WO2023008571A1

WO2023008571A1 - Individual identification system and individual identification method

Info

Publication number: WO2023008571A1
Application number: PCT/JP2022/029346
Authority: WO
Inventors: 亮人小泉; ティンカオ; 美江大越
Original assignee: アニコムホールディングス株式会社
Priority date: 2021-07-30
Filing date: 2022-07-29
Publication date: 2023-02-02
Also published as: JPWO2023008571A1

Abstract

Provided is an individual identification system or an individual identification method for an animal, the system comprising a means which identifies, by means of a simple method, an individual by using an image of the animal.　The individual identification system comprises: a database which stores feature amounts extracted by using a trained model from face images of animals and pieces of individual identification information about the animals; a reception means which receives an input of a face image of an animal; a feature amount extraction means which includes the trained model and extracts feature amounts from the face image input to the reception means; and a determination means which, by comparing the feature amounts stored in the database and the feature amounts extracted by the feature amount extraction means, determines whether the face image of the animal input to the reception means is an image of an individual identical or similar to an animal individual related to the individual identification information stored in the database.

Description

Individual identification system and individual identification method

The present invention relates to an individual identification system and an individual identification method, and more particularly to a system and method for identifying individual animals from facial images of animals.

Pet animals such as dogs, cats, and rabbits, and livestock such as cows and pigs are irreplaceable existences for humans. In recent years, pet animals, so-called pets, are increasingly treated in the same way as humans, and there is a growing need for individual identification of pets.

For example, pet-friendly stores and services are becoming more familiar, such as trimming salons for pets, pet cafes that allow pets, dog runs, and lodging facilities that allow pets to stay together. When trying to operate a store or service that can be used by such pets, a fee is collected for each individual pet, or an individual pet is registered and the service is provided while referring to the past service provision history. In some cases, there is a need to identify and distinguish individual pets.

In addition, when a pet goes missing, a pet search service may be requested. When searching for a missing pet, the pet is searched for based on the photo of the pet provided by the owner. Since it is a problem to determine the identity between the individual in the photograph and the individual suspected to be the animal to be searched, individual identification from the photograph of the pet is required.

Also, when providing pet insurance, it is necessary to distinguish between insured pets and uninsured pets. For example, in order to provide a policyholder with an insurance card for pet insurance so that the policyholder can receive insurance payment by presenting the insurance card at the window of a veterinary hospital, The individual pet being treated must be the same as the individual pet actually undergoing medical care. Currently, when trying to provide such pet insurance, the pet insurance management company issues an insurance card with a photo of the pet's face to the policyholder, and the policyholder presents the insurance card at the reception of the animal hospital. However, veterinary hospitals have no choice but to visually compare the photograph of the pet on the insurance card with the pet that has undergone a medical examination to determine the identity of the individual.

Therefore, there is a need for a simple method for identifying individual pets.

Patent Document 1 discloses a network camera, a presence detection sensor that detects whether or not a wild animal photographed by the camera still exists inside a capture facility, and a smartphone that can read e-mails received from the network camera. , a tablet terminal, a user information terminal such as a personal computer, etc., and a still image of an e-mail received only once or only once from the above network camera. Disclosed is an individual identification wild animal capture system that allows remote control of network cameras and capture equipment while viewing.

Further, in Japanese Patent Laid-Open No. 2002-200003, a user terminal has a photographing means for producing a sound of a frequency that causes a photographed object to turn around when photographing, and an animal identification information registering means transmits photograph data of a photograph photographed by the photographing means to the user terminal. and registers the received photo data in the management DB as animal identification information, and the notification means sends the animal-related information specified by matching that compares the images of the photos and calculates the similarity to the user terminal An animal search system that notifies is disclosed.

Further, Patent Document 3 discloses a pet individual identification system that performs individual identification of a plurality of pets. Then, another image of the pet is acquired, the feature amount is extracted from the other image, and the feature amount is compared with the stored feature amount. If the identification result is judged to be incorrect, the user is prompted to correct it. , a pet individual identification system that updates the identification result when a user's correction is obtained, and registers at least one of the another image and the feature amount extracted from the other image in association with the pet identifier. ing.

However,

Patent Documents

1 and 2 do not disclose means for identifying individuals from animal images using trained models. Further, the individual identification system described in Patent Document 3 prompts the user to correct the identification result when it is determined that the identification result is incorrect, updates the identification result when the user's correction is obtained, and At least one of the feature amount extracted from the image and another image is associated with the pet identifier and registered, and is premised on correction of the identification result from the user.

Japanese Patent Application Laid-Open No. 2019-4703 JP 2016-224640 A JP 2019-71895 A

Therefore, it is an object of the present invention to provide an individual animal identification system and an individual identification method for animals that include means for identifying individuals using animal images in a simple manner.

Insurance companies that operate health insurance for animals, so-called pet insurance, have accumulated a huge number of images of animals and information on individual animals, and the inventors of the present invention use these images. I have been trying to find out if the above problem can be solved. As a result, the inventors have found that individual identification can be easily performed by comparing features extracted from animal images using trained models, and have completed the present invention.

