WO2021083381A1

WO2021083381A1 - Animal identity recognition method, apparatus and system

Info

Publication number: WO2021083381A1
Application number: PCT/CN2020/125884
Authority: WO
Inventors: 陆冬云
Original assignee: 北京观海科技发展有限责任公司
Priority date: 2019-11-01
Filing date: 2020-11-02
Publication date: 2021-05-06
Also published as: CN112785620A

Abstract

An animal identity recognition method, apparatus and system. The method comprises: continuously obtain images of a tracked animal object in a tracking session (S110); calculating a temporary feature value set of the tracked animal object on the basis of the images (S120); and when the temporary feature value set meets a first condition, converting the temporary feature value set into a formal feature value set, assigning a unique identifier to the formal feature value set, and determining the unique identifier as the identity identifier of the tracked animal object (S130). This method can be used at any position in a movement area of an animal, does not need to install a recognition device on the body of the animal, is not subject to environmental electromagnetic interference, and has low cost.

Description

Animal identification method, device and system

Technical field

This application relates to the technical field of machine vision and image recognition, and in particular to an animal identification method, device and system.

Background technique

The lean management of livestock farms requires precise identification of each animal, and the uniqueness of each animal is determined by the assigned unique identifier (referred to as ID for short). Traditional animal identification methods usually use physical equipment such as ear tags, collars, and foot rings on the animal to identify the animal’s identity. The physical equipment stores the assigned unique number as the animal’s ID, which is read by a dedicated reading device . For example, in a cattle breeding farm, when specific activities such as feeding, milking, breeding, vaccination, etc. occur, it is necessary to obtain the ID, and use this to record the activity information of the individual cattle.

However, this traditional method of animal identification requires the installation of special equipment on animals and the deployment of special reading equipment at specific locations in the animal activity area. The larger the number of animals, the more equipment needs to be deployed, which poses a high cost to the enterprise. Management cost; in addition, ID storage devices and reading devices can only work at short distances, and cannot provide IDs in the entire activity space of the animals. The reading position is limited, and the reading accuracy is interfered by the environment.

Summary of the invention

The embodiments of the present application provide an animal identification method, device, and system, which are used to solve the problems of high cost, limited use location, and susceptibility to environmental interference in animal identification in the prior art.

In the first aspect, an embodiment of the present application provides an animal identification method, which includes the following steps:

In a tracking session, continuously obtain images of the animal being tracked;

Based on the image, calculating a temporary feature value set of the tracked animal object;

When the temporary feature value set meets the first condition, transform the temporary feature value set into a formal feature value set, assign a unique identifier to the formal feature value set, and determine that the unique identifier is the tracked animal object ’S identity.

In a preferred embodiment, the method further includes:

When the identity of the tracked animal object is not determined, calculate the similarity between the temporary feature value set and all formal feature value sets, and the uniqueness of at least one formal feature value set in all formal feature value sets The probability of being identified as the identity of the tracked animal object;

When the similarity between the temporary feature value set and a formal feature value set in all formal feature value sets satisfies the second condition, it is determined that the unique identifier of the one formal feature value set is the tracked animal object ’S identity.

In a preferred embodiment, in a tracking session, continuously acquiring images of the tracked animal object includes:

Using a multi-target tracking model to identify the location of the tracked animal object from the collected continuous frame images;

Acquiring a tracking frame of the animal object being tracked in consecutive frames of images based on the position;

Extracting multiple images of the tracked animal object from the tracking frame.

In a preferred embodiment, the calculation of the temporary feature value set of the tracked animal object based on the image includes:

In the tracking session, establish an initial temporary feature value set for the tracked animal object;

Continuously calculating the current feature value of the tracked animal object from the multiple images of the tracked animal object;

If the current characteristic value satisfies the third condition, the current characteristic value is added to the initial temporary characteristic value set.

In a preferred embodiment, that the temporary feature value set satisfies the first condition includes: the number of feature values in the temporary feature value set is greater than a first threshold.

In a preferred embodiment, the similarity between the temporary feature value set and one formal feature value set in all formal feature value sets satisfies the second condition includes: the similarity is greater than a second threshold.

In a preferred embodiment, the multi-target tracking model includes a first convolutional neural network model.

In a preferred embodiment, continuously calculating the current feature value of the tracked animal object from the multiple images of the tracked animal object includes: calculating the current feature value of the tracked animal object using a second convolutional neural network model value.

In a preferred embodiment, the current feature value satisfying the third condition includes: the minimum value of the distance between the current feature value and all feature values in the initial temporary feature value set is greater than a third threshold; wherein, the The distance includes the cosine distance.

