KR102125633B1 - System and method for detecting and specifiing animal - Google Patents

Abstract

The present invention relates to an animal individual identifying and specifying system and method thereof including: an image capturing unit acquiring a nose image of an animal; an input unit for information input on the animal; an image alignment unit aligning the size and angle of the nose image acquired by the image capturing unit focusing on the concave part of the upper lip and the nostril; a feature extraction unit extracting a species classification feature corresponding to the nostril and the nose outline from the image aligned by the image alignment unit and extracting an individual classification feature corresponding to dot and net patterns with respect to certain parts between the nostrils and above the gap; a registration/search unit causing the feature extracted by the feature extraction unit and the animal-related information to be transmitted to an external system and registered as data and causing the animal-related information to be searched for by comparing the feature extracted by the feature extraction unit with data registered in the external system; and a display unit displaying the searched animal-related information or the data registered by the registration/search unit. According to the present invention, dual extraction is performed with respect to the species classification feature and individual classification feature, and thus it is possible to identify an animal such as a pet dog and accurately specify the individual. In addition, the accuracy of animal individual recognition can be increased by image alignment and so on regardless of the recognition target animals′ postures.

Description

System and method for identification and characterization of animals and individuals{SYSTEM AND METHOD FOR DETECTING AND SPECIFIING ANIMAL}

The present invention relates to an animal identification and characterization system and method of an animal, and more specifically, by extracting the doubles for the total classification feature and the individual classification feature, to accurately identify and specify an animal entity such as a companion dog, and to align images The present invention relates to an animal identification and characterization system and method for improving the accuracy of animal recognition even in various postures of a recognized animal.

In recent years, the number of organic and lost animals has increased significantly, and dogs account for a large proportion of them, which is a public health problem in that it is a vehicle for important diseases such as rabies, brucellosis, and heartworm infections. In spite of the fact that they are becoming, many of these facilities are lacking.

To solve this problem, in accordance with the current animal protection law, organic animals, including organic dogs, were protected in an animal shelter for 10 days and then euthanized if the owner did not appear, which had other problems in bioethics.

Therefore, in order to solve these social problems, the government has implemented an animal registration system so that it is easy to find the owner of the organic animal, and the animal registration system allows the owner of the pet to minimize the organic animal. It is legislated to register with the military and ward offices, and is divided into an internal type that inserts a wireless identification chip into the companion animal's body and an external type that hangs the identification tag attached with the wireless identification chip on the animal's neck.

However, in the case of the built-in type, there is a problem that stability is relatively low compared to the external type in that a procedure for planting the wireless identification chip in the animal's body is required, and in the case of the external type, the owner forgets to wear the identification tag on the companion animal or is not allowed. There is a problem in that the convenience of management is poor, such as the loss of the identification tag by animals.

In order to solve the above-described problems in the prior art, the present invention can identify animals such as companion dogs by accurately extracting the total classification feature and the individual classification feature, and accurately identify the individual, and recognize the object by image alignment, etc. It is possible to improve the accuracy of recognition of individual animals even in various postures of animals, so that the owner of animals such as stray dogs can be found accurately, making it easy and quick to find owners of missing dogs, etc. In addition, it aims to contribute to the systematic management of companion animals and to improve the welfare of animals.

In order to solve the problems as described above, according to an aspect of the present invention, an image photographing unit for acquiring a nose image of an animal; An input unit for inputting related information about the animal; An image aligning unit that aligns the size and angle of the image around the nose hole and the concave portion of the upper lip of the nose image obtained by the image photographing unit; From the image aligned by the image alignment unit, a vertical classification feature corresponding to a nostril and an outline line of the nose is extracted, and an individual classification characteristic corresponding to a dot pattern and a mesh pattern is targeted to a predetermined region between the nostrils and the nostrils. Feature extraction unit for extracting; The feature extracted by the feature extracting unit is transmitted to an external system together with the related information about the animal to be registered as data, and the feature extracted by the feature extracting unit is compared with data registered in the external system. A registration/search unit to search related information of animals; And a display unit configured to display data registered by the registration/search unit or related information of the searched animal. An object identification and characterization system for animals is provided.