That is, the present invention is the following [1] to [16].
[1] A database storing feature values extracted from animal face images using trained models and animal individual identification information;
a receiving means for receiving an input of an animal face image;
a feature quantity extracting means for extracting a feature quantity from the face image input to the receiving means using the learned model;
By comparing the feature amount stored in the database and the feature amount extracted by the feature amount extraction means, the facial image of the animal input to the reception means is associated with the individual identification information stored in the database. and determination means for determining whether the image is of an individual that is the same as or similar to the animal individual.
[2] The individual identification system of [1], wherein the trained model is a neural network.
[3] The individual identification system of [2], wherein the trained model is a convolutional neural network.
[4] of [1] to [3], wherein the trained model included in the feature quantity extraction means is the same trained model as the trained model used to extract the feature quantity stored in the database; Any individual identification system.
[5] The individual identification system according to any one of [1] to [4], wherein the feature amount stored in the database is a feature amount obtained by averaging feature amounts extracted from multiple images of the same individual. .
[6] The feature quantity extraction means extracts a feature quantity from a plurality of images of the same individual, and the feature quantity used for comparison with the feature quantity stored in the database in the determination means is the same individual. The individual identification system according to any one of [1] to [5], which is a feature amount obtained by averaging feature amounts extracted from a plurality of captured images.
[7] A database storing feature values extracted from animal face images using trained models and animal individual identification information;
reception means for receiving an input of a facial image of an animal to enter or exit;
a feature quantity extracting means for extracting a feature quantity from the face image input to the receiving means using the learned model;
By comparing the feature amount stored in the database and the feature amount extracted by the feature amount extraction means, the facial image of the animal input to the reception means is associated with the individual identification information stored in the database. An entrance/exit management system, comprising: determination means for determining whether the image is of an individual that is the same as or similar to an animal individual.
[8] The entrance/exit management system of [7], wherein the database stores, as individual identification information, one or more tendencies selected from the group consisting of barking tendency, aggression, activity and sociability.
[9] further comprising alert means for issuing an alert when the determining means determines that the animal about to enter or leave is the same individual as the individual stored in the database as having a predetermined propensity [8] ] entry/exit management system.
[10] A database storing feature values extracted from animal face images using trained models and animal individual identification information;
a receiving means for receiving an input of a face image of an animal to be examined at a veterinary hospital;
a feature quantity extracting means for extracting a feature quantity from the face image input to the receiving means using the learned model;
By comparing the feature amount stored in the database and the feature amount extracted by the feature amount extraction means, the facial image of the animal input to the reception means is associated with the individual identification information stored in the database. A medical examination management system, comprising determination means for determining whether the image is of an individual that is the same as or similar to an animal individual.
[11] A database storing feature amounts and individual identification information of animals extracted from facial images of animals using trained models;
a receiving means for receiving an input of a facial image of an animal that is about to be trimmed or shampooed at a pet salon;
a feature quantity extracting means for extracting a feature quantity from the face image input to the receiving means using the learned model;
By comparing the feature amount stored in the database and the feature amount extracted by the feature amount extraction means, the facial image of the animal input to the reception means is associated with the individual identification information stored in the database. A reception management system comprising a determination means for determining whether the image is of an individual that is the same as or similar to an animal individual.
[12] A database storing feature amounts and individual identification information of animals extracted from facial images of animals using trained models;
receiving means for receiving an input of a facial image of an animal that is a candidate for an animal to be searched;
a feature quantity extracting means for extracting a feature quantity from the face image input to the receiving means using the learned model;
By comparing the feature amount stored in the database and the feature amount extracted by the feature amount extraction means, the facial image of the animal input to the reception means is associated with the individual identification information stored in the database. A search management system comprising a determination means for determining whether the image is of an individual that is the same as or similar to an animal individual.
[13] accepting an input of an animal face image;
a feature quantity extraction step of extracting a feature quantity from the face image input to the reception means using the learned model;
By comparing the feature amount extracted from the face image of the animal using the trained model and the feature amount stored in a database storing individual identification information of the animal and the feature amount extracted by the feature amount extraction means and a determination step of determining whether or not the face image of the animal input to the receiving means is an image of an individual that is the same as or similar to the individual animal associated with the individual identification information stored in the database. Method.
[14] The face image of the animal input to the receiving means by the determining means comparing the cosine similarity between the feature quantity stored in the database and the feature quantity extracted by the feature quantity extracting means. The individual identification system according to any one of [1] to [6], wherein the image is of an individual that is the same as or similar to an individual animal associated with individual identification information stored in a database.
[15] A database storing feature values extracted from face images of young animals using trained models and individual identification information of the young animals;
a receiving means for receiving an input of a facial image of an adult animal;
a feature extraction unit that includes a trained model and extracts a feature from the face image input to the reception unit using the trained model;
By comparing the feature amount stored in the database with the feature amount extracted by the feature amount extraction means, the face image of the adult animal input to the reception means is obtained as the individual identification information stored in the database. and determination means for determining whether or not the image is of an individual that is the same as or similar to the individual animal.
[16] a database storing feature amounts extracted from facial images of adult animals using trained models and individual identification information of the adult animals;
a receiving means for receiving an input of a facial image of a young animal;
a feature extraction unit that includes a trained model and extracts a feature from the face image input to the reception unit using the trained model;
By comparing the feature amount stored in the database and the feature amount extracted by the feature amount extraction means, the face image of the young animal input to the reception means is identified as the individual identification stored in the database. and determination means for determining whether or not the image is of an individual that is the same as or similar to an individual animal related to information.

According to the present invention, it is possible to provide an individual animal identification system and an individual identification method that include means for identifying individuals using animal images in a simple manner. The individual identification system and individual identification method of the present invention can be used for individual identification when entering and exiting a dog run, breeder's kennel, playground, individual identification when receiving a hotel, hospital, or pet salon, individual identification when searching for a lost child, and the like. Various applications are possible.

It is a figure showing an example of the suitable face image of an animal. It is a figure showing an example of the suitable face image of an animal. 1 is a schematic configuration diagram showing one embodiment of an individual identification system of the present invention; FIG. 1 is a schematic configuration diagram showing one embodiment of an entrance/exit management system of the present invention; FIG. It is a flowchart figure showing an example of the flow of entrance/exit management by the entrance/exit management system of this invention. It is an example of an animal face image used in the example. It is an example of an animal face image used in the example.

<Individual identification system>
The individual identification system of the present invention includes a database that stores feature values extracted from animal face images using a trained model and individual animal identification information, a reception means that receives an input of an animal face image, and a trained model. A feature amount extracting means including a model and extracting a feature amount from the face image input to the receiving means by the learned model; a feature amount stored in the database; and a feature extracted by the feature amount extracting means. Determining means for determining whether or not the face image of the animal input to the receiving means is an image of an individual that is the same as or similar to the individual animal associated with the individual identification information stored in the database, by comparing the amounts. And prepare. A similar individual means an individual whose appearance is similar to that of the target individual.