In a preferred embodiment, the method further includes: if the one tracking session ends, destroying the temporary feature value set, and invalidating the identity of the tracked animal object determined in the one tracking session.

In a preferred embodiment, the image of the tracked animal object includes an image of fur patterns on the surface of the animal's body.

In the second aspect, an embodiment of the present application provides an animal identification device, including:

The tracking module is configured to continuously obtain images of the animal being tracked in a tracking session;

A feature calculation module configured to calculate a temporary feature value set of the tracked animal object based on the image;

The first identification module is configured to convert the temporary feature value set into a formal feature value set when the temporary feature value set satisfies the first condition, assign a unique identifier to the formal feature value set, and determine the unique The identifier is the identity identifier of the tracked animal object.

In a preferred embodiment, the device further includes:

The matching module is configured to calculate the similarity between the temporary feature value set and all formal feature value sets, and at least one of all the formal feature value sets when the identity of the tracked animal object is not determined The unique identifier of the formal feature value set is the probability of the identity identifier of the tracked animal object;

The second identification module is configured to determine the uniqueness of the formal feature value set when the similarity between the temporary feature value set and a formal feature value set in all formal feature value sets satisfies a second condition The identifier is the identity identifier of the tracked animal object.

In a preferred embodiment, the tracking module includes:

The first tracking sub-module is configured to use a multi-target tracking model to identify the position of the tracked animal object from the collected continuous frame images;

The second tracking sub-module is configured to obtain the tracking frame of the tracked animal object in the continuous frame image based on the position;

The image extraction sub-module is configured to extract multiple images of the tracked animal object from the tracking frame.

In a preferred embodiment, the feature calculation module includes:

The first calculation sub-module is configured to establish an initial temporary feature value set for the tracked animal object in the tracking session;

The second calculation sub-module is configured to continuously calculate the current characteristic value of the tracked animal object from the multiple images of the tracked animal object;

The third calculation sub-module is configured to add the current characteristic value to the initial temporary characteristic value set if the current characteristic value satisfies a third condition.

In a preferred embodiment, the second calculation sub-module is configured to use a second convolutional neural network model to calculate the current feature value of the tracked animal object.

In a preferred embodiment, the device further includes: a clearing module configured to destroy the temporary feature value set at the end of the one tracking session, and make the tracked animal object determined in the one tracking session The identity of is invalid.

In the third aspect, an embodiment of the present application provides an animal identification system, including:

An image acquisition unit including at least one image acquisition device;

The processing unit is connected to the image acquisition unit via a network, and is configured to:

In a tracking session, continuously obtain images of the animal being tracked;

In a fourth aspect, an embodiment of the present application further provides a computing device, including a memory and a processor; wherein the memory is used to store at least one computer program, wherein the program is executed by the processor to implement the foregoing implementation manners The steps of the method.

In a fifth aspect, the embodiments of the present application also provide a computer-readable storage medium on which a computer program is stored, and the computer program is executed by a processor to implement the steps of the method described in the foregoing embodiments.

Compared with the prior art, the embodiment of the present application can be used at any position where an animal moves, and there is no need to install an identification device on the animal's body, it is not subject to environmental electromagnetic interference, and the cost is low.

Description of the drawings

The present invention will be more fully understood through the following detailed description in conjunction with the accompanying drawings, in which similar elements are numbered in a similar manner, in which:

Figure 1 is a schematic diagram of the deployment of surveillance cameras in an embodiment of the present application;

Fig. 2 is a schematic flowchart of an animal identification method according to an embodiment of the present application;

FIG. 3 is a schematic diagram of a partial flow of an animal identification method according to another embodiment of the present application;

FIG. 4 is a schematic diagram of a partial flow of an animal identification method according to another embodiment of the present application;

FIG. 5 is a schematic diagram of a partial flow of an animal identification method according to another embodiment of the present application;

Fig. 6 is a schematic structural diagram of a second convolutional neural network model in an embodiment of the present application;

FIG. 7 is a diagram of an application example of cattle tracking and identification in a cattle breeding farm according to an embodiment of the present application;

FIG. 8 is a schematic structural diagram of an animal identification device according to an embodiment of the present application;

FIG. 9 is a schematic diagram of a partial structure of an animal identification device according to another embodiment of the present application;

FIG. 10 is a partial structural diagram of an animal identification device according to another embodiment of the present application;

Fig. 11 is a schematic structural diagram of an animal identification system according to an embodiment of the present application.