The feature extracting unit may include: an image normalizing unit for extracting a quadrangular area including a nose region from the image aligned by the image aligning unit and normalizing to a predetermined pixel size; A vertical classification feature that extracts a vertical classification feature vector by applying an inception V2 based DNN using deep learning-based object recognition technology by inputting RGB image with 3 channels to the nose region image normalized by the image normalization unit. Vector extraction unit; An image dividing unit for dividing the nose region image into local square blocks of overlapping predetermined pixel sizes; An image conversion unit that converts a local channel into a single channel image by applying an edge enhancement filter to each of the local square blocks divided by the image division unit; A pattern feature code vector extracting unit that extracts a pattern feature code vector corresponding to an individual classification feature for each of the local square blocks by applying a DNN-based autoencoder to each of the local square blocks; And a feature vector connecting unit constituting the entire feature by connecting the vertical feature vector and the pattern feature code vector obtained for each of the local square blocks.

When the feature extraction is not performed by the feature extraction unit, if it is determined that the image is damaged and the resolution is insufficient, the image photographing unit guides a new image acquisition, and if the alignment of the image fails, the image extraction The image alignment unit may control to retry alignment of the image, and may further include a microcomputer that performs control necessary for the overall operation.

According to another aspect of the present invention, obtaining an image of an animal nose by an image capturing unit and inputting relevant information about the animal by an input unit; Aligning the size and angle of the image with the image alignment unit centering on the nose hole and the concave portion of the upper lip of the nose image obtained by the image photographing unit; From the image aligned by the image alignment unit, a feature classification unit extracts a vertical division feature corresponding to a nostril and an outline line of the nose, and a dot pattern and a net pattern are targeted to a defined portion of the upper portion between the nostrils and between the nostrils. Extracting an individual classification feature corresponding to; And registering the feature extracted by the feature extraction unit with external information related to the animal by registering/searching unit to register it as data. Methods are provided.

According to another aspect of the invention, obtaining an image of an animal nose by an image capturing unit and inputting relevant information about the animal by an input unit; Aligning the size and angle of the image with the image alignment unit centering on the nose hole and the concave portion of the upper lip of the nose image obtained by the image photographing unit; From the image aligned by the image alignment unit, a feature classification unit extracts a vertical division feature corresponding to a nostril and an outline line of the nose, and a dot pattern and a net pattern are targeted to a defined portion of the upper portion between the nostrils and between the nostrils. Extracting an individual classification feature corresponding to; Comparing the features extracted by the feature extraction unit with data registered in the external system by the registration/search unit to search for relevant information of the corresponding animal; And displaying related information of the animal searched by the registration/search unit through a display unit. A data search method for identifying and specifying an individual of an animal is provided.

The step of extracting the vertical classification feature and the individual classification feature comprises: extracting a quadrangular area including a nose region from an image aligned by the image alignment unit and normalizing to a predetermined pixel size. ; The nose region image normalized by the image normalization unit is input into three channels of RGB by the vertical division feature vector extraction unit, and is subjected to vertical division by applying inception V2 based DNN using deep learning-based object recognition technology. Extracting a feature vector; Dividing the nose region image into local square blocks of a predetermined pixel size overlapped by the image segmentation unit; Converting a single square image by applying an edge enhancement filter by an image conversion unit to each of the local square blocks divided by the image division unit; Extracting a pattern feature code vector corresponding to an individual classification feature for each of the local square blocks by applying a DNN-based autoencoder to each of the local square blocks by a pattern feature code vector extractor; And constituting the overall feature by connecting the feature pattern vector obtained for each of the vertical feature vector and the local square block by a feature vector connecting unit.

In the step of extracting the longitudinal classification feature and the individual classification feature, if extraction is not performed, if it is determined by the microcomputer that image damage and resolution are insufficient for the image, a new image is acquired by the image photographing unit If the alignment of the image fails, the image alignment unit may be controlled to retry alignment of the image.

According to the animal identification and characterization system and method according to the present invention, by extracting in duplicate the total classification feature and the individual classification feature, it is possible to accurately identify the individual by identifying animals such as companion dogs, and avoid image alignment, etc. It is possible to improve the accuracy of recognition of individual animals even in various postures of the recognized animal, so that it is possible to accurately find the owner of the animal, such as a lost companion dog, so that it is easy and quick to find the owner of the missing dog. In addition, it is possible to construct a database system that enables easy and accurate management of the system based on the companion animal registration system, and can also contribute to the systematic management of the companion animal and to improve animal welfare.

