Image-based animal weight measurement device and method

Abstract

An image-based animal weight measurement device according to one embodiment of the present invention comprises: a communication module for receiving an image from at least one camera that captures an object; a memory in which an animal weight measurement program is stored; and a processor for executing the program stored in the memory, wherein the animal weight measurement program inputs a received image into an animal detection model so as to extract feature point information of an individual animal, connects the extracted feature point information so as to calculate the horizontal length and the vertical length of the individual animal, and inputs the calculated horizontal length and vertical length into a weight estimation model so as to calculate the weight of the individual animal.

Description

Image-based animal weight measurement device and method

[0001] The present invention relates to an image-based animal weight measurement device and method.

[0002] In livestock farming, accurately measuring livestock weight is a key component of farm operations. Weight is a key indicator of health, growth, and productivity, and it not only determines economic value but also influences important decisions regarding nutrition, disease prevention, and breeding.

[0003] However, existing weight measurement methods are limited by their high cost and potential for stress to livestock. Recent advances in computer vision and machine learning technologies have led to the development of new methods to address these issues.

[0004] These technologies reduce costs and animal stress through automated data collection, and utilize 2D and 3D imaging to identify livestock and estimate weight. This helps monitor animals in real time and optimize farm management. They can also be integrated with existing farm management systems to support real-time decision-making.

[0005] In relation to this, a system or method is being developed to photograph an animal and estimate its volume, weight, etc. through the image, but there is a problem of reduced accuracy due to errors caused by interference factors (sunlight, animal movement, etc.) and other objects (ground, structures, etc.).

[0006] A prior patent document related to this is Republic of Korea Patent No. 10-2389931 (Title of invention: Mobile device with animal weight inference application installed).

[0007] The present invention is intended to solve the above-mentioned problem, and the purpose of the present invention is to provide an image-based animal weight measurement device and method that estimates weight from the horizontal and vertical lengths of an animal object calculated using characteristic point information of the animal object.

[0008] However, the technical task that this embodiment seeks to achieve is not limited to the technical task described above, and other technical tasks may exist.

[0009] As a technical means for solving the above-described technical problem, an image-based animal weight measurement device according to one embodiment of the present invention includes a communication module for receiving an image from at least one camera that photographs an object; a memory in which an animal weight measurement program is stored; and a processor for executing the program stored in the memory, wherein the animal weight measurement program inputs the received image into an animal detection model to extract feature point information of an animal object, connects the extracted feature point information to calculate the horizontal length and vertical length of the animal object, and inputs the calculated horizontal length and vertical length into a weight estimation model to calculate the weight of the animal object.

[0010] A method for measuring animal weight based on an image using a weight measuring device according to another embodiment of the present invention comprises the steps of: (a) inputting an image received from at least one camera photographing an object into an animal detection model to extract feature point information of the animal; (b) calculating the horizontal and vertical lengths of the animal by connecting the extracted feature point information; and (c) calculating the weight of the animal by inputting the calculated horizontal and vertical lengths into a weight estimation model.

[0011] An image-based animal weight measurement device according to another embodiment of the present invention comprises: a communication module for receiving an image from at least one camera photographing an object; a memory for storing an animal weight measurement program; And a processor that executes a program stored in the memory, wherein the animal weight measurement program inputs the received image into an animal detection model to extract information on multiple feature points of the animal, the horizontal and vertical lengths of the animal, and information on a bounding box formed to fit the animal, and inputs the horizontal length of the animal, the vertical length of the animal, information on multiple feature points, information on the bounding box, and the installation height of the camera that captured the image into a weight estimation model to output the weight of the animal, wherein the feature point information includes the end position of the head of the animal (nose), the neck position (neck), the first back position (back1) close to the head divided into three parts by the back, the second back position (back2) in the middle, the third back position (back3) close to the tail, the right shoulder position (right shoulder), the left shoulder position (left shoulder), the front armpit position (front armpit), or the end position of the torso (hip), and the information on the bounding box includes the coordinates of the center point of the bounding box (Xc, Yc), the width of the bounding box (W), the length of the bounding box (H), and the angle (theta) by which the bounding box is rotated with respect to the reference axis. It includes, and the weight estimation model is trained to estimate the weight of an animal object based on the horizontal length of the animal object, the vertical length of the animal object, information on multiple feature points, information on a bounding box, and the installation height of the camera that captured the image.

[0012] According to any one of the aforementioned means for solving the problem of the present invention, weight can be measured efficiently by processing the received image to detect and measure the animal.

[0013] In addition, since the weight is estimated by considering the actual size and shape of the animal using a weight estimation model based on multiple regression analysis, it can provide more accurate results than existing weight measurement methods.

[0014] Additionally, it provides a non-invasive measurement method for animals, allowing weight measurement without causing stress to the animals.

[0015] FIG. 1 is a block diagram illustrating the configuration of an image-based animal weight measurement device according to one embodiment of the present invention.

[0016] FIG. 2 is a flowchart illustrating an image-based animal weight measurement method according to one embodiment of the present invention.

[0017] FIG. 3A and FIG. 3B are diagrams for explaining a method for calculating the horizontal and vertical lengths of an animal object by connecting feature point information of an animal object extracted by an animal detection model according to one embodiment of the present invention.

[0018] FIG. 4 is a drawing for explaining detailed components of an image-based animal weight measurement device according to one embodiment of the present invention.

[0019] FIG. 5 is a drawing for explaining an animal detection model of an image-based animal weight measurement device according to one embodiment of the present invention.

[0020] FIG. 6a and FIG. 6b are diagrams for explaining observation data collection of a weight estimation model according to one embodiment of the present invention.

[0021] Figure 7 illustrates comparative performance indicators of a weight estimation model according to one embodiment of the present invention.

[0022] FIG. 8 is a diagram for comparing and explaining the weight estimation value and the correct weight value of the OLS model according to one embodiment of the present invention.

[0023] FIG. 9 is a drawing for explaining the accuracy of a weight estimation value of an image-based animal weight measurement device according to one embodiment of the present invention.

[0024] FIG. 10 is a drawing for explaining a weight estimation model according to another embodiment of the present invention.

[0025] FIG. 11 is a diagram for explaining the learning process of a weight estimation model according to another embodiment of the present invention.

[0026] FIG. 12 is a diagram for explaining the performance of a weight estimation model according to another embodiment of the present invention.

