Image analysis system, image analysis method, and control program

Abstract

To provide an image analysis system, an image analysis method, and a control program that can detect a livestock area included in a photographed image with relatively high accuracy.SOLUTION: An image analysis system includes a camera and a computer. The camera generates video data by photographing livestock. The video data includes a plurality of frames. The plurality of frames include a first frame and a second frame that is generated temporally later than the first frame. The computer inputs the second frame into the trained model to extract a plurality of regions, each of which is a candidate region that includes the livestock. The computer selects, on the basis of an optical flow generated on the basis of on the first and second frames, one of the plurality of regions as a region including the livestock.SELECTED DRAWING: Figure 6

Description

[Technical field]

[0001] The present invention relates to an image analysis system, an image analysis method, and a control program. [Background technology]

[0002] Japanese Patent Application Laid-Open No. 2022-122735 (Patent Document 1) discloses a behavior tracking system that tracks the behavior of individual animals. In this behavior tracking system, a frame surrounding an individual animal included in a captured image is detected (see Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2022-122735 A Summary of the Invention [Problem to be solved by the invention]

[0004] An object of the present invention is to provide an image analysis system, an image analysis method, and a control program that are capable of detecting livestock areas included in a captured image with relatively high accuracy. [Means for solving the problem]

[0005] An image analysis system according to an aspect of the present invention includes a camera and a computer. The camera generates video data by photographing livestock. The video data includes a plurality of frames. The plurality of frames includes a first frame and a second frame that is generated temporally later than the first frame. The computer inputs the second frame to a trained model to extract a plurality of regions, each of which is a candidate region including livestock. The computer selects one of the plurality of regions as a region including livestock, based on an optical flow generated based on the first and second frames.

[0006] According to this image analysis system, optical flow is referenced when selecting an area including livestock, so that an area including livestock can be selected from a plurality of areas with relatively high accuracy.

[0007] In the above-mentioned image analysis system, when selecting one of a plurality of regions as an area containing livestock, the computer may use different algorithms when the speed of the livestock indicated by the optical flow is equal to or greater than a predetermined speed and when the speed is less than the predetermined speed.

[0008] According to this image analysis system, when selecting an area containing livestock, an appropriate algorithm is used depending on the speed of the livestock (e.g., whether the livestock are moving or not), so that an area containing livestock can be selected from multiple areas with relatively high accuracy.

[0009] In the above image analysis system, when the speed is equal to or greater than a predetermined speed, the computer may select one of the multiple regions as the region containing livestock by referring to the position of the livestock predicted based on the optical flow, while when the speed is less than the predetermined speed, the computer may select one of the multiple regions as the region containing livestock by referring to the region containing livestock in the first frame.

[0010] According to this image analysis system, when selecting an area containing livestock, an appropriate target according to the speed of the livestock is referenced, so that an area containing livestock can be selected from multiple areas with relatively high accuracy.

[0011] In the above image analysis system, when the speed is equal to or greater than a predetermined speed, the computer may select, among the multiple regions, the region that has the greatest degree of overlap with the position of the livestock predicted based on the optical flow as the region containing the livestock, whereas when the speed is less than the predetermined speed, the computer may select, among the multiple regions, the region that has the greatest degree of overlap with the region containing the livestock in the first frame as the region containing the livestock.

[0012] According to this image analysis system, when selecting an area containing livestock, an appropriate target is referenced depending on the speed of the livestock (e.g., whether the livestock are moving or not), so that an area containing livestock can be selected from multiple areas with relatively high accuracy.

[0013] In the image analysis system, the computer may detect at least one of the amount of movement, posture, and tail-lifted state of the livestock based on image data of an area including the livestock that is continuously selected for each frame.

[0014] In the image analysis system, the computer may detect signs of parturition in livestock based on at least one of an amount of movement, posture, and tail-raising state of the livestock.