Detailed ways

The technical solutions of the present application will be clearly and completely described below through embodiments in conjunction with the drawings, but the present application is not limited to the embodiments described below. Based on the following embodiments, all other embodiments obtained by a person of ordinary skill in the art without creative work shall fall within the protection scope of this application. For the sake of clarity, parts that are not related to the description of the exemplary embodiments are omitted in the drawings.

It should be understood that terms such as "including" or "having" in this application are intended to indicate the existence of the features, numbers, steps, actions, components, or combinations thereof disclosed in this specification, and do not exclude one or more other features , Numbers, steps, behaviors, components or combinations thereof exist or are added possibilities.

As mentioned above, in the prior art, the physical device storing the identification ID of the animal is used to identify the individual animal. This method has high cost, limited use location, easy reading accuracy and environmental interference, etc. problem. In order to overcome the problems of the prior art, this application proposes a method and system for obtaining animal IDs based on multi-target tracking visual recognition technology, using distributed image acquisition devices, such as surveillance cameras or cameras, based on multi-target tracking technology (MOT), you can get the ability to identify the animal's identity at any position of the animal's activity. In this application, an animal identity (ID) refers to a string of characters or numbers assigned to each individual animal and used to identify the uniqueness of the individual animal.

In the following embodiments, in order to facilitate a clear understanding of the technical solutions of the present application, a cattle breeding farm is used as an application scenario to describe the embodiments of the present application.

Fig. 1 is a schematic diagram of the deployment of a surveillance camera in an embodiment of the present application. Taking the deployment of surveillance cameras in a cattle breeding farm as an example, the embodiment of the application can deploy surveillance cameras in a distributed manner within the range of movement of the cattle in the cowshed, and the deployment density and angle must satisfy the range of movement of the cattle without visual blind spots. At the same time, try to avoid the cattle from blocking each other visually. In the example shown in Figure 1, a surveillance camera that can monitor 4 directions at the same time can be deployed on the central axis of the cowshed, and the surveillance camera can achieve 360-degree coverage. The distance between the surveillance cameras on the central axis can be about 6-8 meters, and the distance between the feeding area and the outside of the cow house is between 12-14 meters, which can achieve no visual blind spots in the space from the ground to the height of 2 meters. The embodiment of the present application can collect video images of multiple surveillance cameras, and through the multi-target tracking technology, real-time tracking of any cow can be realized in any area covered by the camera. In the embodiments of the present application, the cattle or other animal individuals being tracked are also referred to as tracked animal objects.

Fig. 2 is a schematic flowchart of an animal identification method according to an embodiment of the present application. As shown in FIG. 2, the animal identification method of the embodiment of the present application includes the following steps:

Step S110, in a tracking session, continuously obtain images of the tracked animal object;

Step S120: Calculate a temporary feature value set of the tracked animal object based on the image;

Step S130, when the temporary feature value set meets the first condition, transform the temporary feature value set into a formal feature value set, assign a unique identifier to the formal feature value set, and determine that the unique identifier is the Track the identity of animal objects.

In the embodiments of the present application, when multi-target tracking of animals is implemented based on surveillance cameras, the continuity of animal tracking is usually destroyed due to the fact that animals are blocked by each other or by the environment, and technically, there are situations in which the tracked animal objects are lost. . Therefore, for each tracked animal object, from the creation of a tracked object to its loss, a tracking session is formed.

In a tracking session, the embodiment of the application can continuously acquire images of the tracked animal object, and calculate the temporary feature value set of the tracked animal object based on the acquired image, and the temporary feature value set is calculated for the same tracked animal object The characteristic value (RID) can be a vector calculated from the image of the tracked animal object as input, and is used to characterize the external characteristics of the tracked animal object. In one embodiment, the image of the tracked animal object includes an image of fur patterns on the surface of the animal's body. The image of the fur pattern on the surface of the animal body has significant individual differences. The characteristic value (RID) calculated by collecting the image of the fur pattern of the individual animal under different illumination, viewing angle and distance has the following characteristics: the same individual under different conditions The similarity of the calculated eigenvalues is high, and the similarity of the eigenvalues calculated by different individuals is low. For example, for individual cows, the image feature values formed by bovine flower pieces such as milk cow flower pieces can be used to distinguish the differences between individual cows. When the temporary feature value set of the tracked animal object meets predetermined conditions and has the ability to distinguish other animal objects, the temporary feature value set is converted into a formal feature value set, and a unique identifier is assigned to it, which can then be used Determine the identity of the animal subject to be tracked. The method can be used at any position where the animal moves, and there is no need to install identification equipment on the animal body, it is not subject to environmental electromagnetic interference, and has low cost.