[0027] Hereinafter, the present invention will be described in detail with reference to the attached drawings. However, the present invention can be implemented in various different forms and is not limited to the embodiments described herein. In addition, the attached drawings are only intended to facilitate easy understanding of the embodiments disclosed in the present specification, and the technical ideas disclosed in the present specification are not limited by the attached drawings. In order to clearly explain the present invention in the drawings, parts that are not related to the description are omitted, and the size, shape, and shape of each component shown in the drawings can be variously modified. The same / similar drawing reference numerals are assigned to the same / similar parts throughout the specification.

[0028] The suffixes "module" and "part" used in the following description for components are assigned or used interchangeably solely for the convenience of writing the specification, and do not in themselves have distinct meanings or roles. In addition, when describing the embodiments disclosed in this specification, if a detailed description of a related known technology is judged to obscure the gist of the embodiments disclosed in this specification, the detailed description thereof has been omitted.

[0029] Throughout the specification, when a part is said to be "connected (connected, in contact with, or coupled)" to another part, this includes not only cases where it is "directly connected (connected, in contact with, or coupled)" but also cases where it is "indirectly connected (connected, in contact with, or coupled)" with another member in between. Furthermore, when a part is said to "include (have or provide)" a certain component, this does not mean that it excludes other components, but rather that it may "include (have or provide)" other components, unless otherwise specifically stated.

[0030] As used herein, ordinal terms such as "first," "second," etc., are used solely to distinguish one component from another and do not limit the order or relationship of the components. For example, the first component of the present invention may be referred to as the "second component," and similarly, the second component may also be referred to as the "first component."

[0031] Hereinafter, an embodiment of the present invention will be described in detail with reference to the attached drawings.

[0032] FIG. 1 is a block diagram illustrating a configuration of an image-based animal weight measurement device according to one embodiment of the present invention, and FIG. 2 is a flowchart illustrating an image-based animal weight measurement method according to one embodiment of the present invention.

[0033] Referring to FIG. 1, the image-based animal weight measurement device (100) includes a communication module (110), a memory (120), a processor (130), and may further include a database (140). The image-based animal weight measurement device (100) receives images captured in real time through multiple CCTVs or other various cameras (10) placed in a livestock barn, and performs an operation of measuring the weight of an individual animal using the images.

[0034] The image-based animal weight measurement device (100) can be implemented as a computer or portable terminal that can connect to a network. Here, the computer includes, for example, a notebook, a desktop, a laptop, etc., and the portable terminal can include, for example, a wireless communication device that guarantees portability and mobility, and all kinds of handheld-based wireless communication devices such as various smart phones, tablet PCs, smart watches, etc. In addition, the image-based animal weight measurement device (100) can function as a server that provides the weight measurement result of an animal individual for an animal photographed image received from an external source. At this time, the server can operate in a cloud computing service model such as SaaS (Software as a Service), PaaS (Platform as a Service), or IaaS (Infrastructure as a Service), or can be built in the form of a private cloud, a public cloud, or a hybrid cloud.

[0035] A network is a connection structure that enables information exchange between each node, such as terminals and devices, and includes a local area network (LAN), a wide area network (WAN), the Internet (WWW), wired and wireless data communication networks, telephone networks, and wired and wireless television communication networks. Examples of wireless data communication networks include, but are not limited to, 3G, 4G, 5G, 3GPP (3rd Generation Partnership Project), LTE (Long Term Evolution), WIMAX (World Interoperability for Microwave Access), Wi-Fi, Bluetooth communication, infrared communication, ultrasonic communication, visible light communication (VLC), LiFi, etc.

[0036] The communication module (110) can receive an image captured by a camera (10) installed within a predetermined height range of an animal farm and transmit it to the processor (130). At this time, the object includes various types of animals such as cows, pigs, dogs, etc. Here, a detailed description of the placement location of the camera (10) will be provided later.

[0037] The communication module (110) may be a device including hardware and software necessary to transmit and receive signals such as control signals or data signals through wired or wireless connections with other network devices.

[0038] The memory (120) may record an animal weight measurement program. The animal weight measurement program inputs the received image into an animal detection model to extract feature point information of the animal, connects the feature point information to calculate the horizontal and vertical lengths of the animal, and inputs the horizontal and vertical lengths of the animal into a weight estimation model to calculate the weight of the animal.

[0039] At this time, the memory (120) should be interpreted as a general term for a non-volatile storage device that maintains stored information even when no power is supplied and a volatile storage device that requires power to maintain the stored information. The memory (120) may perform a function of temporarily or permanently storing data processed by the processor (130). In addition to a volatile storage device that requires power to maintain the stored information, the memory (120) may include a magnetic storage media or a flash storage media, but the scope of the present invention is not limited thereto.

[0040] The processor (130) executes an animal weight measurement program (hereinafter, “program”) stored in the memory (120) and provides a function of controlling the hardware of the image-based animal weight measurement device (100) according to the execution of the program. That is, the processor (130) can perform hardware control functions such as a necessary file system, memory allocation, network, basic library, timer, device control (display, media, input device, 3D, etc.), and other utilities according to the execution of the program.

[0041] Referring to FIG. 2, the processor (130) inputs the received image into an animal detection model to extract feature point information of the animal (S110), connects the feature point information to calculate the horizontal and vertical lengths of the animal (S120), and inputs the horizontal and vertical lengths of the animal into a weight estimation model to calculate the weight of the animal (S130). In addition, each specific step of the animal weight measurement process according to the execution of the program will be described later with reference to FIGS. 3 to 9.

[0042] In another embodiment, the processor (130) inputs the received image into an animal detection model to extract information about a plurality of feature points of the animal object, the horizontal and vertical lengths of the animal object, and information about a bounding box formed to fit the animal object, and inputs the horizontal length of the animal object, the vertical length of the animal object, information about a plurality of feature points, information about the bounding box, and the installation height of the camera that captured the image into a weight estimation model to output the weight of the animal object. Here, a detailed description of the weight estimation process of the animal according to the execution of the program will be described later with reference to FIGS. 10 to 12.