[0015] An image analysis method according to another aspect of the present invention includes a step of generating moving image data by photographing livestock, the moving image data including a plurality of frames, the plurality of frames including a first frame and a second frame generated temporally later than the first frame, and further includes a step of extracting a plurality of regions, each of which is a candidate region containing livestock, by inputting the second frame into a trained model, and a step of selecting one of the plurality of regions as the region containing livestock based on an optical flow generated based on the first and second frames.

[0016] According to this image analysis method, optical flow is referenced when selecting an area including livestock, so that an area including livestock can be selected from a plurality of areas with relatively high accuracy.

[0017] A control program according to another aspect of the present invention causes a computer to generate moving image data by photographing livestock, the moving image data including a plurality of frames, the plurality of frames including a first frame and a second frame generated temporally later than the first frame, and further causes the computer to extract a plurality of areas, each of which is a candidate area for an area containing livestock, by inputting the second frame into a trained model, and to select one of the plurality of areas as the area containing livestock based on an optical flow generated based on the first and second frames.

[0018] According to this control program, optical flow is referenced when selecting an area including livestock, so that an area including livestock can be selected from a plurality of areas with relatively high accuracy. Effect of the Invention

[0019] According to the present invention, it is possible to provide an image analysis system, an image analysis method, and a control program that are capable of detecting livestock areas included in a captured image with relatively high accuracy. [Brief description of the drawings]

[0020] [Figure 1] FIG. 1 is a diagram illustrating a configuration of an image analysis system. [Diagram 2] FIG. 2 illustrates an example of a hardware configuration of a server. [Diagram 3] FIG. 2 is a diagram showing a configuration of software modules realized in the server. [Figure 4] FIG. 13 is a diagram for explaining an algorithm used when livestock are not moving. [Diagram 5] FIG. 13 is a diagram for explaining an algorithm used when livestock are moving. [Figure 6] 13 is a flowchart showing a procedure for detecting a livestock area. [Figure 7] 13 is a flowchart showing a procedure for a first selection process. [Figure 8]13 is a flowchart showing a procedure for a second selection process. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0021] Hereinafter, an embodiment according to one aspect of the present invention (hereinafter also referred to as "the present embodiment") will be described in detail with reference to the drawings. Note that the same or corresponding parts in the drawings are given the same reference numerals and their description will not be repeated. Also, each drawing is drawn in a schematic manner with objects appropriately omitted or exaggerated for ease of understanding.

[0022] [1. Overview] It is known that the probability of a livestock birth accident occurring can be significantly reduced if a human is present during livestock (e.g., a cow) birth. In order for a human to be present during birth, a human needs to wait in the birthing pen before the livestock begins to birth. If the accuracy of detecting signs of livestock birth is improved, the labor required to monitor birth can be reduced, for example, because a human only needs to wait in the birthing pen after the signs of birth are detected. In the image analysis system 10 according to the present embodiment, signs of livestock birth are detected with a relatively high degree of accuracy.

[0023] 1 is a diagram illustrating a schematic configuration of an image analysis system 10. As illustrated in FIG. 1, the image analysis system 10 includes a camera 100 and a server 200.

[0024] The camera 100 is installed in a barn 300 that houses livestock (e.g., cows) 400. The camera 100 includes an imaging element such as a complementary metal-oxide-semiconductor (CMOS) image sensor or a charge coupled device (CCD) image sensor. The camera 100 is configured to continuously capture images of the livestock 400 in the barn 300 and continuously generate video data. The frame rate of the video data generated by the camera 100 is, for example, 60 fps.

[0025] A plurality of farrowing stalls are provided in the cowshed 300, and each farrowing stall houses a maximum of one livestock 400. Adjacent farrowing stalls are separated by, for example, a fence. For example, a plurality of cameras 100 are installed in the cowshed 300. The plurality of cameras 100 are associated with different farrowing stalls. Each camera 100 photographs the livestock 400 in the associated farrowing stall. That is, each camera 100 is installed in a position where the associated farrowing stall is within the shooting range of the camera 100. Each camera 100 may be installed, for example, near any of the four corners of the associated farrowing stall. The height position of each camera 100 may be, for example, a position where the livestock 400 in the associated farrowing stall can be photographed from diagonally above. Furthermore, the orientation of each camera 100 may be, for example, such that the approximate center of the angle of view of the camera 100 is the center of the associated farrowing stall.