1 is a configuration diagram showing an animal identification and characterization system according to an embodiment of the present invention.
2 is a flowchart illustrating a method for registering data of individual identification and characterization of animals according to another embodiment of the present invention.
3 is a flowchart illustrating a data retrieval method for identification and characterization of an individual of an animal according to another embodiment of the present invention.
4 is a flowchart specifically illustrating a feature extraction step in a data registration method for an individual object identification and characterization of an animal and a data search method for an individual object identification and characterization for an animal according to embodiments of the present invention.
5 is an image showing an image area used in an animal identification and characterization system and method according to an embodiment of the present invention.
6 is an image showing image blocking in an animal identification and characterization system and method according to an embodiment of the present invention.

Since the present invention may have various embodiments by various changes, specific embodiments will be described by way of example in the drawings. In addition, it should be understood that the present invention is not limited to these specific embodiments, and includes all modifications, equivalents, and substitutes included in the technical spirit of the present invention.

Hereinafter, exemplary embodiments according to the present invention will be described in detail with reference to the accompanying drawings, and the same reference numerals are assigned to the same or corresponding components regardless of reference numerals, and redundant descriptions are omitted. Shall be

1 is a configuration diagram showing an animal identification and characterization system according to an embodiment of the present invention.

Referring to FIG. 1, an animal identification and characterization system 100 according to an embodiment of the present invention includes an image photographing unit 110, an input unit 181, an image alignment unit 120, and a feature extraction unit 130. , A registration/search unit 140 and a display unit 150.

The image photographing unit 110 allows the nose image of an animal to be acquired. For example, various types of cameras or camera modules for image acquisition may be used.

The input unit 181 allows information related to animals to be input, and various input devices may be used, including a keyboard, a touch panel, buttons, and switches.

The image alignment unit 120 aligns the size and angle of the image around the nose hole and the concave portion of the upper lip of the nose image obtained by the image photographing unit 110.

The feature extraction unit 130 extracts the vertical segmentation features corresponding to the nostrils and the outlines of the nose from the image aligned by the image alignment unit 120, and the dot pattern is targeted to a predetermined region between the nostrils and the nostrils. Individual classification features corresponding to and net patterns are extracted. The feature extraction unit 130, for example, extracts a quadrangular area including a nose region from an image aligned by the image alignment unit 120, and normalizes it to a predetermined pixel size, as in the present embodiment. And, input the nose region image normalized by the image normalization unit 131 as three channels of RGB, and apply the inception V2-based DNN using deep learning-based object recognition technology to extract the longitudinal feature vector The subdivision feature vector extracting unit 132, the image dividing unit 133 for dividing the nose region image into local square blocks of overlapping predetermined pixel sizes, and the local square block divided by the image dividing unit 133. An image conversion unit 134 that converts to a single channel image by applying an edge enhancement filter to each, and a pattern corresponding to the individual classification feature for each local square block by applying a DNN-based autoencoder to each of the local square blocks. Includes a pattern feature code vector extracting unit (135) for extracting a feature code vector, and a feature vector connecting portion (136) for concatenating the overall features by connecting the pattern feature code vectors obtained for each of the vertical division feature vector and the local square block. Can be.

The registration/search unit 140 transmits the features extracted by the feature extraction unit 130 together with the related information about the animal to the external system 200 to register as data, and extracts the features by the feature extraction unit 130 The compared features are compared with data registered in the external system 200 to search for relevant information of the corresponding animal. Here, the external system 200 is a service that allows access to the terminal through the communication, but is provided with a service for individual identification and characterization of animals to a terminal that has been authenticated by the user by inputting an assigned ID and a set password or other various methods. The server 210 and the extracted features provided by being connected to the service server 210 and related information of the corresponding animal are stored as data, and if necessary, through the service server 210, the systems 100 to terminals of the present invention It may include a database unit 220 to provide to, in addition to this, a system that stores and provides features extracted by various methods and related information of the animal.

The display unit 150 displays data registered by the registration/search unit 140 or related information of the searched animal.

Animal identification and characterization system 100 according to an embodiment of the present invention, in addition to the configuration as described above, including the external system 200 and Wi-Fi, 3G, LTE, 5G. The communication module 160 to perform various types of wired or wireless communication, the microcomputer 170 to perform control necessary for the overall operation, and the power supplied from external power or internal power such as a battery are operated. It may further include a power supply unit 182 to be supplied with power required for the operation, and a memory unit 183 for storing data and programs or applications required for operation.