In an embodiment, the temporary feature value set satisfying the first condition may include that the number of feature values in the temporary feature value set is greater than a predetermined threshold, that is, the temporary feature value set contains enough to characterize the animal object. The eigenvalues of the external features satisfy the conditions for transforming into the formal eigenvalue set of the animal object.

In an implementation manner, as shown in FIG. 3, the embodiment of the present application may further include the following steps after the above step 120:

Step S140: When the identity of the tracked animal object is not determined, calculate the similarity between the temporary feature value set and all formal feature value sets, and at least one formal feature value in all formal feature value sets The unique identifier of the set is the probability of the identity of the tracked animal object;

Step S150: When the similarity between the temporary feature value set and a formal feature value set in all formal feature value sets meets the second condition, determine that the unique identifier of the formal feature value set is the being Track the identity of animal objects.

In this embodiment, on the basis that the formal feature value set of the animal object has been collected, in a tracking session, through the similarity between the temporary feature value set and all the formal feature value sets, it is obtained that a certain tracked animal object belongs to The probability ranking of a certain identity mark, when the similarity satisfies certain conditions, the temporary characteristic value set can be considered to be equivalent to a certain formal characteristic value set, so that the identity mark of the tracked animal object can be determined. This embodiment can further improve the real-time performance of animal identification, and there is no need to sample a large amount of data on the animal object in advance.

In an embodiment, the similarity between the temporary feature value set and a formal feature value set in all formal feature value sets satisfies the second condition may include that the similarity is greater than a predetermined threshold.

In an embodiment, the following method can be used to calculate the similarity between the temporary feature value set and the formal feature value set:

Assuming that the number of elements in the temporary feature value set is m and the number of elements in the formal feature value set is n, the distance D _x between the two feature value sets can be calculated as:

Among them, TRID _i represents a certain feature value in a temporary feature value set, i=1 to m; FRID _p represents each feature value in a formal feature value set, p=1 to n. The meaning of the operator min is to calculate _{the minimum value of the square of the difference between a temporary feature value TRID i} in a temporary feature value set and all feature values in a certain formal feature value set. Here, m is theoretically smaller than n. FRID _j is each remaining feature value in the formal feature value set after the first part of the right side of the above equation is calculated for min, j = 1 ~ nm.

After calculating the distance between the temporary feature value set and the formal feature value set, the similarity can be obtained. The similarity S _x between the temporary feature value set and the formal feature value set is defined as:

Among them, if D _x =0, the similarity is 1, that is, 100% similarity.

In one embodiment, the probability that the unique identification of a certain formal feature value set in all the formal feature value sets is the identity of the tracked animal object can be calculated by the following method, specifically:

Wherein, the value of N is the number of formal feature set; S _i represents the degree of similarity between a feature value N formal formal set of feature values and the temporary collection of feature values, the greater the degree of similarity, which represents a formal feature The greater the probability that the unique identification of the value set is the identification of the tracked animal object.

In one embodiment, if a tracking session ends, the temporary feature value set of the tracked animal object can be destroyed, and the identity of the tracked animal object determined in the one tracking session can be invalidated. The identification also includes a probabilistic identification calculated for the animal being tracked. In the next tracking session, try to track again.

In one embodiment, as shown in FIG. 4, in the above step 110, in a tracking session, continuously acquiring images of the tracked animal object may include:

Step S210, using a multi-target tracking model to identify the position of the tracked animal object from the collected continuous frame images;

Step S220: Obtain a tracking frame of the tracked animal object in the continuous frame image based on the position;

Step S230: Extract multiple images of the tracked animal object from the tracking frame.

Wherein, in an embodiment, the multi-target tracking model may include a first convolutional neural network model. The first convolutional neural network model may be a general convolutional neural network (CNN) model. This embodiment first obtains the first frame image of the tracked animal object, and uses CNN-based detection technology to obtain the initial position of each tracked animal object in the picture; in subsequent frame images, Kalman filter can be used Calculate the position of the tracked animal object in the next frame of image; and in subsequent frame images, it can also be based on the feature extraction of CNN to enhance the prediction of the position of the tracked animal object. Further, the tracking frame that wraps the tracked animal object in each frame of image is obtained through CNN, for example, the smallest possible rectangular frame that wraps the tracked animal object. The rectangular frame does not include other images that do not contain the tracked animal object, such as the background. Wait. Subsequently, based on the identified tracking frame, an image of the tracked animal object can be extracted. This method can improve the ability to retrieve the target after a brief loss, and can reduce tracking loss caused by occlusion between animals.