[0026] The video data generated by the camera 100 is continuously transmitted to the server 200 via the network N1. That is, the camera 100 has a communication function, and video streaming data is transmitted from the camera 100 to the server 200. The server 200 judges the birth signs of the livestock 400 based on the received video data.

[0027] In order to determine the birth sign of the livestock 400, the server 200 detects the area of ​​the livestock 400 in each frame of the video data. The server 200 detects the birth sign based on at least one of the movement amount, posture, and tail-raised state of the livestock 400. For example, the birth sign may be detected when at least one of the following is satisfied: the movement amount is equal to or more than a predetermined amount, the number of posture changes (e.g., the number of times the livestock stands up and sits down) is equal to or more than a predetermined number, and the tail-raised state continues for a predetermined period of time or more. In order to detect the movement amount, posture, tail-raised state, etc. of the livestock 400 with high accuracy, the area of ​​the livestock 400 needs to be detected with high accuracy in each frame of the video data. The image analysis system 10 according to the present embodiment is designed to be able to detect the area of ​​the livestock 400 with high accuracy in each frame of the video data. The configuration of the image analysis system 10 will be described below in order.

[0028] [2. Configuration] <2-1. Hardware configuration> Fig. 2 is a diagram showing an example of a hardware configuration of server 200. Server 200 is realized by, for example, a general-purpose computer. As shown in Fig. 2, server 200 includes a control unit 210, a storage unit 220, a communication module 230, and an input / output I / F 240. The components included in server 200 are electrically connected via a bus.

[0029] The control unit 210 includes a central processing unit (CPU) 211, a random access memory (RAM) 212, a read only memory (ROM) 213, etc., and is configured to control each of the components in accordance with information processing.

[0030] The storage unit 220 is configured with, for example, an auxiliary storage device such as a hard disk drive or a solid state drive. The storage unit 220 stores, for example, a control program 221, a first trained model 222, a second trained model 223, and a third trained model 224. The control program 221 is a control program of the server 200, and is executed by the control unit 210.

[0031] The first trained model 222 is configured to extract candidates for areas including the livestock 400 in the frame (hereinafter also referred to as "livestock area candidates") in response to input of a frame included in the video image data generated by the camera 100. The first trained model 222 outputs multiple livestock area candidates for one frame, for example. A fence separating a farrowing pen and the livestock 400 in an adjacent farrowing pen may enter the input frame, and for example, an area including them may be extracted as a livestock area candidate, so multiple livestock area candidates are output. The first trained model 222 is generated in advance through machine learning using a large amount of image data (teacher data) each including an area of ​​the livestock 400. Note that various known methods such as deep learning, neural network, decision tree learning, association rule learning, and Bayesian network can be applied as the machine learning.

[0032] The second trained model 223 is configured to output information (hereinafter also referred to as "posture information") regarding the posture of the livestock 400 in response to input of image data of an area including the livestock 400 (hereinafter also referred to as "livestock area"). The second trained model 223 is generated in advance through machine learning using a large amount of image data (teacher data) including an area of ​​the livestock 400 each in a sitting state, and a large amount of image data (teacher data) including an area of ​​the livestock 400 each in a standing state. Note that various known methods such as deep learning, neural networks, decision tree learning, association rule learning, and Bayesian networks can be applied as the machine learning.

[0033] The third trained model 224 is configured to output information on the state of the tail of the livestock 400 (hereinafter also referred to as "tail state information") in response to input of image data of the livestock area. The tail state information indicates, for example, whether the tail of the livestock 400 is erected or not. The third trained model 224 is generated in advance through machine learning using a large amount of image data (teacher data) including areas of the livestock 400 each with its tail erected, and a large amount of image data (teacher data) including areas of the livestock 400 each with its tail not erected. Note that various known methods such as deep learning, neural networks, decision tree learning, association rule learning, and Bayesian networks can be applied as machine learning.