[Database]
The database of the present invention stores feature amounts and individual identification information of animals extracted from facial images of animals using learned models. Preferably, the feature amount extracted from the facial image of the animal and the individual identification information of the individual animal related to the facial image are linked and stored. Examples of individual identification information include species, breed, sex, age, weight, and body length of the animal. Also, the face image of the animal itself may be stored as the individual identification information. Alternatively, an ID number may be assigned to each animal. Furthermore, health-related information such as the animal's hospital visit history, surgery history, and drug administration history, and information related to pet salon usage history such as trimming history, shampoo history, nail clipping history, and the like may be attached to the individual identification information. Also, as the individual identification information, one or more tendencies selected from the group consisting of barking tendency, aggression, activity and sociability can be stored. The database may be managed in the form of a database server, managed on a cloud server, or distributed.

Artificial intelligence (AI) is preferable as a trained model that extracts the feature values stored in the database. Artificial Intelligence (AI) is a software or system that imitates the intellectual work of the human brain on a computer. A computer program, etc., that performs and learns from experience. Artificial intelligence may be general-purpose or specialized, and may be deep neural networks, convolutional neural networks, or the like, and open software can be used. It is preferable that the trained model for extracting the feature quantity stored in the database and the trained model included in the feature quantity extraction means described later are the same trained model.

Also, when the learned model is updated, it is preferable that the feature values stored in the database are automatically updated. Specifically, when the trained model is updated, the image stored in the database and used to extract the feature quantity is searched, and the updated trained model is used to extract the feature quantity from the image again. preferably have a program to The load on the system can be suppressed by executing this program during a time period when the system operation rate is low, for example, at night.

The feature values stored in the database are preferably obtained by averaging feature values extracted from multiple images of the same individual. By averaging the feature amounts, blurring of the feature amounts stored in the database can be suppressed. The term "shake" as used herein refers to movement of the subject such as blinking and facial expression changes at the time of image capturing, and movement due to the environment such as light and focal position. By averaging the feature amounts in this way, it becomes possible to store the unique features of the individual in the database, and the determination accuracy can be improved. In addition, the database only needs to store the averaged feature amount, and it is not necessary to store the feature amount of all images. Therefore, according to the present invention, the load on the database can also be reduced.

Averaging in the present invention refers to a method of deriving representative feature values from feature values extracted from multiple images. Therefore, in addition to the method of calculating the arithmetic mean and the geometric mean, which are general mean values, the method of obtaining the median value and the mode value is treated as averaging in the present invention.

Also, when averaging each feature amount, outliers may be excluded before averaging. Determination accuracy can be improved by excluding outliers. In addition, it can be tested whether it is an outlier by a known method. For example, a test statistic obtained by dividing the deviation by the unbiased standard deviation is obtained, and whether or not this value is greater than the significance point is tested.

Also, when averaging each feature amount, a weighted average may be performed by multiplying the feature amount by a weight according to the importance of the image. Determination accuracy can be improved by weighting the image. For example, an image with a natural expression and facing the front is given a larger weight, and a sideways image with a blink is given a smaller weight.

If the median value or the mode value is used as the average value, weighted averaging cannot be performed, but the determination accuracy can be improved by changing the count number according to the importance of the image. For example, a newly uploaded image is treated as two different images, and an old image is treated as a single image.　

[Receiving means]
The receiving means of the present invention is means for receiving input of face images of animals other than humans. Preferably, an image of an individual to be determined as to whether it is the same or similar to an individual stored in the database is accepted. The image may be a still image or a moving image. In this specification, the term "image" includes both still images and moving images. In the case of moving images, a face image is a moving image in which a face is shown. It is sufficient if the moving image includes a scene in which the face is shown. A moving image is a set of continuous still images and can be processed in the same way as for still images. Examples of animals include mammals such as dogs, cats, rabbits and ferrets, and pets such as birds and reptiles. Mammals are preferred, and dogs and cats are more preferred. The image reception method may be any method such as scanning, inputting and transmitting image data, importing images taken on the spot, and uploading moving images taken in real time. The format of the face image is not particularly limited, but the face image is preferably a photograph of the animal's face taken from the front, and more preferably a photograph showing the animal's face enlarged as shown in FIG. Such photographs include photographs such as those of a person's driver's license. Also preferred are images used on animal health insurance cards, as in FIG. In addition, if a camera installed at the entrance of a section for which entry/exit is to be managed captures an image of the animal and uploads the resulting still image or moving image to the server, the reception means of the server receives the image. Individual identification at the time of entry and exit is possible.

The image may be black and white, grayscale, or color. Images that do not show the entire animal's face, images whose shape has been edited with image editing software, images that include multiple animals, images that have faces so small that their eyes and ears cannot be recognized, or images that are unclear. I don't like it. As for the image, it is preferable that the image is normalized and the resolution and the like are unified.

[Feature quantity extraction means]
The feature amount extracting means of the present invention is means for extracting a feature amount from the face image input to the receiving means by using the learned model. Artificial intelligence (AI) is preferred as the trained model. Artificial Intelligence (AI) is a software or system that imitates the intellectual work of the human brain on a computer. A computer program, etc., that performs and learns from experience. Artificial intelligence may be general-purpose or specialized, and may be deep neural networks, convolutional neural networks, or the like, and open software can be used.

In order to generate a trained model, let artificial intelligence learn using teacher data. Learning may be machine learning or deep learning, but deep learning is preferred. Deep learning is a development of machine learning, and is characterized by automatically finding feature quantities.

