JP7438690B2 - Information processing device, image recognition method, and learning model generation method - Google Patents

Description

The present invention relates to an information processing device, an image recognition method, and a learning model generation method, and particularly relates to an information processing device, an image recognition method, and a learning model generation method for recognizing the face of a person to be recognized.

Technology is being developed to recognize people's faces from videos. In particular, in recent years, techniques for generating a learning model using machine learning such as deep learning and reinforcement learning and recognizing images of people using the learning model have been actively used. Generating a learning model that performs highly accurate recognition requires a large amount of training data.

By the way, broadcasting stations and the like possess a large number of videos of actors, athletes, politicians, and the like. In particular, it is possible to hold not only normal faces but also images of faces with various expressions such as smiling faces and angry faces, for actors who have appeared in many appearances. On the other hand, with a few exceptions, athletes and politicians do not appear on or are broadcast on television on a daily basis, and there are not many opportunities for them to be preserved as footage, such as during tournaments and elections. That is, it is impossible to maintain facial images of various expressions for such people, and when recognizing these people as images, the absolute number of training data is insufficient.

SUMMARY OF THE INVENTION The present invention provides an information processing device, an image recognition method, and a learning model generation method that can prevent a decrease in recognition accuracy due to a lack of absolute number of training data when performing image recognition of a person's face using a learning model based on machine learning. The purpose is to provide a method.

One aspect of the present invention includes an expression change face image generation unit that generates, from a basic face image of a face of a recognition target person, an expression change face image of a face with a different expression from the facial expression of the basic face image; a learning model that inputs a face image of a person and outputs a value related to the person to be recognized, which is generated using training data including identification information of the person, a basic face image of the person to be recognized, and a facial expression change facial image; A video image comprising: a face image extraction unit that extracts a face image of a person from the video and inputs it to the learning model; and a recognition target person recognition unit that recognizes the recognition target person from the video based on an output value of the learning model. It is a processing device.

One aspect of the present invention is to generate, from a basic facial image of a face of a person to be recognized, a facial expression change face image having a facial expression different from the facial expression of the basic facial image, and to generate identification information of the person to be recognized, the recognition information, and the like. Using training data including basic facial images and expression-changing facial images of the target person, a learning model is generated that inputs the person's facial image and outputs values related to the recognition target person, and extracts the facial image of the person from the video. In this image recognition method, the information is input to the learning model, and the recognition target person is recognized from the video based on the output value of the learning model.

One aspect of the present invention is to generate, from a basic facial image of a face of a person to be recognized, a facial expression change face image having a facial expression different from the facial expression of the basic facial image, and to generate identification information of the person to be recognized, the recognition information, and the like. This learning model generation method uses training data including a basic face image and facial expression change facial images of a target person to generate a learning model whose input is a face image of a person and whose output is a value related to the recognition target person.

One aspect of the present invention is to generate a time-varying face image corresponding to elapsed time information of the face of the recognition target person from a basic face image of the face of the recognition target person, and to generate a time-varying face image corresponding to elapsed time information of the recognition target person's face, information, a basic face image of the person to be recognized, and training data including a face image that changes over time to generate a learning model that inputs the face image of the person and the elapsed time information and outputs a value related to the person to be recognized. This is a learning model generation method.

According to the present invention, when performing image recognition of a person's face using a learning model based on machine learning, it is possible to prevent a decrease in recognition accuracy due to a lack of absolute number of teacher data.

FIG. 1 is a block diagram showing the overall configuration of the first embodiment. FIG. 2 is a block diagram of the video processing device 2 according to the first embodiment. FIG. 3 is a diagram for explaining generation of a facial image with a change in expression. FIG. 4 is a diagram for explaining specific operations of the information processing device 2. As shown in FIG. FIG. 5 is a diagram for explaining specific operations of the information processing device 2. As shown in FIG. FIG. 6 is a diagram for explaining specific operations of the information processing device 2. As shown in FIG. FIG. 7 is a diagram for explaining specific operations of the information processing device 2. As shown in FIG. FIG. 8 is a block diagram of the video processing device 2 as a modification of the first embodiment. FIG. 9 is a diagram for explaining generation of a facial image with an article attached. FIG. 10 is a diagram for explaining the face image extraction unit 24 of the second modification of the first embodiment. FIG. 11 is a diagram for explaining the face image extraction unit 24 of the third modification of the first embodiment. FIG. 12 is a diagram for explaining the face image extraction unit 24 of the third modification of the first embodiment. FIG. 13 is a block diagram of the video processing device 2 according to the second embodiment. FIG. 14 is a diagram for explaining the specific operation of the second embodiment. FIG. 15 is a diagram for explaining the specific operation of the second embodiment. FIG. 16 is a block diagram of the video processing device 2 according to the third embodiment. FIG. 17 is a block diagram of a video processing device 2 according to the fourth embodiment. FIG. 18 is a diagram for explaining the specific operation of the information processing device 2 according to the fourth embodiment. FIG. 19 is a diagram for explaining specific operations of the information processing device 2 according to the fourth embodiment. FIG. 20 is a diagram for explaining the specific operation of the information processing device 2 according to the fourth embodiment. FIG. 21 is a block diagram of a video processing device 2 configured by a computer system.