In some embodiments, as shown in FIG. 5, in the above step 120, the calculation of the temporary feature value set of the tracked animal object based on the image may include:

Step S310, in the tracking session, establish an initial temporary feature value set for the tracked animal object;

Step S320, continuously calculating the current feature value of the tracked animal object from the multiple images of the tracked animal object;

Step S330: If the current characteristic value satisfies the third condition, the current characteristic value is added to the initial temporary characteristic value set.

Among them, for each tracking session, an initial temporary feature value set of the tracked animal object is established. As the tracking of the target continues to obtain the image of the tracked animal object, this embodiment continues to calculate the current feature value (RID) of the image, If the distance between the current eigenvalue and all eigenvalues in the temporary eigenvalue set satisfies a certain condition, that is, there is a certain degree of difference between the current eigenvalue and the existing eigenvalues in the temporary eigenvalue set, then the current eigenvalue is added to the initial Temporary feature value set, otherwise the current feature value is discarded, and the temporary feature value set of the tracked animal object is calculated. In an embodiment, if the minimum value of the distance between the current feature value and all feature values in the temporary feature value set is greater than a predetermined threshold, the current feature value may be added to the initial temporary feature value set.

In an embodiment, the distance between the current feature value and all feature values in the temporary feature value set can be calculated by using the cosine distance or the Euclidean distance between the feature values or a combination of the two. Taking the calculation of cosine distance as an example, it is to calculate the cosine value of the angle between the eigenvalue vectors to measure the distance difference between the eigenvalues. The specific calculation formula is as follows:

Among them, x and y respectively represent two eigenvalue vectors, _xi is an element in the eigenvalue vector x, and y _i is an element in the eigenvalue vector y.

The value range of T(x,y) is [-1,1]. The larger the value, the larger the angle, the farther the eigenvalues are, and the smaller the similarity.

In an implementation manner, in this embodiment, a specially trained second convolutional neural network model may be used to calculate the characteristic value RID of the tracked animal object. The second convolutional neural network model cannot be used for image detection or classification, and is dedicated to calculating the RID of the image. In use, it does not require pre-training of animal objects.

Fig. 6 exemplarily shows a schematic structural diagram of the second convolutional neural network model. As shown in Figure 6, the process of calculating the feature value RID by the second convolutional neural network model is mainly divided into three steps: inputting the original image, CNN convolution, and outputting RID. The second convolutional neural network model structure is similar to the main structure of the standard CNN model. Any CNN model (including but not limited to SSD and YOLO models) that can be used for target detection (including but not limited to SSD and YOLO models) can be applied to the embodiments of this application. Compared with the standard CNN model, the main change of the second convolutional neural network is to remove the last layer of the standard CNN model, namely the fully connected layer (FC), which is the classification layer of the CNN neural network. The output RID of the second convolutional neural network is the input of the original FC layer, which is a one-dimensional vector, and the length of the vector can be arbitrary.

Fig. 7 is a diagram of an application example of cattle tracking and identification in a cattle breeding farm according to an embodiment of the present application. As shown in Figure 7, cows can be tracked in any area covered by the camera. During a tracking session, images of each tracked object are periodically collected. The collected images are partial screenshots in a rectangular frame as small as possible that contain the tracked object, and do not contain other images of the tracked object. Such as background images. The collected partial images of the cattle (C1, C2, C3 in the figure) are input into the above-mentioned second convolutional neural network model to calculate the corresponding feature value RID (RID1, RID2, RID3 in the figure), when the calculation When the RID meets the predetermined conditions, it is stored in a temporary RID set corresponding to the tracked object until the temporary RID set confirms to match a formal RID set, or the temporary RID set meets the conditions for transforming into a formal RID set. Transformed into a formal RID, and was assigned a formal cattle ID. If the tracking of the currently tracked object is lost, the temporary RID set is destroyed.

Fig. 8 is a schematic structural diagram of an animal identification device according to an embodiment of the present application. As shown in FIG. 8, the animal identification device of the embodiment of the present application includes the following components:

The tracking module 410 is configured to continuously obtain images of the tracked animal object in a tracking session;

The feature calculation module 420 is configured to calculate a temporary feature value set of the tracked animal object based on the image;

The first identification module 430 is configured to convert the temporary feature value set into a formal feature value set when the temporary feature value set meets the first condition, assign a unique identifier to the formal feature value set, and determine the The unique identifier is the identity identifier of the tracked animal object.