When the feature extraction by the feature extraction unit 130 is not performed, the microcomputer 170 guides the image taking unit 110 to acquire a new image when it is determined that the image is damaged and the resolution is insufficient, If the alignment of the image fails, the image alignment unit 120 may control to retry alignment of the image.

Animal identification and characterization system 100 according to an embodiment of the present invention is a smart phone, a tablet PC, a notebook, to perform a predetermined operation by the installation and operation of an application or program configured to achieve the object of the present invention, Various information processing and communication devices, including PCs, may be applied, and in order to avoid overlapping descriptions of the above-described configurations (110, 120, 130, 10, 150, 160, 170, 181, 182, 183), the object identification and characterization of animals described later is more detailed in a data registration method, etc. It will be described in detail.

2 is a flowchart illustrating a method for registering data of individual identification and characterization of animals according to another embodiment of the present invention.

Referring to FIG. 2, a method for registering an individual object and identifying an animal according to another embodiment of the present invention, first, acquires the nose image of the animal by the image photographing unit 110 (S11). Related information is input by the input unit 181. Here, the related information on the animal may vary depending on the purpose, for example, individual number, dog name, breed, date of birth, sub-individual number, parent (母) individual number, current status (normal/organic status) Etc., etc., as the owner-related information, such as the owner's name, address, contact information (phone number, e-mail), etc., as input information, may be an input institution/inputtor, contact information, registration date and time.

The nose image obtained by the image photographing unit 110 is aligned by the image alignment unit 120 with respect to the size and angle of the image centering on the nose hole and the concave portion of the upper lip (S12). To this end, the size and angle of the image are aligned with the nose image of the photographed animal, for example, a companion dog, around the nose hole and the concave portion of the upper lip (lower center portion of the nose). For the alignment, various correction methods such as correction to an image corresponding to a front view or correction to an image corresponding to a perspective view at a specific angle may be applied, which is based on the nose of the dog and the dog's nose as the reference point. Arrange images and extract the overall shape and color features.

From the image aligned by the image alignment unit 120, the feature extraction unit 130 extracts the vertical segment feature A corresponding to the nostril and the nose outline line, as shown in FIG. 5, and between the nostrils. An individual classification feature (B) corresponding to a dot pattern and a mesh pattern is extracted from a target portion of the upper part between nostrils (S13).

The step (S13) of extracting the longitudinal classification feature and the individual classification feature, first, as shown in FIG. 4, first, a quadrangular shape including the nose region by the image normalization unit 131 from the image aligned by the image alignment unit 120 The area of is extracted and normalized to a predetermined pixel size (S31). Here, a quadrilateral area that completely includes the nose area can be extracted from the aligned nose image of the dog and normalized to a size of 224x224 pictures.

For the nose region image normalized by the image normalization unit 131, input into the nose region image of RGB 3 channels by the vertical feature vector extraction unit 132, and inception using deep learning-based object recognition technology. V2-based DNN is applied to extract the vector of feature classification (S32). DNN based on Inception v2 uses intrinsic properties of Inception v2 Net, based on the color and structural characteristics of Gaeko, to generate an effective feature vector that can be used for macroscopic discrimination such as dog species. It consists of CNN layers based on Res model. On the other hand, the deep learning-based technology presented in the present invention corresponds to a known technology, and of course, it can be followed as it is, and this is all applied below.

The nose region image is divided into local square blocks of a predetermined pixel size overlapped by the image division unit 133 (S33). Here, the nose region image can be divided into overlapping local square blocks of 64x64 pixels.

An edge enhancement filter is applied by the image conversion unit 134 to each of the local square blocks divided by the image division unit 133 to convert it into a single channel image (S34). Here, an edge enhancement filter is a type of spatial filtering, which is a type of filter that enables to discover and enhance edge details of an image.

DNN-based autoencoder is applied to each of the local square blocks by the pattern feature code vector extractor 135 to extract the pattern feature code vector corresponding to the individual classification feature for each of the local square blocks (S35). The autoencoder used here is pre-trained on the nose pattern pattern of 64x64 pixel size extracted from the gaeko image area.