<First embodiment>
A first embodiment will be described.

FIG. 1 is a block diagram showing the overall configuration of the first embodiment. In FIG. 1, 1 is a camera, 2 is a video processing device, and 3 is a display device.

Camera 1 is a camera that takes pictures. The camera 1 is preferably a 4K or 8K camera capable of capturing high-quality images for recognizing people, but is not limited to these.

The video processing device 2 inputs the video shot by the camera 1, recognizes a specific person (hereinafter referred to as a recognition target person) from among the people in the video, and outputs the result to the display device 3. It is. In this embodiment, the video processed by the video processing device 2 is the video output from the camera 1, but the video that has been shot by the camera 1 and once stored in the storage device (not in real time) may also be used. good.

The display device 3 is a display on which the captured video and the recognition results output from the video processing device 2 are output. However, the display device 3 may have not only a display function but also a display having a touch panel function, such as a tablet terminal.

Next, the video processing device 2 will be explained. FIG. 2 is a block diagram of the video processing device 2. As shown in FIG.

The video processing device 2 includes a basic face image storage section 21 , an expression change face image generation section 22 , a learning model 23 , a face image extraction section 24 , and a recognition target person recognition section 25 .

The basic face image storage unit 21 is a storage unit that stores basic face images of a plurality of recognition target persons who are recognition target persons in a video. The basic face image is a face image that shows a face with a basic expression of the person to be recognized.

The expression-changing facial image generation unit 22 uses the basic facial image to generate an image of the face of the person to be recognized (hereinafter referred to as an expression-changing facial image) having a facial expression different from the facial expression of the basic facial image. For example, the face images include a smiling face, a crying face, an angry face, a fearful face, a tired face, etc. as shown in FIG. A known technique can be used to generate the expression-changing facial image. For example, by extracting facial feature points of a basic facial image and changing the feature points according to a certain rule, it is possible to generate an expression-changing facial image with a different expression from that of the basic facial image.

The learning model 23 is a learning model that inputs a face image of a person and outputs a value related to the person to be recognized. The learning model is generated using the identification information (for example, name) of each recognition target person, the basic face image from the basic face image storage unit 21 of each recognition target person, and the facial expression change face image generation unit 22 of each recognition target person. Learning is performed using deep learning, reinforcement learning, a combination of these, etc., using training data including expression-changing facial images generated by . The value related to the person to be recognized, which is the output of the learning model 23, is, for example, the probability that the person in the input face image is the person to be recognized.

The face image extraction unit 24 detects a person's face from the input video, extracts a face image of the face, and outputs the face image and identification information that can identify the face image to the learning model 23. A known technique can be used to extract a person from a video.

The recognition target person recognition unit 25 receives the value from the learning model 23, recognizes the recognition target person, and outputs the recognition result to the display device 3. For example, among the output values of the learning model, a face in a face image with a probability greater than or equal to a predetermined threshold is recognized as the recognition target person, and a marker (for example, the face of the recognition target person) indicating the position of the recognition target person in the video is recognized. (a rectangle surrounding the rectangle) and the identification information (for example, name) of the person to be recognized are output to the display device 3.

Next, the specific operation of the information processing device 2 will be explained.

First, a basic facial image of each person to be recognized is prepared and input to the facial expression change facial image generation section 22. A plurality of facial expression change facial images of each recognition target person are generated.