In an embodiment, the temporary feature value set satisfying the first condition may include that the number of feature values in the temporary feature value set is greater than a predetermined threshold.

In an implementation manner, as shown in FIG. 8, the embodiment of the present application may further include the following components:

The matching module 440 is configured to calculate the similarity between the temporary feature value set and all the formal feature value sets when the identity of the tracked animal object is not determined, and at least among all the formal feature value sets The unique identifier of a formal feature value set is the probability of the identity of the tracked animal object;

The second recognition module 450 is configured to determine the degree of similarity between the temporary feature value set and a formal feature value set in all formal feature value sets when the second condition is satisfied. The unique identifier is the identity identifier of the tracked animal object.

In an embodiment, the calculation of the similarity between the temporary feature value set and the formal feature value set can be calculated using the method described in the foregoing embodiment, which will not be repeated here.

In an embodiment, the probability that the unique identification of a certain formal feature value set in all formal feature value sets is the identity of the tracked animal object can be calculated by the method described in the foregoing embodiment, and will not be repeated here.

In an embodiment, the embodiment of the present application may further include the following components (not shown in the figure): a clearing module configured to destroy the temporary feature value set of the tracked animal object at the end of a tracking session, and make The identity of the tracked animal object determined in the one tracking session becomes invalid. The identification also includes a probabilistic identification calculated for the animal being tracked.

In an embodiment, as shown in FIG. 9, the tracking module 410 may include:

The first tracking sub-module 510 is configured to use a multi-target tracking model to identify the position of the tracked animal object from the collected continuous frame images;

The second tracking sub-module 520 is configured to obtain the tracking frame of the tracked animal object in the continuous frame image based on the position;

The image extraction sub-module 530 is configured to extract multiple images of the tracked animal object from the tracking frame.

Wherein, in an embodiment, the multi-target tracking model may include a first convolutional neural network model. The first convolutional neural network model may be a general convolutional neural network (CNN) model. This embodiment first obtains the first frame image of the tracked animal object, and uses CNN-based detection technology to obtain the initial position of each tracked animal object in the picture; in subsequent frame images, Kalman filter can be used Calculate the position of the tracked animal object in the next frame of image; and in subsequent frame images, it can also be based on the feature extraction of CNN to enhance the prediction of the position of the tracked animal object. Further, the tracking frame that wraps the tracked animal object in each frame of the image is obtained through CNN, for example, the smallest possible rectangular frame that wraps the tracked animal object. The rectangular frame does not include other images that do not contain the tracked animal object, such as the background. Wait. Subsequently, based on the identified tracking frame, an image of the tracked animal object can be extracted. This method can improve the ability to retrieve the target after a brief loss, and can reduce tracking loss caused by occlusion between animals.

In some embodiments, as shown in FIG. 10, the feature calculation module 420 may include:

The first calculation sub-module 610 is configured to establish an initial temporary feature value set for the tracked animal object in the tracking session;

The second calculation sub-module 620 is configured to continuously calculate the current characteristic value of the tracked animal object from the multiple images of the tracked animal object;

The third calculation sub-module 630 is configured to add the current characteristic value to the initial temporary characteristic value set if the current characteristic value satisfies a third condition.

In an embodiment, the distance between the current feature value and all feature values in the temporary feature value set can be calculated by using the cosine distance or the Euclidean distance between the feature values or a combination of the two. Taking the calculation of the cosine distance as an example, it is to calculate the cosine value of the angle between the eigenvalue vectors to measure the distance difference between the eigenvalues. The specific calculation formula is as described above and will not be repeated here.

Fig. 11 is a schematic structural diagram of an animal identification system according to an embodiment of the present application. As shown in FIG. 11, the animal identification system of the embodiment of the present application may include the following units:

An image acquisition unit 710 including at least one image acquisition device;

The processing unit 720 is connected to the image acquisition unit 710 via a network, and is configured to: continuously acquire images of the tracked animal object in a tracking session; and calculate the temporary characteristics of the tracked animal object based on the image Value set; when the temporary feature value set meets the first condition, the temporary feature value set is transformed into a formal feature value set, a unique identifier is assigned to the formal feature value set, and the unique identifier is determined to be the Track the identity of animal objects.

Among them, the image acquisition device may include a camera or a camera, which is distributed in the space of the animal activity place. The processing unit 720 may be a server or other computing processing devices connected to the image capture device through a network.