[0043] The processor (130) may include any type of device capable of processing data. For example, it may refer to a hardware-embedded data processing device having a physically structured circuit to perform a function expressed by a code or command included in a program. Examples of such hardware-embedded data processing devices include processing devices such as a microprocessor, a central processing unit (CPU), a processor core, a multiprocessor, an application-specific integrated circuit (ASIC), and a field programmable gate array (FPGA), but the scope of the present invention is not limited thereto.

[0044] The database (140) stores or provides data required for the image-based animal weight measurement device (100) under the control of the processor (130). For example, the database (140) cumulatively stores weights extracted by the animal weight estimation program, and can be utilized in various applications for monitoring the condition of animals based on such weights. This database (140) may be included as a separate component from the memory (120), or may be constructed in a portion of the memory (120).

[0045] FIG. 3A and FIG. 3B are drawings for explaining a method for calculating the horizontal and vertical lengths of an animal object by connecting characteristic point information of the animal object according to one embodiment of the present invention, and FIG. 4 is a drawing for explaining detailed components of an image-based animal weight measurement device according to one embodiment of the present invention.

[0046] First, referring to FIG. 4, the processor (130) may implement detailed modules that perform various functions according to the execution of the animal weight measurement program. For example, the animal weight measurement program may be executed by the processor (130) to implement an animal detection model (200) and a weight estimation model (300).

[0047] Referring to FIG. 3A, the program inputs an input image into an animal detection model (200) and extracts object detection information. The object detection information includes information on a bounding box formed to fit an animal object detected in the input image and feature point information of the animal object. For example, the information on the bounding box of the animal object may include the center point coordinates (Xc, Yc) of the bounding box, the width (W) of the bounding box, the length (H) of the bounding box, and the angle (theta) by which the bounding box is rotated with respect to the reference axis. In addition, the feature point information of the animal object may include the end position of the head of the animal object (nose), the neck position (neck), the first back position (back1) close to the head that divides the back into three equal parts, the second back position (back2) in the middle, the third back position (back3) close to the tail, the right shoulder position (right shoulder), the left shoulder position (left shoulder), the front armpit position (front armpit), and the end position of the torso (hip).

[0048] Additionally, object detection information can include additional information about the type (class) of the animal detected in the image. This information can be used to distinguish between different species or even within the same species by growth stage. For example, in the case of pigs, the types can be categorized as suckling piglets, weaned piglets, growing pigs, fattening pigs, gilts, pregnant pigs, and farrowing pigs.

[0049] Meanwhile, training data for each image can be generated through a process of labeling the coordinates of the center point of the bounding box, the width of the bounding box, the length of the bounding box, the rotation angle of the bounding box, and the feature point information of the animal object as correct data for each image. For this purpose, labeling can be performed by having an expert manually identify the bounding box and determine the coordinates of the center point of the bounding box, the width of the bounding box, and the length of the bounding box. Furthermore, training data can be prepared by displaying the feature point information of the animal object for each image.

[0050] Next, the program connects the feature point information of the animal object extracted from the animal detection model to calculate the horizontal and vertical lengths of the animal object.

[0051] For example, the program can calculate the length of the line segment connecting the right shoulder and left shoulder of an animal as the horizontal length (SWM, body width) of the animal. Additionally, the program can calculate the length of the line segment connecting the neck of the animal to the hip as the vertical length (SLM, body length) of the animal.

[0052] To this end, the program can convert the pixel distance between feature point information of an animal object extracted from an input image into a centimeter distance using a calibration pattern image captured by a camera (10).

[0053] For example, the program can perform distortion correction so that the pixel distances of all chessboard sides are the same by using a calibration pattern image arranged so that the center of the chessboard is located at the exact center of the angle of view of the camera (10). For reference, the chessboard is a grid pattern of multiple squares, as illustrated in FIG. 6a, which will be described later, and is mainly used to correct the distortion of a camera-captured image based on this. At this time, the distortion correction method can uniformly depict the squares of the chessboard in the image by utilizing VPI (Vision Programming Interface) and a polynomial distortion model. Since this distortion correction method corresponds to a prior art, a detailed description thereof will be omitted.

[0054] Next, the program can convert the pixel distance between the feature points of the animal object extracted from the input image into a centimeter distance. This is calculated using the following mathematical equation (1).

[0055] <Mathematical Formula 1>

[0056]

[0057] At this time, chessboard pixel is the pixel distance value of the chessboard side, and chessboard centimeter is the actual length value of the chessboard side.

[0058] For example, the program can determine the horizontal length based on the distance between shoulders among the feature point information, and can determine the vertical length based on the distance between the neck and the end of the torso among the feature point information. For example, the actual horizontal length (SWM) of the animal object can be obtained by multiplying the pixel distance between the right shoulder and the left shoulder point (distright shoulder, left shoulder) by the centimeter-per-pixel ratio value (p2c). In addition, the program can obtain the actual vertical length (SLM) of the animal object by multiplying the centimeter-per-pixel ratio value (p2c) by the sum of the pixel distance between the neck and the first back point, the pixel distance between the first back and the second back point, the pixel distance between the second back and the third back point, and the pixel distance between the third back and the end of the torso point (distneck, back1+distback1, back2+distback2, back3+distback3, hip). In addition, the horizontal length can be determined based on the distance between the feature points at one end and the feature points at the other end along the horizontal direction of the livestock object, and the vertical length can be determined based on the distance between the feature points at one end and the feature points at the other end along the vertical direction of the livestock object. For example, the feature points of the livestock object arranged along the horizontal direction of the livestock object may be the left shoulder, the right shoulder, etc. In addition, the feature points of the livestock object arranged along the vertical direction of the livestock object may be the nose, eyes, head, neck, back, tail, and buttocks, etc.

[0059] Next, the program inputs the calculated horizontal length (SWM) and vertical length (SLM) into the weight estimation model (300) to calculate the weight of the animal.

[0060] Below, we will examine the weight estimation model (300).

[0061] Meanwhile, Fig. 3b illustrates an upper triangular correlation matrix, which shows SLM (body_cm) and SWM (shoulder_cm), which show the highest correlation with the correct weight (weight_answer) among the variables connecting the feature points. Accordingly, in the present invention, in order to construct a weight estimation model (300), the horizontal length (SWM) and vertical length (SLM) among the variables connecting the feature points were used.

[0062] The weight estimation model (300) is a multiple regression analysis model, and can calculate the weight of an individual animal using mathematical formula 2.