FIG. 4 is a diagram for explaining the generation of a plurality of expression-changing facial images for the recognition target person X. A basic facial image of the person to be recognized X is prepared and input to the facial expression change facial image generation section 22. For example, the facial expression change face image generation unit 22 extracts the feature points of the face of the basic face image and changes the feature points according to a certain rule, thereby creating a face different from the face of the basic face image of the person to be recognized. A plurality of facial images with changing expressions (for example, a smiling face, a crying face, an angry face, etc.) are generated. This solves the problem of insufficient training data when recognizing the image of the person X to be recognized. Similarly, other recognition target person Y, recognition target person Z. ．． ．． A basic facial image of a person desired to be recognized is prepared, and the facial expression changing face image generation unit 22 generates a plurality of facial expression changing facial images with different expressions from the face in the basic facial image.

Next, a facial expression changing face is created using identification information (for example, name) of at least one person to be recognized, at least one basic face image of the person to be recognized, and at least one basic face image of the person to be recognized. A learning model 23 for recognizing a person to be recognized from the facial images is generated using the plurality of facial expression change facial images generated by the image generating unit 22 as training data. The output of the learning model 23 is the probability (likelihood) that the face of the input face image is the face of the person to be recognized.

After the learning of the learning model 23 is completed, image recognition processing of the person to be recognized is started. In the following explanation, the camera 1 photographs an election speech, and the video is the video shown in FIG. 5, which is the video displayed on the display device 3.

The image captured by the camera 1 is input to the face image extraction section 24 . The face image extraction unit 24 detects the face of a person in the video of the camera 1. The faces detected in the video of the camera 1 are face A, face B, face C, face D, face E, and face F, as shown in FIG. In the video of FIG. 6, the detected face of the person is shown as a dotted rectangle, but this only conceptually shows the detection, and the dotted rectangle is not displayed on the display device 3. .

The face image extraction unit 24 extracts images of rectangular areas surrounding each of face A, face B, face C, face D, face E, and face F from the video of camera 1, and They are output to the learning model 23 as a face image, a face image of face C, a face image of face D, a face image of face E, and a face image of face F.

The learning model 23 uses the face image of each recognition target person for the face image of face A, the face image of face B, the face image of face C, the face image of face D, the face image of face E, and the face image of face F. Outputs the probability (likelihood) that is. Here, the output of the learning model 23 is, for the person to be recognized image = 0.9, face image of face E = 0.3, face image of face F = 0.1, and for recognition target person Y, face image of face A = 0.1, face image of face B = 0.5, the face image of face C = 0.01, the face image of face D = 0.6, the face image of face E = 0.9, the face image of face F = 0.1, and the recognition target person Z For, the facial image of face A = 0.2, the facial image of face B = 0.1, the facial image of face C = 0.1, the facial image of face D = 0.2, the facial image of face E = 0. 1. The face image of face F = 0.9. ．． ．． , and so on.

The recognition target person recognition unit 25 receives the output value of the learning model 23 as input. Here, the recognition target person recognition unit 25 recognizes a face in a face image having a probability equal to or higher than a predetermined threshold value as a recognition target person. Here, if the threshold value is 0.8, then face B and face D are the faces of recognition target person X, face E is the face of recognition target person Y, and face F is the face of recognition target person Z. recognize. Then, a rectangle surrounding faces B and D and "recognition target person X", a rectangle surrounding face E and "recognition target person Y", and a rectangle surrounding face F and "recognition target person Z" are displayed on the display device 3. Output to.

FIG. 7 is an example of a recognition result displayed on the display device 3, in which faces B and D are surrounded by a rectangle, and "recognition target person X", which is identification information, is displayed below the rectangle. Similarly, the face E is surrounded by a rectangle, and the identification information "person Y to be recognized" is displayed below the rectangle. Similarly, the face F is surrounded by a rectangle, and the identification information "person F to be recognized" is displayed below the rectangle.

According to the first embodiment, it is possible to prevent a decrease in recognition accuracy due to a lack of absolute number of teacher data.

<Modification 1 of the first embodiment>
A modification of the first embodiment will be described.

FIG. 8 is a block diagram of the video processing device 2 as a modification of the first embodiment.

In a modification of the first embodiment, an article mounting image generation unit 26 is added to the information processing device 2 of the first embodiment.