The feature vector connecting unit 136 connects the vertical feature vector and the pattern feature code vector obtained for each of the local square blocks to form the entire feature (S36). Here, the entire feature is constructed by connecting the feature code vector obtained from every block with the feature vector obtained from the RGB 3-channel image that has been extracted, and this feature compares the data in the database unit 220 with partial matching. By applying it, high similarity can be output even when several partial feature vectors are matched, so that strong performance can be ensured even in partial occlusion, wounds, and impurities. In addition, in this method, in order to extract features from the entire image, induction net type DNN is used, and local area segmentation and DNN-based autoencoder are used together to analyze various types of wrinkles in the nasal region. Can pursue.

The feature extracted by the feature extraction unit 130 is transmitted to the external system 200 together with the related information about the animal by the registration/search unit 140 to register as data (S14). At this time, in the external system 200, an object number (ID) may be assigned to an animal, for example, a companion dog, and object registration may be performed, and all related items are stored as data in the database unit 220. At this time, along with the extracted feature information, the input information (photo image, companion dog information, companion dog owner information, etc.) is registered in the database unit 220.

On the other hand, in the step (S13) of extracting the longitudinal classification feature and the individual classification feature, if extraction is not made, if it is determined by the microcomputer 170 that the image is damaged and the resolution is insufficient, the image photographing unit 110 ) Through the display unit 150 to obtain a new image by text or through an additionally provided speaker (not shown) by voice, and if the alignment of the image fails, the software is imaged by the image alignment unit 120. It can be controlled to retry the alignment of the image (S15).

3 is a flowchart illustrating a data retrieval method for identification and characterization of an individual of an animal according to another embodiment of the present invention.

Referring to FIG. 3, in a data retrieval method for identifying and specifying an individual of an animal according to another embodiment of the present invention, an image of the nose of the animal is acquired by the image photographing unit 110 and input information related to the animal 181 ) (S21), and aligning the size and angle of the image with the image alignment unit around the nose hole and the concave portion of the upper lip of the nose image obtained by the image photographing unit 110 ( S22), and by the feature extraction unit 130 from the image aligned by the image alignment unit 120, extracts the nomenclature features corresponding to the nostrils and the outline of the nose, and the upper part between the nostrils and the nostrils The step of extracting the individual classification features corresponding to the dot pattern and the mesh pattern for the designated area (S23), and the feature extracted by the feature extraction unit 130, the registration/search unit 140 by the external system 200 Comparing with the data registered in the step of searching for relevant information of the corresponding animal (S24), and displaying the relevant information of the animal searched by the registration/search unit 140 through the display unit 150 ( S25). On the other hand, the method for retrieving and identifying an individual of an animal according to another embodiment of the present invention is a data retrieval method from the image acquisition step (S21) to the step of extracting the characteristic (S23), according to another embodiment of the present invention. And, since the specificization was described as the same steps (S11 to S13) in the data registration method, redundant description will be omitted. However, in the step S21 of inputting related information about the animal by the input unit 181, the related information about the animal is search information, and may be a breed, a current state, or the like.

In step S24 of searching for related information of the animal, a feature matching the database search is extracted using a feature or a search word extracted from the image input from the dog's object database unit 220.

In step S26 of displaying the relevant information of the animal through the display unit 150, the retrieved data may be displayed through the display unit 150 provided in a terminal such as a smartphone.

In addition, in this embodiment, the step (S23) of extracting the species classification feature and the individual classification feature, as in the step (S13) corresponding to the data registration method for individual identification and characterization of animals of the previous embodiment, FIG. Referring to the image alignment unit 120, the image normalization unit 131 extracts a quadrangular area including a nose region by the image normalization unit 131 and normalizes the image to a predetermined pixel size (S31). For the nose region image normalized by the unit 132, input into three channels of RGB by the vertical segmentation feature vector extraction unit 132 to apply inception V2-based DNN using deep learning-based object recognition technology Extracting the vertical classification feature vector (S32), dividing the nose region image into local square blocks of a predetermined pixel size overlapped by the image division unit 133 (S33), and an image division unit 133 Step S34 of applying the edge enhancement filter by the image conversion unit 134 to each of the local square blocks divided by) (S34), and pattern feature code vector extraction unit 135 of each of the local square blocks By applying a DNN-based autoencoder (Autoencoder) to extract the pattern feature code vector corresponding to the individual classification feature for each of the local square block (S35), and the vertical division feature vector and the pattern obtained for each of the local square block The feature code vector may be connected by the feature vector connecting unit 136 to configure the entire feature (S36), and the duplicate description will be omitted, as described above.