The learning method for generating trained models is not particularly limited, and publicly available software can be used. For example, DIGITS (the Deep Learning GPU Training System) published by NVIDIA can be used. In addition, for example, learning may be performed by a known support vector machine method (Support Vector Machine method) published in "Introduction to Support Vector Machines" (Kyoritsu Shuppan). As a training data set, a public data set such as ImageNet can be used.

The training data for learning is animal face images and their breeds. The face image of the animal as training data may be the same as or different from the face image described in the reception method. As the facial image of the animal used as training data, it is preferable to use an image of the entire face including the eyes, nose, mouth, ears, and outline of the face of the animal.

A neural network is preferable as a trained model, and a convolutional neural network is more preferable. It is preferable that the trained model is one that has been specifically trained with images of animal species that are targets for individual identification. Further, it is preferable that the trained model for extracting the feature quantity stored in the database and the trained model included in the feature quantity extraction means are the same trained model. By using the same trained model, the same number of feature values are extracted through the trained model, which facilitates the calculation of the matching rate (cosine similarity, for example). An example of a trained model is EfficientNet trained using ImageNet. In addition, using a publicly available neural network, a trained model that can be trained to classify and estimate breeds, ages, and individual identification from facial images of dogs and extract features from facial images of dogs is available. preferable.

[Determination means]
The determining means of the present invention compares the feature amount stored in the database with the feature amount extracted by the feature amount extracting means so that the facial image of the animal input to the receiving means is stored in the database. It is determined whether or not the image is of an individual that is the same as or similar to the individual animal associated with the individual identification information. The judging means includes software that calculates, for example, the cosine similarity between the feature quantities, and judges that the individuals are the same or that they are similar when the value is equal to or greater than a predetermined value. include. When calculating the cosine similarity, it is also possible to obtain a barycentric feature amount, which is the barycenter of the feature amounts registered in the database, and to use a value obtained by subtracting the barycentric feature amount from each feature amount. Also, the normalized feature amount can be used for cosine similarity calculation. Further, it may include software for searching a database for individuals having a high cosine similarity with the feature quantity extracted by the feature quantity extracting means, and selecting identical or similar individual candidates. In addition, Euclidean distance can be used instead of cosine similarity as a method for determining identical or similar individuals.

[output]
When the determining means of the present invention receives a face image of an animal as input information, it determines the breed of the animal using the learned model.
The output format is not particularly limited. For example, on the screen of a personal computer, an individual that is the same as or similar to the individual in the image input to the input means is displayed with the image and individual identification information to display the individual identification result. can be output. If there are a plurality of individuals with similar matching rates of feature amounts in the database, a plurality of candidates for the same individual may be presented. When a plurality of candidates are presented, individual identification (identical individual determination) and similar individual determination are performed simultaneously. Moreover, as incidental information, the certainty of the individual identification result may be output at the same time. For example, "this child is ○○-chan (reliability: 80%)", "this child resembles △△-chan".
The individual identification system of the present invention may additionally have output means for receiving the determination result from the determination means and outputting the determination result.

In addition, regarding similar individual determination, the age of the individual registered in the database and the age of the individual to be determined may be significantly different. For example, similar individuals may be determined from images of puppies, taking into account growth factors of the puppies. With such a configuration, for example, when searching for a lost individual, if only a photo of the puppy remains at hand, the lost individual can be searched for by using the photo of the puppy. It becomes possible to determine whether the candidate individual is the same as or similar to the individual to be searched. In order to determine similar individuals from images of puppies, taking into account the growing factors of the puppies, learning is performed using images of puppies and images of growing up as trained models. It is preferred to use a ready-made model. In addition, by adopting a configuration in which a similar individual is determined from the image of the puppy after taking into consideration the growth factor of the puppy, for example, using the image of the puppy sold by a breeder or a pet shop, future It is also possible to show the prospective purchaser what the puppy will look like when it grows up by presenting an image of a similar individual.

That is, a preferred aspect of the present invention includes the following individual identification system.
a database storing feature values extracted from face images of young animals using trained models and individual identification information of the young animals;
a receiving means for receiving an input of a facial image of an adult animal;
a feature quantity extracting means for extracting a feature quantity from the face image input to the receiving means using the learned model;
By comparing the feature amount stored in the database and the feature amount extracted by the feature amount extraction means, the face image of the adult animal input to the reception means is obtained as the individual identification information stored in the database. and determination means for determining whether or not the image is of an individual that is the same as or similar to the individual animal.

　Early age is preferably 1 year or younger, more preferably 6 months or younger. An adult preferably refers to an individual who is over the age of 6 months or over the age of 1 year, and specifically includes over 6 months of age, over 1 year of age, and the like.

In this aspect, judging whether the images are of the same or similar individuals means that the juvenile individuals currently registered in the database and the adult individuals input to the receiving means are the same or similar. It may be determined whether or not the individual is an individual, and it may be determined whether or not the grown-up appearance of a juvenile individual registered in the database is similar to the adult individual input to the reception means. .

By using this aspect of the individual identification system, for example, when an image of an adult animal is input, it is possible to pick up individual juvenile animals that may resemble the adult animal in the future. Specifically, on the website of a pet shop incorporating the individual identification system of this aspect or the sales site of pet animals directly sold by breeders, the face image and individual identification information of the individual to be sold, and the face The features extracted from the images by the trained model are registered in a database, and the person who wishes to purchase the pet uploads the adult image of the individual he/she likes to the website. Then, the image of the adult individual uploaded to the website is accepted by the acceptance means of the individual identification system, the feature amount is extracted by the trained model, the feature amount is compared by the determination means, and grows into a similar individual in the future. Images of juvenile individuals likely to be are picked up and presented to prospective pet purchasers. This makes it possible, for example, to purchase juvenile individuals that may become individuals similar to widowed individuals from pet shops or breeders.