Similar to the changed face image generation section 22, the article-mounted face image generation section 26 uses the basic face image to generate an image of the face of the person to be recognized who has worn the article on the face of the basic face image (hereinafter referred to as an article-mounted face image). ) is generated. The article-wearing facial image is, for example, as shown in FIG. 9, a facial image when glasses are worn on the basic facial image of the person to be recognized. Here, the article may be anything as long as it is worn on the face, and includes, for example, glasses, sunglasses, hats, helmets, and accessories.

A modification of the first embodiment has the effect of increasing the recognition accuracy of the learning model 23 by generating a face image of the person to be recognized wearing glasses or the like and using it as training data for the learning model 23. be.

<Modification 2 of the first embodiment>
A second modification of the first embodiment will be described.

The face image extraction unit 24 of the first embodiment detects all faces that appear to be people in the video. However, in videos used in broadcasting and the like, a person located near the center of the video is often an important subject. That is, the person to be recognized is often located near the center of the image. Therefore, the face image extraction unit 24 is configured to limit the area in which faces are recognized and to detect only the faces of people present in that area. In the example shown in FIG. 10, an identification target area is set near the center of the video, and only the faces of people within the identification target area are detected, so the detected faces are face B, face C, and face D. Therefore, the number of detected faces is reduced compared to the first embodiment, and the number of faces to be recognized can be reduced.

Modification 2 of the first embodiment has the effect of speeding up the entire image recognition process by limiting the region in which faces are detected and thereby reducing the number of faces detected by the face image extraction unit 24.

<Modification 3 of the first embodiment>
A third modification of the first embodiment will be described.

In videos used in broadcasting, etc., people near the center of the video are often important subjects. Furthermore, in the case of a group shot, people near the center and people located at the same depth are often equally important. In a third modification of the first embodiment, an example will be described in which these features are utilized to reduce the number of recognition targets and shorten the processing time.

Specifically, the face image extraction unit 24 detects a detectable face in the video, and calculates the size of the face (the size of the rectangle surrounding the face). In the example of FIG. 11, the faces that can be detected are faces A to F, and the sizes of faces A to F are determined.

Next, the face of the person closest to the center of the image is targeted for recognition. However, detecting a person located near the center of the image requires separate processing such as skeleton detection. Therefore, the position of a person's face near the center of the image is assumed, that position is used as a reference point, and the face closest to the reference point is set as the recognition target. Specifically, as shown in FIG. 11, for example, the reference point is the intersection of a line that divides the vertical direction into 30% to 70% from above and a line that divides the horizontal direction into 50% to 50%. do. Then, the face B closest to the reference point is set as the recognition target, and the size of the face B (the size of the rectangle surrounding the face) that is the recognition target is detected, and this is set as the reference size.

Next, faces that are smaller (for example, 70% or less) or larger (140% or more) than the standard size of the face to be recognized are excluded from the recognition target. That is, facial images of those faces are not output to the learning model 23. In the example of FIG. 11, faces that meet the above conditions and are to be recognized in addition to face B are face A and face C, and faces D, E, and F are not to be recognized. Therefore, the face images output to the learning model 23 are the face image of face A, the face image of face B, and the face image of face C, as shown in FIG.

Such processing can reduce the number of face images processed by the learning model 23, and as a result has the effect of speeding up the entire image recognition process.

<Second embodiment>
A second embodiment will be described.

In the second embodiment, image recognition is performed in consideration of changes in the facial expression of a person to be recognized over time.

FIG. 13 is a block diagram of the video processing device 2 according to the second embodiment.

The video processing device 2 of the second embodiment differs from the video processing device 2 of the first embodiment in that the learning model inputs time information (e.g., elapsed time of a match, time code) and recognizes it. The learning model 23 recognizes the recognition target person by taking into account changes in facial expressions of the target person's face over time. Here, changes in facial expressions due to elapsed time refer to changes in facial expressions over time, such as changes in facial expressions due to fatigue over time, changes in facial expressions due to aging, etc. etc.

To give a concrete explanation, as time passes from the start of a competition, athletes become tired and fatigue appears on their faces. This is especially noticeable in events such as marathons.

Therefore, the changed face image generation unit 22 generates a changed face image from the basic face image, taking into account the degree of fatigue according to the time elapsed since the start of the competition. For example, as shown in FIG. 14, a changed face image a that takes into account the degree of fatigue after one hour has elapsed from the start of the competition, and a changed face image b that takes into account the degree of fatigue that has passed two hours after the start of the competition are generated.