[0034] The communication module 230 is configured to communicate with an external device, for example, to communicate with the camera 100. The communication module 230 is configured with, for example, a wireless LAN (Local Area Network) module, a wired LAN module, or a communication module conforming to other communication standards such as LTE (Long Term Evolution).

[0035] The input / output I / F 240 is an input / output device in the server 200. The input / output I / F 240 is composed of, for example, a keyboard, a mouse, and a display.

[0036] <2-2. Software configuration> Fig. 3 is a diagram showing a configuration of software modules realized in the server 200. Referring to Fig. 3, in the server 200, the control unit 210 executes the control program 221 to realize an acquisition unit 250, a livestock area detection unit 252, a movement amount calculation unit 254, a delivery sign detection unit 256, and a notification unit 258.

[0037] The acquisition unit 250 is configured to continuously acquire video data generated by the camera 100. Each frame contained in the video data includes an image of a livestock 400 in a farrowing pen.

[0038] The livestock area detection unit 252 is configured to detect a livestock area of ​​each frame included in the video data acquired by the acquisition unit 250. The livestock area detection unit 252 extracts a plurality of livestock area candidates by inputting a frame to the first trained model 222. The livestock area detection unit 252 calculates an optical flow based on a frame (second frame) input to the first trained model 222 and a frame (first frame) generated one frame before the frame input to the first trained model 222. The calculated optical flow indicates a distribution of apparent speeds of objects (e.g., livestock 400) in the video data. The second frame can also be said to be a frame generated one frame later in time than the first frame. The livestock area detection unit 252 determines whether the livestock 400 is moving based on the calculated optical flow. The livestock area detection unit 252 selects one of the extracted livestock area candidates as a livestock area based on the determination result. Thereby, the livestock area included in the second frame is detected. The livestock region of the first frame in the entire video data may be detected according to a completely different algorithm. For example, the livestock region of the first frame in the entire video data may be detected by referring to the likelihood information of each livestock region candidate output by the first trained model 222.

[0039] In the image analysis system 10 according to the present embodiment, when selecting one of a plurality of livestock region candidates as a livestock region, different algorithms are used when the speed of the livestock 400 indicated by the optical flow is equal to or greater than a predetermined speed and when the speed of the livestock 400 is less than the predetermined speed. More specifically, in the image analysis system 10, when selecting one of a plurality of livestock region candidates as a livestock region, different algorithms are used when the livestock 400 is moving and when the livestock 400 is not moving.

[0040] Fig. 4 is a diagram for explaining an algorithm used when the livestock 400 is not moving. Referring to the upper diagram included in Fig. 4, in this example, it is determined that the livestock 400 is not moving based on an optical flow calculated using a frame F1 and a frame generated immediately before the frame F1 (hereinafter also referred to as the "previous frame"). In this example, areas T1, T2, T3, T4, and T5 are extracted as livestock area candidates.

[0041] Referring to the central diagram included in FIG. 4, the livestock region detection unit 252 (FIG. 3) deletes regions that do not overlap with the livestock region of the previous frame from the livestock region candidates. In this example, the regions T1 and T2 do not overlap with the livestock region of the previous frame. Therefore, the livestock region detection unit 252 deletes the regions T1 and T2 from the livestock region candidates.

[0042] Referring to the lower diagram in FIG. 4, the livestock region detection unit 252 selects the region that has the greatest degree of overlap with the livestock region of the previous frame as the livestock region among the remaining livestock region candidates. In this example, the region T5 has the greatest degree of overlap with the livestock region of the previous frame. Therefore, the livestock region detection unit 252 selects the region T5 as the livestock region.

