as the object detection AI.

[0051] Next, the same license plate that was determined to be that of a motorcycle in step S2 is tracked using multiple images 120 (tracking; FIG. 7, step S3). As mentioned above, motorcycles move quickly on ICs and other roads, so this is to prevent false detections and erroneous judgments caused by the license plate becoming unclear. This tracking is used to determine whether or not the motorcycle is the same from multiple images.

[0052] Next, image classification (secondary inference, second judgment) is performed to judge whether or not it is a predetermined motorcycle license plate (FIG. 7, step S4). Here, when it is judged by the above-mentioned first judgment step that the image 120 of the object contains a motorcycle license plate, the control unit 20 (estimation unit 26a) judges whether or not the license plate contained in the image 120 of the object (motorcycle) is a predetermined license plate (second judgment step).

[0053] To prevent false positives, this image classification may be performed multiple times. For example, if inference is performed for each frame with a camera of about 20 fps, the results may be blurred. That is, in this second judgment step, it may be judged whether the license plate of a motorcycle included in each of the multiple images 120 including the same object is the license plate of a specific motorcycle. When 10 images are taken per second, if it is judged that the specific license plate is present at least three times in five consecutive frames, it may be judged that the specific motorcycle (license plate) is present in the image 120. A false positive filter may be introduced to activate an alert when the judgment of 125cc or less is made for several frames in step S4.

[0054] If it is determined in step S4 that the image 120 contains the license plate of a specific motorcycle, an alert is issued warning the driver not to enter the expressway (FIG. 7, step S5).

[0055] According to the computer 2, program, and information processing method of this embodiment, which are configured as described above, there are provided a reception unit 24a that receives an image 120 of an object captured by the camera 11 and transmitted from the terminal 3 including the camera 11, an estimation unit 26a that estimates whether or not the image 120 contains the license plate of a specified motorcycle from the image 120 of the object received by the reception unit 24a using an estimation model 110, which takes as input the image of the object and outputs as whether or not the image contains the license plate of the specified motorcycle, and which is generated by machine learning using teacher data 100 including images including the license plate of a specified motorcycle whose total displacement is equal to or less than a specified value, images including the license plate of a vehicle other than the specified motorcycle, and images including no license plate, and information regarding whether or not the image contains the license plate of the specified motorcycle, and an output unit 28a that outputs a warning signal when the estimation unit 26a estimates that the image 120 contains the license plate of the specified motorcycle.

[0056] According to the computer 2, program, and information processing method disclosed herein, it is possible to accurately detect a specific motorcycle whose total displacement is equal to or less than a specific value. By introducing images including license plates of vehicles other than the specific motorcycle to be detected and images not including license plates, it is possible to reduce false positives.

[0057] In addition, in the computer 2, program, and information processing method of this embodiment, a model generation unit 22a can be further provided that uses teacher data 100 including an image including a license plate of a specific motorcycle, an image including a license plate of a vehicle other than the specific motorcycle, and an image including no license plate, and information on whether the license plate of the specific motorcycle is included in the image, and generates an estimation model 110 by machine learning, in which the input is an image of an object and the output is whether the license plate of the specific motorcycle is included in the image. Here, in the teacher data 100, as an image including the license plate of the specific motorcycle, an image of a license plate classified based on at least one of the shape of the license plate, the color of the license plate, the size of each number on the license plate, and the position of hiragana or alphabet on the license plate may be used. In this way, by performing image classification based on the characteristics of the specific motorcycle license plate, that is, based on the shape, color, size of the number, and position of hiragana or alphabet, it is possible to more accurately estimate whether the image 120 includes the license plate of the specific motorcycle.

[0058] In addition, in the computer 2, the program, and the information processing method of this embodiment, the teacher data 100 may use images of license plates classified into at least a plurality of types of license plates, including hexagonal license plates, license plates that are not white or dark green, license plates with numbers that are not uniform in size, and license plates with hiragana or alphabet characters to the left of the numbers, as types of images including the license plate of a specific motorcycle. In this way, if the specific motorcycle is a moped, it is possible to learn features other than the major features mentioned above, such as hiragana characters on the left of license plates of motorcycles with a total engine displacement of 125cc or less, and hiragana characters on the upper right of license plates of motorcycles with a total engine displacement of over 125cc. It is also possible to teach that the hexagonal feature is more important than the arrangement of numbers, and that the feature of small characters on the left and large characters on the right in the arrangement of characters is also of high priority.

[0059] In the computer 2, the program, and the information processing method of the present embodiment, the estimation unit 26a may use the estimation model 110 to execute a first determination step of determining whether or not the image 120 of the object received by the reception unit 24a includes a motorcycle license plate from the image 120 of the object received by the reception unit 24a, and a second determination step of determining whether or not the motorcycle license plate included in the image 120 of the object is a license plate of a predetermined motorcycle when it is determined in the first determination step that the image 120 of the object includes a motorcycle license plate. In this way, by performing two-stage inference, it is possible to further suppress erroneous detection and erroneous determination.

[0060] In the computer 2, program, and information processing method of this embodiment, when the estimation unit 26a determines in the first determination step that the image 120 of the object contains a motorcycle license plate, the second determination step determines whether the motorcycle license plates contained in each of the multiple images 120 containing the same object are the license plates of a specific motorcycle, and when the second determination step determines that the motorcycle license plates contained in each of the multiple images 120 are the license plates of the specific motorcycle, the output unit 32a may output a warning signal. In this way, the detection accuracy can be improved.

[0061] The computer 2, the program, and the information processing method according to the present disclosure are not limited to the above-described aspects and combinations, and various modifications can be made.