When the learning model 23 learns, the basic face image of each recognition target person from the basic face image storage unit 21 and the expression change face image generated by the expression change face image generation unit 22 of each recognition target person are used as teacher data. In addition to this, temporal information of facial images with changing expressions is added. In the above example, time information such as 1 lapse after the start of the competition for the changed face image a and 2 hours after the start of the competition for the changed face image b is also used as the teacher data.

On the other hand, during image recognition, in addition to the face image, time information (elapsed time from the start of the competition) is input to the learning model 23 as input data.

FIG. 15 is a diagram using a marathon as an example. After the start of the marathon, the athlete is not fatigued and has a face similar to the basic face image, but after one hour, the athlete's face becomes tired and resembles the changed face image b. Furthermore, after two hours have passed, the player's face becomes even more fatigued and becomes closer to the changed face image c.

The learning model 23 utilizes the above characteristics, and when performing image recognition, inputs the approximate elapsed time from the start of the competition together with the video, and uses this as one of the recognition parameters. Thereby, image recognition can be performed in consideration of elapsed time, and more accurate image recognition can be performed.

<Third embodiment>
A third embodiment will be described.

FIG. 16 is a block diagram of the video processing device 2 according to the third embodiment.

The video processing device 2 of the third embodiment differs from the video processing device 2 of the first embodiment in that the learning model 23 recognizes the person to be recognized using video-related information regarding the video. It is.

For videos such as news, there is video-related information such as the manuscript of the news. In many cases, a manuscript or the like includes the names and the like of the people present in the video. Therefore, the learning model 23 recognizes the identification information (for example, name) of the person to be recognized extracted from text data such as a manuscript, using the correspondence relationship between the text data and the video (for example, time code, etc.). input to the learning model 23 when the video is displayed.

The learning model 23 weights the probability of the recognition target person corresponding to the input identification information of the recognition target person. Thereby, the recognition accuracy of the learning model 23 can be improved.

<Fourth embodiment>
FIG. 17 is a block diagram of a video processing device 2 according to the fourth embodiment.

The difference between the video processing device 2 of the fourth embodiment and the video processing device 2 of the first embodiment is that the recognition target person recognition unit 25 detects changes over time in the recognized face of the recognition target person. The present invention further includes a recognition target exclusion unit 27 that excludes recognition target persons who do not do so from recognition targets.

Although the learning model 23 can identify the face of the recognition target candidate, it cannot distinguish between the recognition target person who is actually present at the location and the recognition target person who appears on a poster, painting, figurine, or the like. For example, as shown in FIG. 7, in the first embodiment, the face of the person to be recognized who is actually present at the location and the face of the person to be recognized on the poster are not distinguished, and the face of the person to be recognized is It is detected as.

However, there may be cases where it is desired to distinguish between the person to be recognized who is actually present at that location and the person to be recognized who is shown on a poster, painting, figurine, etc., and to identify only the person to be recognized who is actually present at that location.

Therefore, the person to be recognized who is actually in the location moves or smiles over time, and their movements and facial expressions change. Utilizing such properties, the recognition target exclusion unit 27 detects changes over time in the faces of these recognition target persons and excludes faces of recognition target persons for which no change over time is detected. It is possible to exclude the faces of people in photographs, figurines, etc., and only recognize the faces of the people who are actually present at the location.

Here, the term "change over time" refers to a change in the face image of the person to be recognized over time, such as a change in the position of a feature point extracted from the face image of the person to be recognized. The recognition target exclusion unit 27 specifies, as a recognition target person, a person corresponding to a face image in which feature points have changed among the face images of the recognition target person.

Next, the specific operation of the fourth embodiment will be explained.

As shown in FIG. 18, the recognition target person recognition unit 25 identifies face B as "recognition target person X," face D as "recognition target person X," face E as "recognition target person Y," and face F as "recognition target person X." Target person F' is detected.

The recognition target exclusion unit 27 acquires a predetermined frame of facial images of each recognition target person's face, and detects changes in the feature points of each recognition target person's face. Here, face B is the face of the person to be recognized who is actually present at that location, and face D, face E, and face F are facial photographs of the person to be recognized that are shown on the poster. Therefore, as shown in FIG. 19, a change in the feature points of face B is detected, but a change in the feature points of face D, face E, and face F is not detected.