[0043] FIG. 5 is a diagram for explaining an algorithm used when the livestock 400 is moving. Referring to the upper diagram included in FIG. 5, in this example, it is determined that the livestock 400 is moving based on an optical flow calculated using the frame F2 and a frame (previous frame) generated immediately before the frame F2. In addition, the livestock region detection unit 252 estimates the livestock region in the frame F2 based on the previous frame and the optical flow. Hereinafter, the estimated livestock region is also referred to as an "estimated livestock region." In this example, regions T6, T7, T8, T9, and T10 are extracted as livestock region candidates. Note that, since the frame rate of the moving image data generated by the camera 100 as described above is, for example, 60 fps, the position of the livestock 400 does not change so much between adjacent frames even if the livestock 400 is moving. The frame rate of the moving image data generated by the camera 100 does not necessarily have to be 60 fps, but may be 50 fps, 30 fps, 25 fps, 24 fps, 3 fps, or 1 fps. The frame rate of the moving image data generated by the camera 100 need not be extremely low, such as 1 / 60 fps.

[0044] Referring to the central diagram included in FIG. 5, the livestock region detection unit 252 (FIG. 3) deletes regions that do not overlap with the estimated livestock region from the livestock region candidate. In this example, the regions T6 and T7 do not overlap with the estimated livestock region. Therefore, the livestock region detection unit 252 deletes the regions T6 and T7 from the livestock region candidate.

[0045] Referring to the lower diagram in FIG. 5, the livestock region detection unit 252 selects the region that has the greatest degree of overlap with the estimated livestock region from among the remaining livestock region candidates as the livestock region. In this example, the region T10 has the greatest degree of overlap with the estimated livestock region. Therefore, the livestock region detection unit 252 selects the region T10 as the livestock region.

[0046] 3 again, the livestock area detection unit 252 inputs the image data input from the acquisition unit 250 and information indicating the livestock area to the movement amount calculation unit 254, and inputs the image data of the livestock area to the second trained model 223 and the third trained model 224. In the second trained model 223, posture information is generated, and in the third trained model 224, tail state information is generated.

[0047] The movement amount calculation unit 254 is configured to calculate the amount of movement of the livestock 400 based on the information input from the livestock area detection unit 252. The movement amount calculation unit 254 is configured to accumulate the amount of movement of the livestock 400 by continuously analyzing the video data.

[0048] The birth sign detection unit 256 is configured to detect a birth sign of the livestock 400 based on the movement amount input from the movement amount calculation unit 254, the posture information input from the second trained model 223, and the tail state information input from the third trained model 224. The birth sign detection unit 256 detects a birth sign of the livestock 400, for example, when the movement amount, posture information, and tail state information of the livestock 400 satisfy a predetermined condition. The birth sign detection unit 256 may detect a birth sign, for example, when at least one of the following is satisfied: the movement amount of the livestock 400 is equal to or greater than a predetermined amount, the number of changes in posture of the livestock 400 (for example, the number of times the livestock 400 stands or sits) is equal to or greater than a predetermined number, and the livestock 400 keeps its tail raised for a predetermined period of time or more.

[0049] The notification unit 258 is configured to notify the user when a birth sign is detected in the livestock 400. For example, the notification unit 258 sends an email to a communication terminal owned by the user when a birth sign is detected in the livestock 400.

[0050] [3. Livestock area detection operation] 6 is a flowchart showing a procedure for detecting a livestock area. The process shown in this flowchart is repeatedly executed by the control unit 210 of the server 200 when video image data is continuously received from the camera 100.

[0051] 6, the control unit 210 acquires information about a plurality of livestock region candidates by inputting a frame included in the video data received from the camera 100 to the first trained model 222 (step S100). The control unit 210 calculates an optical flow based on the frame (second frame) input to the first trained model 222 and the frame (first frame) generated immediately before the frame input to the first trained model 222 (step S110).

[0052] The control unit 210 determines whether the speed of the livestock 400 is equal to or greater than a predetermined speed (whether the livestock 400 is moving) based on the calculated optical flow (step S120). If the control unit 210 determines that the speed of the livestock 400 is equal to or greater than the predetermined speed (YES in step S120), it executes a first selection process (step S130), and if the control unit 210 determines that the speed of the livestock 400 is less than the predetermined speed (NO in step S120), it executes a second selection process (step S140).