In addition, by using the individual identification system of this aspect, for example, when the owner has only an image of the lost individual when it was a puppy, the image of the puppy is registered in the database, and the individual identification information and the face are registered. If the feature values extracted by the trained model from the images are also registered in the database, when searching for a lost individual, a candidate individual can be found and photographed. The information is received by the receiving means, the feature amount is extracted by the trained model, the feature amount is compared by the determination means, and it is determined whether or not the candidate individual is the same as or similar to the lost individual. This makes it possible to search for a lost child even if there is only a photo of the puppy.

Still another preferred aspect is the following individual identification system.
a database storing feature amounts extracted from facial images of adult animals using trained models and individual identification information of the adult animals;
a receiving means for receiving an input of a facial image of a young animal;
a feature quantity extracting means for extracting a feature quantity from the face image input to the receiving means using the learned model;
By comparing the feature amount stored in the database and the feature amount extracted by the feature amount extraction means, the face image of the young animal input to the reception means is identified as the individual identification stored in the database. and determination means for determining whether or not the image is of an individual that is the same as or similar to an individual animal related to information.

In this aspect, determining whether the images are of the same or similar individuals means that the adult individuals currently registered in the database and the juvenile individuals input to the receiving means are the same or similar. It may be determined whether or not the individual is an individual, or whether or not the adult individual registered in the database resembles the growth figure of the young individual input to the reception means. .

By using the individual identification system of this aspect, for example, when an image of a juvenile animal is input, it is possible to pick up an adult animal individual that may resemble the juvenile animal when it grows up in the future. can. Specifically, on the website of a pet shop in which the individual identification system of this aspect is incorporated, or on the sales site of pet animals directly sold by breeders, the face of an adult individual is previously sold separately from the juvenile individual. Images, individual identification information, and features extracted from facial images by trained models are registered in a database. Then, when a person who wishes to purchase a pet animal selects an infant individual posted on the website, the face image of the infant individual is received by the acceptance means of the individual identification system, and the characteristics are determined by the trained model. The amount is extracted, the feature amount is compared by the judging means, and the image of the adult individual that the selected juvenile individual may resemble in the future is picked up. This enables the prospective purchaser to have a concrete image of what kind of individual the young animal to be considered for purchase will be when it grows up in the future, by means of images of similar individuals.

An example of an embodiment of the individual identification system of the present invention will be described with reference to FIG.
The database 12 stores in advance facial images, feature amounts extracted from the facial images using trained models, and individual identification information for a plurality of individual animals.
An image including the animal's face is captured by a capturing means 16 such as a camera installed at the entrance of the area where entry/exit of the animal is desired to be managed. Then, the photographed image is input to the receiving means 15 through the network.
The processing calculation unit 17 uses the feature amount extraction means (learned model) 18 to extract the feature amount from the image input to the reception means 15 . After that, the processing operation unit 17 refers to the database 12, and uses the determination unit 11 to match the feature amount related to the animal face image stored in the database 12 with the feature amount obtained from the received image. Calculate the rate and determine whether or not they are the same individual. The judgment is repeated for each of the animal face images stored in the database 12, and if it is judged that they are the same individual, the face image of the animal and the individual identification information of the animal are output by the output means 14. Output on the screen of the terminal 13 .

In FIG. 3, the terminal 13 is, for example, a terminal installed at the entrance of an area where entry/exit of animals is desired to be managed. Examples of the terminal 13 include a personal computer and a tablet terminal. The terminal 13 includes a processing unit such as a CPU, a storage unit such as a hard disk, ROM or RAM, a display unit such as a liquid crystal panel, an input unit such as a mouse, keyboard, touch panel, etc., and a communication unit such as a network adapter. .

In the present embodiment, an image of an animal is photographed by a photographing means such as a camera, and individual identification is performed. It is also possible to use a camera to photograph the face of the target animal, and input and transmit the photograph. For example, the user takes a photograph of the face of the animal to be insured according to instructions displayed on the screen of the terminal 13 , and sends it to the reception means 15 when an appropriate photograph is taken. At this time, the server is separately provided with a photographing assistance means consisting of an image determination program, and the photographing assistance means is an image of the individual, such as that the entire face of the animal is imaged, or that the animal's face is photographed from the front. It may be determined whether or not the photograph is suitable for identity determination, and the determination result may be transmitted to the user through an interface or terminal.

Although the server is configured by a computer in this embodiment, it may be any device as long as it has the functions of the present invention. The server may be a server on the cloud. In addition, in the present embodiment, an embodiment has been described in which the determination means and reception means are stored in the server and connected to the terminal via connection means such as the Internet or LAN, but the present invention is not limited to this. A mode in which the means and the interface unit are stored in one server or device, or a mode in which a separate terminal is not required may be employed.

The storage unit 10 is composed of, for example, a ROM, a RAM, or a hard disk. The storage unit 10 stores an information processing program for operating each unit of the server, and particularly stores the determination unit 11 .

As described above, the determination means 11 is software that calculates the matching rate of the feature amount for each image and determines whether or not the individuals are the same individual based on the matching rate value.

The processing calculation unit 17 executes identity determination using the determination means 11 stored in the storage unit.

The interface unit (communication unit) includes reception means 15 and output means 14, for example, receives a face image of an animal from a photographing means, and transmits to the terminal the determination result of identity or similarity, and the photographed individual. Output individual identification information about the same or similar individuals.

<Entrance control system>
The entrance/exit management system of the present invention includes a database storing feature values extracted from animal facial images using a trained model and animal individual identification information, and an input of an animal's facial image entering or exiting. Receiving means for receiving, feature quantity extracting means for extracting a feature quantity from the face image input to the receiving means, including a trained model, feature quantity stored in the database, and the feature quantity extracting means for extracting By comparing the obtained feature amounts, it is determined whether or not the face image of the animal input to the reception means is an image of an individual that is the same as or similar to the individual animal associated with the individual identification information stored in the database. and determination means.