The recognition target exclusion unit 27 excludes faces D, E, and F for which no changes in feature points have been detected from the faces of the recognition target person, and displays only face B for which a change in feature points has been detected as the recognition target person on the display device. Output to 3. FIG. 20 is a display example of the display device 3 in the fourth embodiment. Compared to FIG. 7, in the example of FIG. 20, only the face B of the person to be recognized X who is actually present at the location is surrounded by a rectangle, and the person to be recognized X is displayed below the rectangle.

The fourth embodiment does not recognize the face of the person to be recognized that is reflected in a poster, painting, figure, etc., but can only recognize the face of the person to be recognized who is actually present at that location.

Specifically, the video processing device 2 described above can also be realized by a computer system having a processor that performs various calculation processes.

FIG. 21 is a block diagram of a video processing device 2 configured by a computer system.

As shown in FIG. 21, the video processing device 2 includes a processor 101, a memory (ROM or RAM) 102, a storage device (hard disk, semiconductor disk, etc.) 103, an input device (keyboard, mouse, touch panel, etc.) 104, and a communication device 105. It can be configured by a computer 100 having a computer 100.

The video processing device 2 includes a facial expression change face image generation unit 22, a learning model 23, a face image extraction unit 24, and a recognition target by loading the program stored in the storage device 103 into the memory 102 and executing it by the processor 101. The functions of the person recognition section 25, the article-wearing face image generation section 26, and the recognition target exclusion section 27 are realized. Further, the basic face image storage unit 21 corresponds to the storage device 103. Note that the basic face image storage unit 21 may be physically provided outside the computer 100 and connected to the computer 100 via a network such as a LAN.

Although the present invention has been described above with reference to preferred embodiments, it is not necessary to include the configurations of all the embodiments, and not only can they be implemented in combination as appropriate, but the present invention is not necessarily limited to the above embodiments. The present invention is not limited to the above, and can be modified and implemented in various ways within the scope of the technical idea.

1 Camera 2 Video processing device 3 Display device 21 Basic facial image storage unit 22 Facial expression change facial image generation unit 23 Learning model 24 Facial image extraction unit 25 Recognition target person recognition unit 26 Article-mounted facial image generation unit 27 Recognition target exclusion unit

Claims (8)

a face image extraction unit that extracts a face image of a person from the video;
a learning model that outputs a probability that the extracted person's face image is the face of the recognition target person;
a recognition target person recognition unit that recognizes the recognition target person from the video using the probability output by the learning model;
a changed face image generation unit that generates a time-varying face image in which the face of the recognition target person changes over time from a basic face image of the face of the recognition target person;
Equipped with
The changed face image generation unit generates the time-changed face image every predetermined elapsed time,
The learning model is a learning model trained using teacher data including a basic face image of the recognition target person, the time-varying face image, and the elapsed time of the time-varying face image,
The extracted face image of the person and the elapsed time are input to the learning model.
Video processing device. The elapsed time is the time that has elapsed since the start of the predetermined competition,
The video processing device according to claim 1 . The changing face image generation unit generates the time-changing face image that takes into account fatigue of the recognition target person over time that has passed since the start of a predetermined competition.
The video processing device according to claim 2 . The changed facial image generation unit generates, from the basic facial image, a facial expression changed facial image with a facial expression different from the facial expression of the basic facial image,
The training data of the learning model further includes the facial expression change facial image.
The video processing device according to any one of claims 1 to 3 . The changed face image generation unit generates an article-attached face image in which an article is attached to the face of the recognition target person from the basic face image,
The training data of the learning model further includes the image of the face wearing the article.
The video processing device according to any one of claims 1 to 4 . The information processing device
Generating a time-varying face image in which the face of the recognition target person changes over time from a basic face image of the recognition target person's face at every predetermined elapsed time ;
Generating a learning model using training data including a basic face image of the recognition target person, the time-varying face image, and the elapsed time of the time-varying face image ;
extracting a person's face image from the video, inputting the extracted person's face image and the elapsed time to the learning model,
Recognizing the recognition target person from the video using an output value of the probability that the face of the recognition target person is the face of the learning model;
Image recognition method. The elapsed time is the time that has elapsed since the start of the predetermined competition,
The image recognition method according to claim 6 . the information processing device generates the time-varying facial image that takes into account fatigue of the recognition target person over time that has elapsed since the start of a predetermined competition;
The image recognition method according to claim 7 .

Images