[0053] Fig. 7 is a flow chart showing the procedure of the first selection process. Referring to Fig. 7, the control unit 210 estimates the livestock region in the second frame based on the livestock region and the optical flow in the first frame (step S200). The control unit 210 deletes livestock region candidates that do not overlap with the estimated livestock region from among the multiple livestock region candidates acquired in step S100 (step S210). The control unit 210 selects the livestock region candidate that has the greatest degree of overlap with the estimated livestock region from among the remaining livestock region candidates as the livestock region (step S220).

[0054] Fig. 8 is a flow chart showing the procedure of the second selection process. Referring to Fig. 8, the control unit 210 deletes livestock region candidates that do not overlap with the livestock region of the previous frame from among the livestock region candidates acquired in step S100 (step S300). The control unit 210 selects the livestock region candidate that has the greatest degree of overlap with the livestock region of the previous frame as the livestock region from among the remaining livestock region candidates (step S310).

[0055] 6, when the livestock area is selected in step S130 or step S140, the control unit 210 outputs image data within the selected livestock area (step S150). The control unit 210 detects a birth sign of the livestock 400 by using the output image data.

[0056] [4. Features] As described above, in the image analysis system 10 according to the present embodiment, one of the multiple livestock region candidates is selected as the livestock region based on the optical flow generated based on the first and second frames. According to the image analysis system 10, the optical flow is referred to in selecting the livestock region, so that the livestock region can be selected from the multiple livestock region candidates with relatively high accuracy.

[0057] Furthermore, in the image analysis system 10, when one of a plurality of livestock region candidates is selected as a livestock region, different algorithms are used when the speed of the livestock 400 indicated by the optical flow is equal to or greater than a predetermined speed and when the speed of the livestock 400 is less than the predetermined speed. According to the image analysis system 10, an appropriate algorithm is used according to the speed of the livestock 400 (for example, whether the livestock 400 is moving or not) when selecting a livestock region, so that a livestock region can be selected from a plurality of livestock region candidates with relatively high accuracy.

[0058] Furthermore, in the image analysis system 10, when the speed of the livestock 400 is equal to or greater than a predetermined speed, one of the multiple livestock region candidates is selected as the livestock region by referring to the position of the livestock 400 predicted based on the optical flow (estimated livestock region), while when the speed of the livestock 400 is less than the predetermined speed, one of the multiple livestock region candidates is selected as the livestock region by referring to the livestock region in the previous frame. According to the image analysis system 10, an appropriate target according to the speed of the livestock 400 is referred to when selecting the livestock region, so that the livestock region can be selected from the multiple livestock region candidates with relatively high accuracy.

[0059] Furthermore, in the image analysis system 10, when the speed of the livestock 400 is equal to or greater than a predetermined speed, the region among the multiple livestock region candidates that overlaps the livestock position predicted based on the optical flow is selected as the livestock region, whereas when the speed of the livestock 400 is less than the predetermined speed, the region among the multiple livestock region candidates that overlaps the livestock region in the first frame is selected as the livestock region. According to the image analysis system 10, an appropriate target according to the speed of the livestock 400 is referenced when selecting the livestock region, so that the livestock region can be selected from the multiple livestock region candidates with relatively high accuracy.

[0060] 5. Other embodiments The concept of the above embodiment is not limited to the embodiment described above. Hereinafter, an example of another embodiment to which the concept of the above embodiment can be applied will be described.

[0061] In the above embodiment, when a livestock region is selected from a plurality of livestock region candidates, the livestock candidate is selected after the region that does not overlap with the livestock region of the previous frame or the region that does not overlap with the estimated livestock region is deleted from the livestock region candidate. However, when a livestock region is selected from a plurality of livestock region candidates, the region that does not overlap with the livestock region of the previous frame or the region that does not overlap with the estimated livestock region does not necessarily have to be deleted from the livestock region candidate. When a livestock region is selected from a plurality of livestock region candidates, the livestock region candidate with the greatest degree of overlap with the livestock region in the previous frame or the livestock region candidate with the greatest degree of overlap with the estimated livestock region may be selected as the livestock region without deleting the region that does not overlap with the livestock region of the previous frame or the region that does not overlap with the estimated livestock region.