INDUSTRIAL APPLICABILITY The entrance/exit management system of the present invention can be used, for example, in pet hotels, dog runs, pet cafes, etc., where it is desired to grasp and manage the entrance and exit of animals to specific areas or rooms.
The database preferably stores, as individual identification information, one or more tendencies selected from the group consisting of barking tendency, aggression, activity and sociability. In addition, in the entrance/exit management system of the present invention, when the determining means determines that an animal about to enter or exit is an individual identical or similar to an individual having a predetermined propensity stored in the database, It is preferable to further have alert means for alerting. Alert means is not particularly limited.

The reception means is not limited, and for example, it is possible to receive, through the network, images captured by photography means such as a surveillance camera fixed in the room, a network camera, or a camera attached to a smartphone.

The database, feature quantity extraction means, determination means, and output means are the same as above.

An example of an embodiment of the entrance/exit management system of the present invention will be described with reference to FIG. 4 using a case of a pet hotel.
First, the user of the pet hotel registers as a member on the website of the pet hotel through the terminal 13, and then registers the species, breed, name, date of birth, weight, sex, and vaccination status of the pet to be accommodated at the pet hotel. The face image of the pet is registered together with the individual identification information such as. The individual identification information of the registered pet is stored in the database 12 managed by the pet hotel. Also, a feature amount is extracted from the face image registered using the trained model, and stored in the database in association with the individual identification information.
Next, the user makes a reservation for the pet hotel on the website through the terminal 13, and visits the pet hotel with the pet on the reserved date and time.
An employee of the pet hotel photographs the pet by a photographing means 16 such as a camera installed in the pet hotel. Then, the photographed image is input to the receiving means 15 through the network.
The processing calculation unit 17 uses the feature amount extraction means (learned model) 18 to extract the feature amount from the image input to the reception means 15 . After that, the processing operation unit 17 refers to the database 12, and uses the determination unit 11 to match the feature amount related to the face image of the pet stored in the database 12 with the feature amount obtained from the received image. Calculate the rate and determine whether or not they are the same individual. The determination is repeated for each face image of the pet stored in the database 12, and when it is determined that they are the same individual, the face image of the pet and the individual identification information of the pet are output by the output means 14. Output on the screen of the hotel terminal. The determination result may also be output to the terminal 13 owned by the user of the pet hotel.
At this time, the pet hotel employee uses the output determination result and pet individual identification information to determine that the pet trying to use the pet hotel is the same individual as the pet registered as a member. and allow the pet to enter.

In FIG. 4, a terminal 13 is a terminal used by a user. Examples of the terminal 13 include a personal computer and a tablet terminal. The terminal 13 includes a processing unit such as a CPU, a storage unit such as a hard disk, ROM or RAM, a display unit such as a liquid crystal panel, an input unit such as a mouse, keyboard, touch panel, etc., and a communication unit such as a network adapter. .

When applying or registering as a member, the user may use a smartphone camera to take a picture of the target animal's face on the spot, enter and send it. For example, the user takes a photograph of the face of the animal to be insured according to the instructions displayed on the screen of the terminal 13 and sends it to the database 12 when an appropriate photograph is taken. At this time, the server is separately provided with a photographing assistance means consisting of an image determination program, and the photographing assistance means is an image of the individual, such as that the entire face of the animal is imaged, or that the animal's face is photographed from the front. It may be determined whether or not the photograph is suitable for identity/similarity determination, and the determination result may be transmitted to the user through an interface or a terminal.

Although the server is configured by a computer in this embodiment, it may be any device as long as it has the functions of the present invention. The server may be a server on the cloud.

As described above, the determination means 11 is software that calculates the matching rate of the feature amount for each image and determines whether the individuals are the same individual or similar individuals based on the matching rate value.

The processing calculation unit 17 uses the determination means 11 stored in the storage unit to perform identity/similarity determination.

The interface unit (communication unit) includes reception means 15 and output means 14. For example, the face image of the animal is received from the photographing means, and the determination result of identity/similarity and the identity of the photographed individual are sent to the terminal. Or output individual identification information about similar individuals.

With the entrance/exit management system of this embodiment, the pet hotel can easily determine that the pet brought by the user is the same individual as the pet registered as a member, and the reception of the pet hotel. can be done smoothly.

Fig. 5 shows a flowchart of individual identification based on the embodiment of the entrance/exit management system of the present invention. An employee of the pet hotel uses a photographing means to photograph the face of the pet brought by the user, and inputs it to the reception means of the server using a terminal (step S1). The processing operation unit of the server compares the feature amount of the uploaded face image and the face image of the animal registered in advance in the database using the determination means, and determines whether or not they are the same individual or similar. It is determined whether or not the object is an individual (step S2). The output means outputs the judgment result by displaying it on the terminal screen, etc., and presents it to the employee of the pet hotel (step S3).

<Medical examination management system>
The medical examination management system of the present invention includes a database storing feature amounts extracted from animal facial images using learned models and individual identification information of animals, and facial images of animals to be examined at a veterinary hospital. Receiving means for receiving input; feature quantity extracting means for extracting a feature quantity from the face image input to the receiving means, the feature quantity being stored in the database; By comparing the extracted feature amounts, it is determined whether or not the face image of the animal input to the receiving means is the image of an individual that is the same as or similar to the individual animal associated with the individual identification information stored in the database. and determination means for determining.

The medical examination management system of the present invention can be used, for example, when a pet insurance provider wants to know whether a pet covered by pet insurance has been examined.

<Reception management system>
The reception management system of the present invention includes a database storing feature values extracted from animal face images using learned models and animal individual identification information, and a face of an animal about to be trimmed or shampooed at a pet salon. Receiving means for receiving an input of an image; feature quantity extracting means for extracting a feature quantity from the face image input to the receiving means, including a trained model; feature quantity stored in the database; By comparing the feature values extracted by the extracting means, it is determined whether or not the facial image of the animal input to the receiving means is the image of an individual that is the same as or similar to the individual animal associated with the individual identification information stored in the database. A reception management system comprising: determination means for determining whether the

The reception management system of the present invention can be used, for example, when a pet salon wants to know whether a pet to be serviced is a pet registered as a member and the past service provision history.