[0063] <Mathematical Formula 2>

[0064]

[0065] Here, W is weight, SLM (Spine Length Measure) is the vertical length of the animal, SWM (Shoulder Width Measure) is the horizontal length of the animal, and β0, β1, and β2 are regression coefficients that adapt to the growth trajectory of the animal.

[0066] In order to build a weight estimation model (300), the program collects observation data matching the horizontal length, vertical length, and correct weight values ​​of an animal individual, and generates an OLS (Ordinary Least Squares) regression model based on the observation data to determine the regression coefficient of Equation 1. At this time, the observation data collection process and the OLS regression model generation process will be described later with reference to FIGS. 6a to 9.

[0067] The regression coefficients (β0, β1, and β2) can be optimized by minimizing e (error) of the multiple regression formula.

[0068] The program transforms Equation 2 into a multiple linear regression matrix form, as in Equation 3, and calculates the β matrix that minimizes the SSE function of multiple regression according to Equation 4, thereby obtaining the optimal β value.

[0069] <Mathematical Formula 3>

[0070]

[0071] <Mathematical Formula 4>

[0072]

[0073] Below, we will examine an animal detection model that extracts feature point information of the animal individuals described above.

[0074] FIG. 5 is a drawing for explaining an animal detection model of an image-based animal weight measurement device according to one embodiment of the present invention.

[0075] The animal detection model (200) used in the present invention is constructed based on learning data that matches multiple images containing at least one animal object and object detection information (feature point information) for the animal objects included in each image. After being trained through a learning process, the animal detection model (200) can automatically output object detection information for an input image through an inference process for an actual input image.

[0076] The learning data used in the learning process of the animal detection model (200) includes multiple images and object detection information matched to each image. At this time, the object detection information is manually extracted for each image. That is, an expert can manually view each image and input object detection information using an appropriate SW tool, or an expert can automatically input the information using an existing animal detector that has been fully developed and then modify / supplement it. For example, the expert displays a bounding box for each animal object included in the image, considering the direction of rotation relative to the reference axis of the animal object, and generates information such as the coordinates of the center point of each bounding box, the width of the bounding box, the length of the bounding box, and the angle at which the bounding box is rotated with respect to the reference axis. In addition, the worker can additionally extract information on the type of animal object and use it as learning data.

[0077] As illustrated in FIG. 4, the animal detection model (200) may include a back part (210), a neck part (220), and a head part (230).

[0078] The backbone (210) extracts features from input images and is a commonly used component in deep neural network-based image analysis and processing methods. The backbone (210) primarily takes the form of a 2D convolutional accumulation, and has been enhanced to incorporate various neural network structures to improve its efficiency. Backbones of various structures commonly receive images and extract intermediate information, which is then transmitted to the neck (220).

[0079] The neck unit (220) collects intermediate information from each layer of the back unit (210) based on the features extracted from the back unit (210). The neck unit (220) is a lower neural network that constitutes a universal object detector and performs the role of collecting and interpreting the intermediate information of each layer of the back unit (210). Since the resolution of the image interpreted by each layer is different, the neck unit (220) extracts intermediate information of each layer to effectively detect animals of various sizes according to the body type of the animal, whether the target is far away or close, and provides it to the head unit (230). The specific configuration of the neck unit (220) varies depending on the shape of the back unit (210) described above, and the number of layers and hyperparameters of each layer of the specific neural network constituting the neck unit (220) may vary depending on the shape of the back unit (610).

[0080] The head unit (230) outputs object detection information based on the intermediate information collected from the neck unit (220). The head unit (230) receives the intermediate information obtained from the neck unit (220) and outputs object detection information. The head unit (230) receives the intermediate information of each layer of the neck unit (220) and outputs object detection information recognized for each layer. In particular, the head unit (230) of the present invention includes a plurality of animal detection subnets, and each animal detection subnet includes a subnet for extracting bounding boxes and feature point information, and a subnet for extracting animal types, as shown in FIG. 5.

[0081] Meanwhile, an NMS (Non-maximum Suppression) module may be further combined at the output terminal of the head unit (230). This is an algorithm for selecting a bounding box with the highest similarity when multiple bounding boxes are created for the same object, and since it corresponds to prior art, a detailed description thereof will be omitted.

[0082] The subnet that extracts bounding boxes and feature information consists of cascaded multi-lane deep convolutional networks. The cascaded multi-lane deep convolutional networks are constructed according to a causal order to find bounding boxes and feature information for a given animal image. The following causal order is followed to define a single object detection information in each image.

[0083] That is, as illustrated in Fig. 3a, first, the center point (Xc, Yc) and the aforementioned feature point information (nose, ..., hip) are indicated. Next, a tangent line is drawn that crosses the center point and at least one point of the feature point information (excluding the right shoulder position and the left shoulder position). Finally, the area (surface) through which the tangent line passes through the center is specified.

[0084] This type of cascaded multipath deep convolutional network outputs information according to the causal order described above. That is, the first path outputs center point and feature point information, the second path outputs the direction of the tangent (theta), and the third path outputs the width (W) and height (H) of the region (bounding box) containing the tangent and center points.

[0085] Meanwhile, the subnet that extracts the animal type is obtained through a general structure, i.e., a single-path deep convolutional network.

[0086] An animal detection model (200) like this can also be expressed in the following mathematical formula.

[0087]

[0088] Here, A={RK,c,p} denotes vectorized object detection information, M(x) denotes an animal detection model, I denotes an input image matrix (having dimensions of image width x image height x image channel), and E(A) denotes encoded animal detection information. In addition, B(x), N(x), and H(x) denote the back part (210), the neck part (220), and the head part (230), respectively.

[0089] When the input image matrix is ​​input to the animal detection model M(x), the output is trained to be identical to the encoded animal detection information E(A), and the animal detection model is built through a process in which the weights of the animal detection model are repeatedly updated through backpropagation learning.