In an embodiment, the processing unit 720 is further configured to:

In one embodiment, the processing unit 720 is further configured to: at the end of a tracking session, destroy the temporary feature value set of the tracked animal object, and invalidate the identity of the tracked animal object determined in the one tracking session . The identification also includes a probabilistic identification calculated for the animal being tracked.

In one embodiment, the processing unit 720 is further configured to: use a multi-target tracking model to identify the position of the tracked animal object from the acquired continuous frame images; and obtain information about the tracked animal object in the continuous frame image based on the position. Tracking frame; extracting multiple images of the tracked animal object from the tracking frame.

In an embodiment, the multi-target tracking model may include a first convolutional neural network model. The first convolutional neural network model may be a general convolutional neural network (CNN) model.

In some embodiments, the processing unit 720 is further configured to: in the tracking session, establish an initial temporary feature value set for the tracked animal object; and continuously calculate the tracked animal object from multiple images of the tracked animal object. The current feature value of the tracked animal object; if the current feature value satisfies the third condition, the current feature value is added to the initial temporary feature value set.

In an embodiment, the current feature value satisfying the third condition includes that the minimum value of the distance between the current feature value and all feature values in the temporary feature value set is greater than a predetermined threshold.

In an embodiment, the distance between the current feature value and all feature values in the temporary feature value set can be calculated by using the cosine distance or the Euclidean distance between the feature values or a combination of the two. Taking the calculation of the cosine distance as an example, the specific calculation formula is as described above, and will not be repeated here.

The steps, units, or modules involved in the embodiments of the present application can be implemented by software, hardware, or a combination thereof. The described steps, units, or modules may also be provided in the processor of the computing device, where the name of the unit or module does not constitute a limitation on the unit or module itself.

The method described in the embodiments of the present application can be implemented as a computer software program. For example, an embodiment of the present application may include a computer program product, which includes a readable storage medium storing one or more computer programs, and the computer program includes program code for executing the method described in the present application. On the other hand, the embodiments of the present application may also include a computer-readable storage medium that stores one or more programs, and the one or more programs are executed by one or more processors. When executed, the method described in this application can be implemented.

The methods and devices described in this application can be implemented by computing devices such as personal computers and servers. The computing devices usually include a processor for executing various programs and a memory for storing programs, where the programs are loaded into the processor and run. The method described in this application can be implemented at this time.

The implementation of the application is not limited to the above-mentioned embodiments. Without departing from the spirit and scope of the application, a person of ordinary skill in the art can make various changes and improvements in the form and details of the application, all of which are It is considered to fall into the scope of protection of this application.

Claims

An animal identification method, characterized in that it comprises the following steps:

In a tracking session, continuously obtain images of the animal being tracked;

Based on the image, calculating a temporary feature value set of the tracked animal object;

When the temporary feature value set meets the first condition, transform the temporary feature value set into a formal feature value set, assign a unique identifier to the formal feature value set, and determine that the unique identifier is the tracked animal object ’S identity.
The animal identification method according to claim 1, characterized in that it further comprises:

When the identity of the tracked animal object is not determined, calculate the similarity between the temporary feature value set and all formal feature value sets, and the uniqueness of at least one formal feature value set in all formal feature value sets The probability of being identified as the identity of the tracked animal object;

When the similarity between the temporary feature value set and a formal feature value set in all formal feature value sets satisfies the second condition, it is determined that the unique identifier of the one formal feature value set is the tracked animal object ’S identity.
The animal identification method according to claim 1, characterized in that, in a tracking session, continuously acquiring images of the tracked animal object comprises:

Using a multi-target tracking model to identify the location of the tracked animal object from the collected continuous frame images;

Acquiring a tracking frame of the animal object being tracked in consecutive frames of images based on the position;

Extracting multiple images of the tracked animal object from the tracking frame.
The animal identification method according to claim 3, wherein the calculation of the temporary feature value set of the tracked animal object based on the image comprises:

In the tracking session, establish an initial temporary feature value set for the tracked animal object;

Continuously calculating the current feature value of the tracked animal object from the multiple images of the tracked animal object;