<Search management system>
The management system of the present invention includes a database storing feature values extracted from animal face images using a trained model and animal individual identification information, and face images of animals that are candidates for animals to be searched. a feature extraction means for extracting a feature from the face image input to the acceptance means including a trained model; a feature stored in the database; and the feature extraction By comparing the feature values extracted by the means, whether or not the face image of the animal input to the receiving means is an image of an individual that is the same as or similar to the individual animal associated with the individual identification information stored in the database. and determination means for determining the.

The animal management system of the present invention can be used, for example, when searching for a lost pet and wanting to check whether the found animal is the pet to be searched.

<Individual identification method>
The individual identification method of the present invention includes a step of receiving an input of an animal face image, a feature extraction step of extracting a feature from the face image input to the receiving means using a trained model, and an animal face By comparing the feature amount extracted from the image using the trained model and the feature amount stored in a database storing the individual identification information of the animal and the feature amount extracted by the feature amount extraction means, the reception a determination step of determining whether the face image of the animal input to the means is an image of an individual that is the same as or similar to the individual animal associated with the individual identification information stored in the database.

The step of accepting an input of an animal's facial image is, for example, a step of accepting an input of an animal's facial image using the accepting means described above.
The feature quantity extraction step is a step of extracting a feature quantity from the face image input to the reception means using the learned model. The trained model is the same as the trained model described in the individual identification system above.
In the determination step, the feature amount extracted from the face image of the animal using the trained model and the feature amount stored in the database storing the individual identification information of the animal, and the feature amount extracted by the feature amount extraction means. It is a step of determining whether or not the face image of the animal input to the reception means is the image of an individual that is the same as or similar to the individual animal associated with the individual identification information stored in the database. For example, it is the step of calculating the matching rate of the feature amount using the above determination means and determining whether the individuals are the same or similar.

[Example 1]
Face images of each of nine miniature dachshunds (images showing only the eyes and their surroundings. As an example, the color photographs of FIGS. 6(A) to (C). They were unified to 500×500 pixels.) were prepared. .
Each image was input to the trained model and the feature amount was extracted.

A model with ImageNet weights in EfficientNetB7 was used as a trained model.

Next, a face image A of a miniature dachshund (individual to be identified; FIG. 7) for which individual identification is to be performed was prepared, and feature amounts were extracted using the above-described trained model. The same individual as the identification target individual is included in the above nine animals.

For the feature values of each of the nine images, the cosine similarity with the feature value of the face image A is calculated, and the softmax function of the scaled cosine similarity is used so that the sum of the nine images is 100%. adjusted to
As a result, it was found that the individual with the highest facial image match rate of 93% among the nine individuals was the image of the same individual as the individual to be identified.

[Example 2]
A plurality of face images similar to those in Example 1 were prepared for each of nine individuals (miniature dachshunds A to I). These images are called "database images".
Using the same trained model as in Example 1, feature amounts were extracted from each face image. Averaging (arithmetic mean) was performed on the feature amounts of multiple face images for each of the nine individuals. As shown in Table 1, the number of face images used for averaging was changed to 2, 4, 8, 16, 32, 64, and 128, and averaging was performed. rice field. In Table 1, the number of images used for averaging is "1", which is an example in which one image is used and no averaging is performed. The feature quantity obtained for each of the nine individuals is defined as the "average feature quantity".
Next, 10 facial images of miniature dachshunds to be identified were prepared for each of miniature dachshunds A to I, and feature amounts were extracted using the same trained model as in Example 1 above. These feature amounts are referred to as "discrimination target feature amounts".
For each of the nine individuals, cosine similarity was calculated between the feature amount to be identified and the averaged feature amount, and individuals with high cosine similarity were determined to be the same individual.
As a judgment result, the case where the same individual could actually be judged to be the same was defined as "correct answer". For example, it is "correct" if the image of miniature dachshund A to be identified can be determined to be the image of the same individual as the image of miniature dachshund A from among the database images. The percentage of correct answers out of a total of 90 images of 9 individuals×10 images was defined as the “correct answer rate”.
Table 1 shows the relationship between the number of sheets used for averaging and the accuracy rate.
As is clear from Table 1, the accuracy rate can be increased by averaging the feature values.

[Example 3] (Similar individual determination)
One face image was prepared for each of 452 puppies. The puppies are 3-6 month old puppies whose breed is specified by pedigree or DNA test, and the images are 500×500 pixel color images showing at least the face and its surroundings. In addition, although the breed was not limited in this test, Toy Poodle, Chihuahua, Miniature Dachshund, Shiba, Pomeranian, Miniature Schnauzer, Yorkshire Terrier, Shih Tzu, Papillon and French - Including bulldogs.
Each image was input to the trained model and the feature amount was extracted.
For EfficientNetB4, we used a model fine-tuned from a trained model with ImageNet weights.
Next, one face image was prepared for each of 100 adult dogs, and the feature amount was extracted using the learned model. Puppies and adult dogs are separate individuals.

Next, using this model, we conducted a test to see if similar puppies could be selected from 100 adult dogs. Specifically, the cosine similarity between the facial image feature amount of adult dogs (100 dogs) and the facial image feature amount of puppies (452 dogs) was calculated, and the similarities were ranked.

Regarding whether or not similar images could be selected, we first determined whether the varieties matched (objective determination). Specifically, breeds of adult dogs (100) were compared with breeds of puppies determined to have the highest degree of similarity. As a result, puppies of the same breed could be selected for 95 (95%) out of 100 adult dogs.