In addition, in the step (S23) of extracting the longitudinal classification feature and the individual classification feature, if extraction is not performed as in the previous embodiment, if it is determined by the microcomputer 170 that the image is damaged and the resolution is insufficient, , It is possible to guide the image acquisition unit 110 to acquire a new image, and if the alignment of the image fails, the image alignment unit 120 may be controlled to retry alignment of the image.

According to the animal identification and characterization system and method according to the present invention, the pattern of the animal's nose is used to characterize the animal, and the surface image of the animal's nose, such as a companion dog, is as shown in FIG. And the fact that there is a certain dot pattern or net pattern around the concave portion of the upper lip.

To this end, first, the nose image is aligned by using a concave portion from the nose hole and the upper lip. At this time, between the nasal cavity and the triangular part of the upper side, there is a net-like pattern such as a mosaic pattern or a dot-like pattern of protrusions. From this pattern, geometric topological features that can be specified for the individual can be extracted. In addition, the overall shape and color of the nose varies depending on the dog's species or animal species. Using these shapes and colors, the dogs are grouped according to species, and by applying deep learning-based confrontational pattern generation method to every net and dot-like patterns, geometric phase structure features are extracted and compared. Since the pattern characteristics are very different, it is possible to specify an object using it.

As a result of testing to date, when an object is specified using this method, the probability that the same overall shape, color, and pattern will appear for different objects will be equal to one tenth of a million, like fingerprints in humans. With the technology realized so far, about 120 dogs were built and learned, and the test of the performance of cognitive control resulted in the success rate over the data retrieval method of identification and characterization of 95 animals in 1:1 comparison. Is showing.

Although the present invention has been described with reference to the accompanying drawings, various modifications may be made without departing from the technical spirit of the present invention. Therefore, the scope of the present invention should not be limited to the above-described embodiments, but should be defined by the claims, and equivalents to the claims.

110: image photographing unit 120: image alignment unit
130: feature extraction unit 131: image normalization unit
132: vertical division feature vector extraction unit 133: image segmentation unit
134: image conversion unit 135: pattern feature code vector extraction unit
136: feature vector connection unit 140: registration / search unit
150: display unit 160: communication module
170: microcomputer 181: input unit
182: power supply 183: memory
210: service server 220: database

Claims (6)