[0062] The above describes the embodiments of the present invention by way of example. That is, the detailed description and the accompanying drawings are disclosed for the purpose of illustrative description. Therefore, some components described in the detailed description and the accompanying drawings may be non-essential components for solving the problem. Therefore, just because the non-essential components are described in the detailed description and the accompanying drawings, it should not be immediately determined that the non-essential components are essential.

[0063] Moreover, the above-described embodiments are merely examples of the present invention in every respect. Various improvements and modifications of the above-described embodiments are possible within the scope of the present invention. For example, at least a part of the configuration of any of the embodiments may be combined with at least a part of the configuration of any of the other embodiments. In other words, in implementing the present invention, specific configurations can be appropriately adopted according to the embodiment. [Explanation of symbols]

[0064] 10 Image analysis system, 100 Camera, 200 Server, 210 Control unit, 211 CPU, 212 RAM, 213 ROM, 220 Memory unit, 221 Control program, 222 First trained model, 223 Second trained model, 224 Third trained model, 230 Communication module, 240 Input / output I / F, 250 Acquisition unit, 252 Livestock area detection unit, 254 Movement amount calculation unit, 256 Delivery sign detection unit, 258 Notification unit, 300 Cowshed, 400 Livestock, F1, F2 Frame, N1 Network, T1-T10 Area.

Claims

1. A camera is provided to capture images of livestock to generate video data; The video data includes a plurality of frames; the plurality of frames include a first frame and a second frame that is generated temporally later than the first frame, The method further includes a computer that extracts a plurality of regions each of which is a candidate region including the livestock by inputting the second frame into a trained model; The computer selects one of the plurality of regions as the region containing the livestock based on an optical flow generated based on the first and second frames.

2. The image analysis system of claim 1, wherein when selecting one of the multiple regions as the region containing the livestock, the computer uses different algorithms when the speed of the livestock indicated by the optical flow is equal to or greater than a predetermined speed and when the speed is less than the predetermined speed.

3. The computer includes: When the speed is equal to or greater than the predetermined speed, selecting one of the plurality of regions as a region including the livestock by referring to a position of the livestock predicted based on the optical flow, The image analysis system of claim 2, further comprising: selecting one of the plurality of regions as the region containing the livestock by referring to the region containing the livestock in the first frame when the speed is less than the predetermined speed.

4. The computer includes: If the speed is equal to or greater than the predetermined speed, a region among the plurality of regions that has the greatest degree of overlap with the position of the livestock predicted based on the optical flow is selected as a region including the livestock, The image analysis system of claim 3, wherein if the speed is less than the predetermined speed, the area among the multiple areas that has the greatest degree of overlap with the area containing the livestock in the first frame is selected as the area containing the livestock.

5. The image analysis system according to any one of claims 1 to 4, wherein the computer detects at least one of the movement amount, posture, and tail-lifted state of the livestock based on image data of an area including the livestock that is continuously selected for each frame.

6. The image analyzing system according to claim 5 , wherein the computer detects signs of parturition in the livestock based on at least one of an amount of movement, a posture, and a state of raising the tail of the livestock.

7. generating video image data by photographing livestock; The video data includes a plurality of frames; the plurality of frames include a first frame and a second frame that is generated temporally later than the first frame, Extracting a plurality of regions, each of which is a candidate region including the livestock, by inputting the second frame into a trained model; and selecting one of the plurality of regions as a region containing the livestock based on an optical flow generated based on the first and second frames.

8. causing a computer to generate video image data by photographing livestock; The video data includes a plurality of frames; the plurality of frames include a first frame and a second frame that is generated temporally later than the first frame, Extracting a plurality of regions, each of which is a candidate region including the livestock, by inputting the second frame into a trained model; and selecting one of the plurality of regions as a region including the livestock based on an optical flow generated based on the first and second frames.

Images