[0090] The learning data used in the learning process of the animal detection model (200) includes multiple images and object detection information matched to each image. At this time, some of the object detection information may be manually extracted for each image. That is, the worker may look at each image, use an appropriate SW tool to specify a bounding box, and directly input information such as the center point coordinates, width, length, and feature points of the animal object of the bounding box, or may use an existing animal detector to automatically input information and then have a person modify / supplement it. For example, the worker may display a bounding box for each animal object included in the image, taking into account the direction of rotation relative to the reference axis of the animal object, and generate the center point coordinates of each bounding box, the width of the bounding box, and the length of the bounding box. Then, the worker generates nine feature point information, including the head end position (nose), neck position (neck), back1, back2, back3, right shoulder position (right shoulder), left shoulder position (left shoulder), front armpit position (front armpit), and torso end position (hip), for each animal contained in each bounding box, and includes the feature point information in the learning data. In addition, the worker additionally extracts information on the type of animal and uses it as learning data. The object detection information contained in the learning data goes through an encoding process before being used in the learning process. The object detection information can be encoded through the following process.

[0091] First, the area of ​​interest (230) by head part ) is specified. At this time, the area of ​​interest ( ) is defined as the processing area x size type x angle type x box ratio for each head part (230).

[0092] Additionally, the degree of overlap (o) between the animal area and the area of ​​interest recorded in the object detection information (A) a,k ) is calculated using the following mathematical formula.

[0093]

[0094] Here, IoU(x,y) calculates the degree of overlap between two bounding boxes.

[0095] Next, the animal region with the highest degree of overlap for each region of interest (R k ) only. Here, k' = argmax k (o a,k )am.

[0096] Next, encoding is performed between the region of interest and the corresponding animal region.

[0097]

[0098] At this time, is processed as follows:

[0099] When,

[0100] As such, it is processed.

[0101] Accordingly, the output is as follows.

[0102]

[0103] likewise, is processed as follows:

[0104] When,

[0105] It is processed as follows.

[0106] Accordingly, the output is as follows.

[0107]

[0108] Through the above process, encoded detection information is used in the process of building an animal detection model (200).

[0109] Meanwhile, backpropagation learning may be used in the learning process of the animal detection model (200). That is, the process of calculating the loss value between the encoded object detection information E(A) and its estimate and updating the neural network parameters constituting the animal detection model (200) so that this loss value becomes smaller is repeated. For example, when calculating the loss value of the bounding box (rbbox) and the keypoints of an animal object, L1 or L2 loss may be used, and the loss value of the type (c) or pose (p) of an animal object may be a discriminative loss such as binary cross entropy loss or focal loss.

[0110] Using this loss function, an animal detection model (200) is built by repeating learning until the total loss falls below the target value.

[0111] In this way, we will examine the process of inferring object detection information (A) for an input image using the constructed animal detection model (200) (M(x)). This can be expressed mathematically as follows.

[0112]

[0113] That is, when an input image is input to the animal detection model (200), encoded detection information (E(A)) can be obtained. Accordingly, a process of decoding the encoded detection information is performed.

[0114] In, c a Only the values ​​when c is higher than the threshold are left. In other words, a' = a if c a > thr, where thr is the detection threshold.

[0115] and, Each of these is performed to obtain decrypted animal detection area and feature point information. The processing process is as follows.

[0116] When,

[0117] It is processed as follows.

[0118] Accordingly, the output is as follows.

[0119]

[0120] likewise, is processed as follows:

[0121] , When,

[0122] It is processed as follows.

[0123] Accordingly, the output is as follows.

[0124]

[0125] The object detection information (R, K, c, p) output through the decryption process may contain multiple overlapping pieces of information for a single animal. To address this, an algorithm that removes overlapping object detection information can be applied.

[0126] In this way, object detection information can be overlaid on an image to visually confirm object detection information.

[0127] FIG. 6a and FIG. 6b are diagrams for explaining observation data collection of a weight estimation model according to one embodiment of the present invention, and FIG. 7 is a table summarizing pig movement sessions of multiple farms for explaining observation data of a weight estimation model according to one embodiment of the present invention.

[0128] Referring to Fig. 6a, the camera (10) installation height was set to 1.8 to 2.5 meters to accommodate various farm infrastructures, with a preferred installation height of 2.2 meters for optimal coverage. Furthermore, the camera's (10) field of view was set to a minimum horizontal length of 0.9 meters to enable complete animal capture. Meanwhile, an automatic adjustment device precisely adjusted the camera's field of view to the width of the corridor, ensuring consistent image quality and data accuracy.

[0129] Figure 6b is a table summarizing pig movement sessions from various farms, categorized by group composition and showing the distribution of O80 and U40 pigs.

[0130] As illustrated in Figure 6b, in one example, data were collected from a total of 290 pig movement sessions across five different farms. This dataset includes the number of pig movements, total weight, and images of pig movements. The present invention tailored the data collection strategy to pig growth stages, categorizing them into O80 (over 80 kg) and U40 (under 40 kg) to ensure age-appropriate accuracy.

[0131] For the O80 group of adult pigs, due to the impracticality of individual weight measurements, which would take 10-20 minutes, transport weight certificates were utilized. These documents record the vehicle weights before and after loading the pigs. The average weight per pig was calculated from the net weight of the pigs, yielding the correct weight (GT) for the adult pigs.

[0132] For the younger U40 group, weights were collected using multi-pig scales, which can efficiently weigh multiple pigs. Accurate average weights were calculated during the farm transfer, allowing accurate weights of the young pigs to be collected.

[0133] This dual data collection can simplify the steps for collecting observational data and enhance the reliability of body weight estimation models for adult and young pigs.

[0134] Using the animal detection model (200), individual pig detection information was extracted, and the lengths between pixel-based feature points were converted into centimeter distances to obtain the pig's vertical and horizontal lengths (SLM / SWM). To accurately estimate body weight from these vertical and horizontal lengths of individual animals, an OLS model was created. The OLS model undergoes an iterative calibration process to minimize the residual, which is the difference between the true body weight and the predicted body weight. The directionality of the pigs as they pass through the imaging area was also taken into account to prevent the viewpoint from distorting the measurements.

[0135] This directional consideration is expressed in the following weight estimation formula:

[0136]

[0137] Here, w* is the improved weight estimate, and N l and N r are the number of pigs moving left and right respectively, and w l,i Wow w r,j is the corresponding estimated weight. Through data integration that takes directionality into account, an accurate weight estimate is obtained.

[0138] The predictive accuracy of the OLS model according to the present invention was improved by harmonizing it with the known average weight of the population. This fine-tuned the predictive accuracy by gradually reducing the prediction error range.

[0139] Figure 7 illustrates comparative performance indicators of a weight estimation model according to one embodiment of the present invention.