If the current characteristic value satisfies the third condition, the current characteristic value is added to the initial temporary characteristic value set.
The animal identification method according to claim 1, wherein the temporary feature value set satisfying the first condition comprises: the number of feature values in the temporary feature value set is greater than a first threshold.
The animal identification method according to claim 2, wherein the similarity between the temporary feature value set and one formal feature value set in all formal feature value sets satisfies the second condition comprises: the similarity The degree is greater than the second threshold.
The animal identification method according to claim 3, wherein the multi-target tracking model comprises a first convolutional neural network model.
The animal identification method according to claim 4, wherein continuously calculating the current characteristic value of the tracked animal object from the multiple images of the tracked animal object comprises: calculating by using a second convolutional neural network model The current feature value of the tracked animal object.
The animal identification method according to claim 4, wherein the current feature value satisfying the third condition comprises: the minimum value of the distance between the current feature value and all feature values in the initial temporary feature value set Greater than the third threshold; wherein, the distance includes a cosine distance.
The animal identification method according to claim 1, further comprising: at the end of the one tracking session, destroying the temporary feature value set, and making the tracked animal object determined in the one tracking session The identity of is invalid.
The animal identification method according to claim 1, wherein the image of the tracked animal object comprises an image of fur patterns on the surface of the animal's body.
An animal identification device, characterized in that it comprises:

The tracking module is configured to continuously obtain images of the animal being tracked in a tracking session;

A feature calculation module configured to calculate a temporary feature value set of the tracked animal object based on the image;

The first identification module is configured to convert the temporary feature value set into a formal feature value set when the temporary feature value set satisfies the first condition, assign a unique identifier to the formal feature value set, and determine the unique The identifier is the identity identifier of the tracked animal object.
The animal identification device according to claim 12, further comprising:

The matching module is configured to calculate the similarity between the temporary feature value set and all formal feature value sets, and at least one of all the formal feature value sets when the identity of the tracked animal object is not determined The unique identifier of the formal feature value set is the probability of the identity identifier of the tracked animal object;

The second identification module is configured to determine the uniqueness of the formal feature value set when the similarity between the temporary feature value set and a formal feature value set in all formal feature value sets satisfies a second condition The identifier is the identity identifier of the tracked animal object.
The animal identification device according to claim 12, wherein the tracking module comprises:

The first tracking sub-module is configured to use a multi-target tracking model to identify the position of the tracked animal object from the collected continuous frame images;

The second tracking sub-module is configured to obtain the tracking frame of the tracked animal object in the continuous frame image based on the position;

The image extraction sub-module is configured to extract multiple images of the tracked animal object from the tracking frame.
The animal identification device according to claim 14, wherein the characteristic calculation module comprises:

The first calculation sub-module is configured to establish an initial temporary feature value set for the tracked animal object in the tracking session;

The second calculation sub-module is configured to continuously calculate the current characteristic value of the tracked animal object from the multiple images of the tracked animal object;

The third calculation sub-module is configured to add the current characteristic value to the initial temporary characteristic value set if the current characteristic value satisfies a third condition.
The animal identification device according to claim 12, wherein the temporary feature value set satisfying the first condition comprises: the number of feature values in the temporary feature value set is greater than a first threshold.
The animal identification device according to claim 13, wherein the similarity between the temporary feature value set and one formal feature value set in all formal feature value sets satisfies the second condition comprises: the similarity The degree is greater than the second threshold.
The animal identification device according to claim 14, wherein the multi-target tracking model comprises a first convolutional neural network model.
The animal identification device according to claim 15, wherein the second calculation sub-module is configured to use a second convolutional neural network model to calculate the current feature value of the tracked animal object.
The animal identification device according to claim 15, wherein the current feature value satisfying the third condition comprises: the minimum value of the distance between the current feature value and all feature values in the initial temporary feature value set Greater than the third threshold; wherein, the distance includes a cosine distance.
The animal identification device according to claim 12, further comprising: a clearing module configured to destroy the temporary feature value set when the one tracking session ends, and make the The identity of the identified animal subject to be tracked is invalid.
The animal identification device according to claim 12, wherein the image of the tracked animal object comprises an image of fur patterns on the surface of the animal's body.
An animal identification system, which is characterized in that it comprises:

An image acquisition unit including at least one image acquisition device;

The processing unit is connected to the image acquisition unit via a network, and is configured to:

In a tracking session, continuously obtain images of the animal being tracked;

Based on the image, calculating a temporary feature value set of the tracked animal object;

When the temporary feature value set meets the first condition, transform the temporary feature value set into a formal feature value set, assign a unique identifier to the formal feature value set, and determine that the unique identifier is the tracked animal object ’S identity.
A computing device, which is characterized by comprising a memory and a processor; wherein,

The memory is used to store at least one computer program, wherein the computer program is executed by the processor to implement the steps of the method according to any one of claims 1-11.
A computer-readable storage medium, characterized in that a computer program is stored thereon, and the computer program is executed by a processor to implement the steps of the method according to any one of claims 1-11.