Next, it was determined whether or not a similar image could be selected by a specialist engaged in animal distribution (subjective determination). Specifically, five images of puppies are selected in descending order of degree of similarity, and when the puppies grow up, whether or not they resemble adult dogs (100 dogs) to be compared is determined based on the experience of experts. Judged. If even one similar puppy image was selected, it was determined as "successful determination".
When comparing the images of an adult dog and a puppy one by one, it is easy to increase the bias of feeling "dissimilar" due to subtle differences. For this reason, five images of puppies are prepared for one image of an adult dog, and an image that is not similar (ranked 5th in similarity) is also added. It weakens the bias you feel. Test results are shown in Table 2.

Claims

a database storing feature values extracted from animal face images using trained models and animal individual identification information;
a receiving means for receiving an input of an animal face image;
a feature quantity extracting means for extracting a feature quantity from the face image input to the receiving means using the learned model;
By comparing the feature amount stored in the database and the feature amount extracted by the feature amount extraction means, the facial image of the animal input to the reception means is associated with the individual identification information stored in the database. and determination means for determining whether the image is of an individual that is the same as or similar to the animal individual.
The individual identification system according to claim 1, wherein the trained model is a neural network.
The individual identification system according to claim 2, wherein the trained model is a convolutional neural network.
4. The learned model according to any one of claims 1 to 3, wherein the trained model included in said feature amount extraction means is the same trained model as the trained model used to extract the feature amount stored in said database. individual identification system.
The individual identification system according to any one of claims 1 to 4, wherein the feature amount stored in the database is a feature amount obtained by averaging feature amounts extracted from a plurality of images of the same individual.
The feature amount extraction means extracts feature amounts from a plurality of images of the same individual, and the feature amounts used for comparison with the feature amounts stored in the database in the determination means are the plurality of images of the same individual. 6. The individual identification system according to any one of claims 1 to 5, wherein the feature amount is obtained by averaging the feature amounts extracted from the images.
a database storing feature values extracted from animal face images using trained models and animal individual identification information;
reception means for receiving an input of a facial image of an animal to enter or exit;
a feature quantity extracting means for extracting a feature quantity from the face image input to the receiving means using the learned model;
By comparing the feature amount stored in the database and the feature amount extracted by the feature amount extraction means, the facial image of the animal input to the reception means is associated with the individual identification information stored in the database. An entrance/exit management system, comprising: determination means for determining whether the image is of an individual that is the same as or similar to an animal individual.
　The entrance/exit management system according to claim 7, wherein the database stores, as individual identification information, one or more tendencies selected from the group consisting of barking tendency, aggression, activity and sociability.
8. An alert means for issuing an alert when the determination means determines that an animal about to enter or leave the facility is the same or similar to an individual stored in the database as having a predetermined propensity. Access control system as described.
a database storing feature values extracted from animal face images using trained models and animal individual identification information;
a receiving means for receiving an input of a face image of an animal to be examined at a veterinary hospital;
a feature quantity extracting means for extracting a feature quantity from the face image input to the receiving means using the learned model;
By comparing the feature amount stored in the database and the feature amount extracted by the feature amount extraction means, the facial image of the animal input to the reception means is associated with the individual identification information stored in the database. A medical examination management system, comprising determination means for determining whether the image is of an individual that is the same as or similar to an animal individual.
a database storing feature values extracted from animal face images using trained models and animal individual identification information;
a receiving means for receiving an input of a facial image of an animal that is about to be trimmed or shampooed at a pet salon;
a feature quantity extracting means for extracting a feature quantity from the face image input to the receiving means using the learned model;
By comparing the feature amount stored in the database and the feature amount extracted by the feature amount extraction means, the facial image of the animal input to the reception means is associated with the individual identification information stored in the database. A reception management system comprising a determination means for determining whether the image is of an individual that is the same as or similar to an animal individual.
a database storing feature values extracted from animal face images using trained models and animal individual identification information;
receiving means for receiving an input of a facial image of an animal that is a candidate for an animal to be searched;
a feature amount extracting means for extracting a feature amount from the face image input to the receiving means, including a trained model;
By comparing the feature amount stored in the database and the feature amount extracted by the feature amount extraction means, the facial image of the animal input to the reception means is associated with the individual identification information stored in the database. A search management system comprising a determination means for determining whether or not the image is of the same individual as the animal individual.
a step of accepting an input of an animal face image;
a feature quantity extraction step of extracting a feature quantity from the face image input to the reception means using the trained model;
By comparing the feature amount extracted from the face image of the animal using the trained model and the feature amount stored in a database storing individual identification information of the animal and the feature amount extracted by the feature amount extraction means and a determination step of determining whether or not the face image of the animal input to the receiving means is an image of an individual that is the same as or similar to the individual animal associated with the individual identification information stored in the database. Method.
The determining means compares the feature amount stored in the database with the cosine similarity of the feature amount extracted by the feature amount extracting means, so that the facial image of the animal input to the receiving means is stored in the database. 7. The individual identification system according to any one of claims 1 to 6, which determines whether or not the image is of an individual that is the same as or similar to the individual animal associated with the individual identification information stored in the image.
a database storing feature values extracted from face images of young animals using trained models and individual identification information of the young animals;
a receiving means for receiving an input of a facial image of an adult animal;
a feature quantity extracting means for extracting a feature quantity from the face image input to the receiving means using the learned model;
By comparing the feature amount stored in the database and the feature amount extracted by the feature amount extraction means, the face image of the adult animal input to the reception means is obtained as the individual identification information stored in the database. and determination means for determining whether or not the image is of an individual that is the same as or similar to the individual animal.
a database storing feature amounts extracted from facial images of adult animals using trained models and individual identification information of the adult animals;
a receiving means for receiving an input of a facial image of a young animal;
a feature quantity extracting means for extracting a feature quantity from the face image input to the receiving means using the learned model;
By comparing the feature amount stored in the database and the feature amount extracted by the feature amount extraction means, the face image of the young animal input to the reception means is identified as the individual identification stored in the database. and determination means for determining whether or not the image is of an individual that is the same as or similar to an individual animal related to information.