[0140] Referring to Figure 7, when an OLS model was created by applying the entire data set, the average relative error was 0.0499. On the other hand, when the U40 and O80 data were separated and individual OLS models were created, the O80 OLS model showed a low error rate, indicating improved accuracy. In particular, the OLS model in which each OLS (U40 and O80) was merged showed an average relative error almost similar to that of the existing DNN model. In this way, the present invention can simplify and reduce computational costs without compromising accuracy compared to DNN by applying OLS to the weight estimation model.

[0141] FIG. 8 is a drawing for comparing and explaining a weight estimation value and an accurate weight value of an OLS model according to one embodiment of the present invention, and FIG. 9 is a drawing for explaining the accuracy of a weight estimation value of an image-based animal weight measurement device according to one embodiment of the present invention.

[0142] As shown in Figure 8, the range of maximum error rates in the farm-specific test set is 5.8% to 17.4%, indicating potential areas for improvement.

[0143] Figure 9 shows the prediction results of the OLS model for two segments, U40 and O80. That is, the OLS model weight estimation results for the U40 and O80 groups indicate the accuracy of the model across various weight classes.

[0144] FIG. 10 is a drawing for explaining a weight estimation model according to another embodiment of the present invention, FIG. 11 is a drawing for explaining a learning process of a weight estimation model according to another embodiment of the present invention, and FIG. 12 is a drawing for explaining the performance of a weight estimation model according to another embodiment of the present invention.

[0145] Hereinafter, a weight estimation model (310) according to another embodiment of the present invention will be described.

[0146] The animal weight measurement program inputs the received image into an animal detection model (200) to extract information on multiple feature points of the animal, the horizontal and vertical lengths of the animal, and information on a bounding box formed to fit the animal, and inputs the horizontal length of the animal, the vertical length of the animal, information on multiple feature points, information on the bounding box, and the installation height of the camera that captured the image into a weight estimation model (310) to output the weight of the animal.

[0147] Here, the feature point information includes the end position of the head of the animal (nose), the neck position (neck), the first back position (back1) close to the head that divides the back into three equal parts, the second back position (back2) in the middle, the third back position (back3) close to the tail, the right shoulder position (right shoulder), the left shoulder position (left shoulder), the front armpit position (front armpit), or the end position of the body (hip). In addition, the information about the bounding box includes the center point coordinates of the bounding box (Xc, Yc), the width of the bounding box (W), the length of the bounding box (H), and the angle (theta) by which the bounding box is rotated with respect to the reference axis.

[0148] For example, the horizontal length of an animal object may be the sum of the pixel distance between the neck and the first back point, the pixel distance between the first back and the second back point, the pixel distance between the second back and the third back point, and the pixel distance between the third back and the end point of the body among the feature point information of the animal object, and the vertical length of the animal object may be the pixel distance between the right shoulder and the left shoulder point among the extracted feature point information of the animal object.

[0149] The weight estimation model (310) is trained to estimate the weight of an animal based on the horizontal length of the animal, the vertical length of the animal, information on multiple feature points, information on a bounding box, and the installation height of the camera that captured the image.

[0150] Referring to FIG. 10, the program generates sequence data based on the horizontal length of the animal object, the vertical length of the animal object, information on multiple feature points, information on the bounding box, and the installation height of the camera that captured the image extracted from the animal detection model (200) described above, and inputs the generated sequence data for each animal object into the weight estimation model (310) to output a weight prediction value for each animal object.

[0151] At this time, referring to FIG. 11, the sequence data is generated by grouping data including the horizontal length (body), vertical length (shoulder), multiple feature point information (keypoint), bounding box information (rbbox), and camera installation height (cam height) of an animal object obtained from multiple frames into a predetermined sequence length (number of frames). For example, the sequence data can be configured in the format (B, 240) when the sequence length (S) is 10 and the number of features per frame (X) is 24, where B is the batch size.

[0152] For example, the weight estimation model (310) can input sequence data and output a weight prediction value of an animal individual by performing the following sequential processing based on the sequence data.

[0153] First, the weight estimation model (310) can perform a multihead attention operation on the input sequence data. This multihead attention operation is designed to simultaneously consider and learn the interrelationships between each frame constituting the sequence from multiple perspectives, thereby integrating motion features in the time series and important frame information to generate a high-dimensional embedding vector (B, E) representing the sequence.

[0154] Next, the weight estimation model (310) can normalize the output value to a range of 0 to 1 by applying a sigmoid (scaled sigmoid) activation function to the embedding vector. At this time, the normalized value plays a role in increasing the stability of model learning and adjusting the value within a limited range so that it can be converted to an actual weight unit in a later stage. For example, the sigmoid activation function is intended to complement the resolution limit of the general sigmoid function, and includes α (alpha) and β (beta) parameters to increase the prediction sensitivity, especially in the low weight range (e.g., 3 to 20 kg). The weight estimation model (310) is designed to maintain a uniform prediction sensitivity across the entire weight range by searching for the optimal α and β values ​​during the learning process.

[0155] Next, the weight estimation model (310) can convert the normalized value into a numerical value within the actual weight range through weight denormalization. For example, the weight estimation model (310) can correct the normalized value (e.g., 0 to 1 value) based on a predefined minimum weight (e.g., 3 kg) and maximum weight (e.g., 200 kg) to ultimately produce a weight prediction value (B, W) in kg.

[0156] Next, the learning process of the weight estimation model (310) according to another embodiment of the present invention will be described.

[0157] The weight estimation model (310) may be constructed using learning data including, for each of a plurality of animal individuals, sequence data grouped by a predetermined sequence length based on a plurality of frames including the horizontal length of the animal individual, the vertical length of the animal individual, information on a plurality of feature points, information on a bounding box, and a camera installation height, and average weight information obtained from a movement record report as the weight correct value of each animal individual corresponding to the sequence data.

[0158] Here, movement records are defined based on shipment records, and each report includes information on the total number of animals shipped and their total weight. Therefore, the average weight based on the report can be used as the correct data, and actual weight information for individual animals often does not exist separately.

[0159] The sequence data used for learning the weight estimation model (310) can be composed based on movement information (sequence) tracked for each object through the animal detection model (200) and the existing tracking model, and each frame is stored in chronological order, and only frames that satisfy a predefined pose for weight measurement can be filtered and utilized for sequence composition.