[0062] For example, an example has been described in which images including license plates of vehicles other than a specified motorcycle and images not including license plates are used as images of the training data 100 for generating the estimation model 110. However, an estimation model can also be generated using only images including the license plate of a specified motorcycle, as shown in FIG. 4(a), which is the minimum necessary image.

[0063] Also, in the image 120 of the object accepted by the accepting unit 24a, a region (ROI: Region on Interest) for determining whether or not there is a license plate can be set. In this case, the ROI can be set on the image 120 according to the camera angle of each location where the terminal 3 is installed. By using the object or license plate in this region for the judgment, it is possible to prevent a car fading out of the screen of the camera 11 from being mistakenly recognized as a motorcycle, and to prevent the motorcycle from being mistakenly determined as a moped of 125cc or less because the license plate becomes small when the motorcycle fades out into the distance. As a result, the accuracy of judging a specified motorcycle is improved.

[0064] Furthermore, although the above describes an example in which the purpose is to detect the license plate of a motorcycle (moped) with a total displacement of 125cc or less, it is also possible to apply the computer 2, program, and information processing method disclosed herein to detect a kick scooter or the like having a license plate. [Explanation of symbols]

[0065] 1. Prohibited Bike Detection System 2. Computer 3. Terminal 11 Camera 12 Communications Department 20 Control section 22 Model Generation Method 24 Reception methods 26 Estimation means 28 Output Method 22a Model generation section 24a Reception 26a Estimation part 28a Output section 30 Storage section 40 Communications Department 50 Display 60 Control section 100 Teacher Data 110 Estimation Model 120 Still Images 130 videos

Claims

1. A reception unit that receives an image of an object captured by a camera transmitted from a terminal including the camera; an estimation unit that estimates whether or not the image contains the license plate of a specific motorcycle from the image of the object accepted by the acceptance unit, using an estimation model that takes as input an image of the object and outputs whether or not the image contains the license plate of the specific motorcycle, the estimation model being generated by machine learning using teacher data including images including the license plate of a specific motorcycle whose total displacement is equal to or less than a specific value, images including the license plate of a vehicle other than the specific motorcycle, and images including no license plate, and information regarding whether or not the image contains the license plate of the specific motorcycle; an output unit that outputs a warning signal when the estimation unit estimates that the image includes the license plate of the specified motorcycle; A computer equipped with

2. The present invention further includes a model generation unit that uses the training data including an image including the license plate of the specified motorcycle, an image including the license plate of a vehicle other than the specified motorcycle, and an image including no license plate, and information on whether the license plate of the specified motorcycle is included in the image, and generates an estimation model by machine learning, in which an image of an object is input and whether the image includes the license plate of the specified motorcycle is output, The computer of claim 1, wherein in the training data, images of license plates classified based on at least one of the shape of the license plate, the color of the license plate, the size of each number on the license plate, and the position of hiragana or alphabet characters on the license plate are used as images including the license plate of the specified motorcycle.

3. The computer of claim 2, wherein in the training data, images of license plates classified into at least a plurality of types including the license plate of the specified motorcycle are used. The image types include license plates that are hexagonal, license plates that are not white or dark green, license plates in which the numbers are not uniform in size, and license plates in which hiragana or alphabet characters are positioned to the left of the numbers.

4. The computer of claim 1, wherein the estimation unit uses the estimation model to execute a first judgment step of judging whether or not the image of the object accepted by the reception unit includes a motorcycle license plate from the image of the object accepted by the reception unit, and a second judgment step of judging whether or not the motorcycle license plate included in the image of the object is the license plate of the specified motorcycle when it is judged by the first judgment step that the image of the object includes a motorcycle license plate.

5. In the estimation unit, when it is determined in the first determination step that the image of the object includes a motorcycle license plate, in the second determination step, it is determined whether or not the motorcycle license plates included in each of a plurality of images including the same object are the predetermined motorcycle license plate, 5. The computer according to claim 4, wherein the output unit outputs a warning signal when the second determination step determines that the motorcycle license plate included in each of the plurality of images is the predetermined motorcycle license plate.

6. A program for causing a computer to function as a receiving means, an estimating means, and an output means, The receiving means receives an image of an object captured by a camera transmitted from a terminal including the camera, the estimation means estimates whether or not the image contains the license plate of the specified motorcycle from the image of the object accepted by the acceptance means, using an estimation model that takes as input an image of the object and outputs as whether or not the image contains the license plate of the specified motorcycle, the estimation model being generated by machine learning using teacher data including images including images including the license plate of a specified motorcycle whose total displacement is equal to or less than a specified value, images including the license plate of a vehicle other than the specified motorcycle, and images including no license plate, and information as to whether or not the image contains the license plate of the specified motorcycle; The output means outputs a warning signal when the estimation means estimates that the image contains the license plate of the specified motorcycle.

7. An information processing method performed by a computer having a control unit, A step of receiving an image of an object captured by a camera by the control unit, the image being transmitted from a terminal including the camera; The control unit estimates whether the image contains the license plate of the specified motorcycle from the image of the object received using an estimation model that takes as input an image of the object and outputs as whether the image contains the license plate of the specified motorcycle, the estimation model being generated by machine learning using teacher data including images including images including the license plate of a specified motorcycle whose total displacement is equal to or less than a specified value, images including the license plate of a vehicle other than the specified motorcycle, and images including no license plate, and information regarding whether the image contains the license plate of the specified motorcycle; a step of outputting a warning signal when the control unit estimates that the image includes the license plate of the predetermined motorcycle; An information processing method comprising:

Images