An image photographing unit acquiring an animal nose image;
An input unit for inputting related information about the animal;
An image aligning unit that aligns the size and angle of the image around the nose hole and the concave portion of the upper lip of the nose image obtained by the image photographing unit;
From the image aligned by the image alignment unit, a vertical classification feature corresponding to a nostril and an outline line of the nose is extracted, and an individual classification feature corresponding to a dot pattern and a mesh pattern is targeted to a predetermined portion of the upper portion between the nostrils and between the nostrils. To extract,
An image normalization unit for extracting a quadrangular area including a nose region from the image aligned by the image alignment unit and normalizing to a predetermined pixel size;
A vertical classification feature that extracts a vertical classification feature vector by applying an inception V2 based DNN using deep learning-based object recognition technology by inputting RGB image with 3 channels to the nose region image normalized by the image normalization unit. Vector extraction unit;
An image dividing unit for dividing the nose region image into local square blocks of overlapping predetermined pixel sizes;
An image conversion unit that converts a local channel into a single channel image by applying an edge enhancement filter to each of the local square blocks divided by the image division unit;
A pattern feature code vector extraction unit for extracting a pattern feature code vector corresponding to an individual classification feature for each of the local square blocks by applying a DNN-based autoencoder to each of the local square blocks; And
A feature extracting unit comprising; a feature vector connecting unit constituting the entire feature by connecting the longitudinal feature vector and the pattern feature code vector obtained for each of the local square blocks;
The feature extracted by the feature extracting unit is transmitted to an external system together with the related information about the animal to be registered as data, and the feature extracted by the feature extracting unit is compared with data registered in the external system. A registration/search unit to search related information of animals; And
A display unit configured to display data registered by the registration/search unit or related information of the searched animal;
System comprising the identification and characterization of the individual of the animal comprising a.
The method according to claim 1,
When the feature extraction is not performed by the feature extraction unit, if it is determined that the image is damaged and the resolution is insufficient, the image photographing unit guides a new image acquisition, and if the alignment of the image fails, the image extraction A system for identifying and specifying an individual of an animal, further comprising a microcomputer that controls re-alignment of the image by the image alignment unit and performs control necessary for the overall operation.
Obtaining an animal nose image by an image capturing unit and inputting related information about the animal by an input unit;
Aligning the size and angle of the image with the image alignment unit centering on the nose hole and the concave portion of the upper lip of the nose image obtained by the image photographing unit;
From the image aligned by the image alignment unit, a feature classification unit extracts a vertical division feature corresponding to a nostril and an outline line of the nose, and a dot pattern and a net pattern are targeted to a defined portion of the upper portion between the nostrils and between the nostrils. Extract the individual classification feature corresponding to,
The step of extracting the species classification feature and the individual classification feature,
Extracting a quadrangular area including a nose region from the image aligned by the image aligning unit by the image normalizing unit, and normalizing to a predetermined pixel size;
The nose region image normalized by the image normalization unit is input into three channels of RGB by the vertical division feature vector extraction unit, and is subjected to vertical division by applying inception V2 based DNN using deep learning-based object recognition technology. Extracting a feature vector;
Dividing the nose region image into local square blocks of a predetermined pixel size overlapped by the image segmentation unit;
Converting a single square image by applying an edge enhancement filter by an image conversion unit to each of the local square blocks divided by the image division unit;
Extracting a pattern feature code vector corresponding to an individual classification feature for each of the local square blocks by applying a DNN-based autoencoder to each of the local square blocks by a pattern feature code vector extractor; And
Comprising the steps of: constituting the overall feature by connecting the feature pattern vector obtained for each of the vertical feature vector and the local square block by a feature vector connecting unit; And
Transmitting a feature extracted by the feature extraction unit to an external system together with related information about the animal by a registration/search unit to register as data;
A method of registering data to identify and characterize an individual of an animal, comprising: The method according to claim 3,
In the step of extracting the longitudinal classification feature and the individual classification feature, if extraction is not performed, if it is determined by the microcomputer that image damage and resolution are insufficient for the image, a new image is acquired by the image photographing unit And controlling to retry alignment of the image by the image alignment unit when alignment of the image fails. Obtaining an animal nose image by an image capturing unit and inputting related information about the animal by an input unit;
Aligning the size and angle of the image with the image alignment unit centering on the nose hole and the concave portion of the upper lip of the nose image obtained by the image photographing unit;
From the image aligned by the image alignment unit, a feature classification unit extracts a vertical division feature corresponding to a nostril and an outline line of the nose, and a dot pattern and a net pattern are targeted to a defined portion of the upper portion between the nostrils and between the nostrils. Extract the individual classification feature corresponding to,
The step of extracting the species classification feature and the individual classification feature,
Extracting a quadrangular area including a nose region from the image aligned by the image aligning unit by the image normalizing unit, and normalizing to a predetermined pixel size;
The nose region image normalized by the image normalization unit is input into three channels of RGB by the vertical division feature vector extraction unit, and is subjected to vertical division by applying inception V2 based DNN using deep learning-based object recognition technology. Extracting a feature vector;
Dividing the nose region image into local square blocks of a predetermined pixel size overlapped by the image segmentation unit;
Converting a single square image by applying an edge enhancement filter by an image conversion unit to each of the local square blocks divided by the image division unit;
Extracting a pattern feature code vector corresponding to an individual classification feature for each of the local square blocks by applying a DNN-based autoencoder to each of the local square blocks by a pattern feature code vector extractor; And
Comprising the steps of: constituting the overall feature by connecting the feature pattern vector obtained for each of the vertical feature vector and the local square block by a feature vector connecting unit;
Comparing the features extracted by the feature extraction unit with data registered in the external system by the registration/search unit to search for relevant information of the corresponding animal; And
Displaying related information of an animal searched by the registration/search unit through a display unit;
A method of retrieving data to identify and characterize an individual of an animal, comprising: The method according to claim 5,
In the step of extracting the longitudinal classification feature and the individual classification feature, if extraction is not performed, if it is determined by the microcomputer that image damage and resolution are insufficient for the image, a new image is acquired by the image photographing unit And controlling to retry alignment of the image by the image alignment unit when alignment of the image fails.

Images