[0160] The learning data configuration process of the weight estimation model (310) is described with reference to FIG. 11.

[0161] For example, track IDs that satisfy a minimum sequence length or more may be selected, and sampling may be performed based on a set sequence length (e.g., 10 frames) among frames listed in time-series order for each object. This sampling may be repeated multiple times (M times) for one track ID, thereby generating multiple sequence samples for the same object.

[0162] For example, each sequence data is composed of a total of 24-dimensional feature vectors including the vertical length (body) of an animal individual, horizontal length (shoulder), bounding box information (rbbox(5)), multiple feature point information (keypoint(16)), and camera installation height (cam_height(1)), and can be composed of an input tensor in the form of (S × 24) when the sequence length S is 10. For one track ID, sequence sampling can be repeated M times, and a total of M sequence data can be generated from trk1_s(1) to trk1_s(M). Each sequence data is input to a weight estimation model (310), and the sample predicted weights (sample_est_weight) from W(1,1) to W(1,M) are calculated, and for the same individual, the average of M individual weight predicted values ​​(object_est_weight) can be calculated. That is, for each identical individual corresponding to track ID 1 to track ID N, individual unit weight prediction values ​​are calculated, and by averaging these, a single report unit average weight prediction value (report_est_avg_weight) can be calculated. At this time, the overall report average prediction value can be calculated by the following mathematical formula.

[0163]

[0164] Here, N means the total number of track IDs included in the report, M means the number of sequence samples generated for each track ID, and w ij is the predicted weight value for the jth sequence sample of the ith object, is the average value of the sequence prediction weight for all objects, and means the average prediction weight (report estimated average weight) of the corresponding report unit.

[0165] Based on this learning data, a weight estimation model (310) can be trained using a combination of five loss functions described below. Each loss function is applied with equal weights, and the overall loss can be calculated as the average of these.

[0166] For example, the embedding loss can be implemented using triplet loss, which minimizes the difference in embedding distance between sequence samples extracted from the same object ID and maximizes the distance between embeddings between different IDs. The margin can be set to 0.1.

[0167] Consistency Loss is a loss that minimizes the deviation between predicted weight values ​​for multiple sequence samples generated from the same ID, and can be implemented as Smooth L1 Loss.

[0168] Target Loss is a loss function that induces the average of the N track IDs in each report and the M sequence prediction values ​​generated for each ID to be similar to the average weight value of the report, and can be configured based on Huber Loss.

[0169] The standard deviation maximum loss (Stddev Max Loss) is a loss that prevents the entire prediction results from being excessively biased toward a specific report average value. It can be used by modifying the Smooth L1 Loss in a way that maintains the variance between the prediction values.

[0170] Rank Loss is a loss that induces the relative order of weight predictions within each report to match the relative order of the area calculated by multiplying the vertical length (body) and vertical length (shoulder) of each animal individual for each sequence, and can be implemented using ListNet Loss.

[0171] Meanwhile, AdamW can be applied as an optimization function for training the weight estimation model (310), and CosineAnnealingWarmRestarts can be applied as a learning rate scheduler. These are widely used components to enhance the learning stability and generalization performance of the model.

[0172] The prediction accuracy of the weight estimation model (310) is described with reference to Figure 12. Verification was performed based on separately collected evaluation reports, not used to train the weight estimation model of the present invention. The evaluation targets were 28 shipment reports collected from five different farms, each containing movement information for a total of 861 pigs.

[0173] Each farm had an average weight range ranging from a minimum of about 5 kg to a maximum of about 130 kg, and thus the weight estimation model of the present invention was shown to have generalizable predictive performance for various weight ranges.

[0174] The model's performance was evaluated based on the mean absolute error (MAE) and mean relative error (MRE), and the prediction results were compared by dividing them into the basic DNN output and the DNN+B result with bias correction applied.

[0175] Specifically, Fig. 12 is a table showing the MAE results and the MRE results, and compares the number of reports for each farm, the number of individuals, the average bias value (DNN bias), and the DNN-based prediction error and the error after correction (DNN+B).

[0176] DNN bias refers to the average value of the difference between the predicted average weight and the actual average weight based on each report, and DNN+B shows the MAE and MRE performance calculated after applying the bias value to the predicted value.

[0177] Based on the average of all evaluation subjects, MAE improved from 2.22 kg before correction to 1.77 kg after correction, and MRE decreased from 3.7% before correction to 3.0% after correction, confirming that the weight estimation model (310) of the present invention provides high prediction accuracy even in an actual application environment.

[0178] In particular, for j_farm, the MAE before correction was 3.22 kg and the MRE was 2.8%, but when DNN+B correction was applied, the MAE was significantly reduced to 0.73 kg and the MRE was 0.6%, showing that post-correction of the model prediction results is effective in improving actual accuracy.

[0179] Additionally, for hs_farm and s2_farm, there was little difference before and after correction, which may mean that the model's predictions for the data were already made without bias.

[0180] Therefore, the weight estimation model (310) according to another embodiment of the present invention can provide stable and reliable prediction performance even in various farm environments and weight ranges.

[0181] Below, the description of the same configuration among the above-described configurations is omitted.

[0182] Referring again to FIG. 2, the image-based animal weight measurement method using a weight measurement device includes a step (S110) of inputting an image received from at least one camera (10) that photographs an object into an animal detection model (200) to extract feature point information of the animal object, a step (S120) of calculating the horizontal and vertical lengths of the animal object by connecting the extracted feature point information, and a step (S130) of calculating the weight of the animal object by inputting the calculated horizontal and vertical lengths into a weight estimation model (300).

[0183] The image-based animal weight measurement method may further include a step of constructing a weight estimation model. In this case, the weight estimation model can calculate the weight of the animal individual using the mathematical equation 2 described above.

[0184] The step of constructing a weight estimation model may include a step of collecting observation data in which horizontal length, vertical length, and correct weight values ​​of an animal individual are matched, and a step of generating an OLS (Ordinary Least Squares) regression model based on the observation data to determine the regression coefficient of the above-described mathematical expression 2.

[0185] Step S120 may include a step of determining a horizontal length based on the distance between shoulders among the feature point information, and a step of determining a vertical length based on the distance between the neck and the end of the torso among the feature point information. In addition, the step may include a step of converting a pixel distance between feature point information into a centimeter distance using a calibration pattern image captured by a camera (10).

[0186] The animal detection model (200) is constructed based on learning data that matches a plurality of images containing at least one animal object and feature point information for the animal object included in each image, and includes a back unit (210) that extracts features from an input image, a neck unit (220) that collects intermediate information from each layer of the back unit (210) based on the features extracted from the back unit (210), and a head unit (230) that outputs feature point information based on the intermediate information collected from the neck unit (220).

[0187] An embodiment of the present invention may also be implemented in the form of a recording medium containing computer-executable instructions, such as program modules, executed by a computer. Computer-readable media may be any available media that can be accessed by a computer, and includes both volatile and nonvolatile media, removable and non-removable media. Computer-readable media may also include computer storage media. Computer storage media includes both volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storing information, such as computer-readable instructions, data structures, program modules, or other data.

[0188] Although the methods and systems of the present invention have been described with respect to specific embodiments, some or all of their components or operations may be implemented using a computer system having a general-purpose hardware architecture.

[0189] The above description of the present invention is for illustrative purposes only, and those skilled in the art will readily appreciate that the present invention can be readily modified into other specific forms without altering the technical spirit or essential characteristics of the present invention. Therefore, the embodiments described above should be understood as illustrative in all respects and not restrictive. For example, each component described as a single entity may be implemented in a distributed manner, and similarly, components described as distributed may be implemented in a combined manner.

[0190] The scope of the present invention is indicated by the claims described below rather than the detailed description above, and all changes or modifications derived from the meaning and scope of the claims and their equivalent concepts should be interpreted as being included in the scope of the present invention.

Claims

1. In an image-based animal weight measuring device, A communication module for receiving images from at least one camera that photographs an object; Memory in which the animal weight measurement program is stored; and A processor that executes a program stored in the above memory, The above animal weight measurement program is an image-based animal weight measurement device that inputs the received image into an animal detection model to extract feature point information of the animal, connects the extracted feature point information to calculate the horizontal and vertical lengths of the animal, and inputs the calculated horizontal and vertical lengths into a weight estimation model to calculate the weight of the animal.

2. In paragraph 1, The above animal weight measurement program is, An image-based animal weight measuring device, wherein the horizontal length is determined based on the distance between shoulders among the above characteristic point information, and the vertical length is determined based on the distance between the neck and the end of the torso among the above characteristic point information.

3. In paragraph 1, The above weight estimation model is an image-based animal weight measuring device that calculates the weight of the animal individual by mathematical formula 1. <Mathematical Formula 1> Here, W is weight, SLM (Spine Length Measure) is the vertical length of the animal, SWM (Shoulder Width Measure) is the horizontal length of the animal, and β0, β1, and β2 are regression coefficients that adapt to the growth trajectory of the animal.

4. In paragraph 3, The above animal weight measurement program is, An image-based animal weight measurement device that collects observation data matching the horizontal length, vertical length, and correct weight values ​​of an animal object, creates an OLS (Ordinary Least Squares) regression model based on the observation data, and determines the regression coefficient of the above mathematical formula 1.

5. In paragraph 1, The above animal weight measurement program is, An image-based animal weight measurement device that converts pixel distances between feature point information of the animal into centimeter distances using a calibration pattern image captured by the camera.

6. In paragraph 1, The above animal detection model is constructed based on learning data that matches a plurality of images including at least one animal object and the feature point information for the animal object included in each image, An image-based animal weight measurement device comprising a back unit for extracting features from an input image, a neck unit for collecting intermediate information from each layer of the back unit based on the features extracted from the back unit, and a head unit for outputting the feature point information based on the intermediate information collected from the neck unit.

7. In an image-based animal weight measurement method using a weight measurement device, (a) a step of inputting images received from at least one camera that captures an object into an animal detection model to extract feature point information of an animal object; (b) a step of calculating the horizontal and vertical lengths of the animal by connecting the extracted feature point information; and (c) An image-based animal weight measurement method, comprising a step of inputting the calculated horizontal and vertical lengths into a weight estimation model to calculate the weight of the animal.

8. In paragraph 7, Step (b) above An image-based animal weight measurement method, comprising: a step of determining a horizontal length based on the distance between shoulders among the above characteristic point information; and a step of determining a vertical length based on the distance between the neck and the end of the torso among the above characteristic point information.

9. In paragraph 7, Further comprising a step of building the above weight estimation model, The above weight estimation model is an image-based animal weight measurement method that calculates the weight of the animal individual by mathematical formula 1. <Mathematical Formula 1> Here, W is weight, SLM (Spine Length Measure) is the vertical length of the animal, SWM (Shoulder Width Measure) is the horizontal length of the animal, and β0, β1, and β2 are regression coefficients that adapt to the growth trajectory of the animal.

10. In paragraph 9, The steps of building the above weight estimation model are: An image-based animal weight measurement method, comprising: a step of collecting observation data in which horizontal length, vertical length, and correct weight values ​​of an animal are matched; and a step of generating an OLS (Ordinary Least Squares) regression model based on the observation data to determine the regression coefficient of the mathematical expression 1.

11. In paragraph 7, Step (b) above An image-based animal weight measurement method, comprising a step of converting a pixel distance between feature point information of the animal into a centimeter distance using a calibration pattern image captured by the camera.

12. In paragraph 7, The above animal detection model is constructed based on learning data that matches a plurality of images including at least one animal object and the feature point information for the animal object included in each image, An image-based animal weight measurement method, comprising: a back unit for extracting features from an input image; a neck unit for collecting intermediate information from each layer of the back unit based on the features extracted from the back unit; and a head unit for outputting the feature point information based on the intermediate information collected from the neck unit.

13. A non-transitory computer-readable recording medium having recorded thereon a computer program for performing an image-based animal weight measurement method according to any one of Articles 7 to 12.

14. In the image-based animal weight measurement device, A communication module for receiving images from at least one camera that photographs an object; Memory in which the animal weight measurement program is stored; and A processor that executes a program stored in the above memory, The above animal weight measurement program inputs the received image into an animal detection model to extract information on a plurality of feature points of the animal, the horizontal and vertical lengths of the animal, and information on a bounding box formed to fit the animal, and inputs the horizontal length of the animal, the vertical length of the animal, the plurality of feature point information, information on the bounding box, and the installation height of the camera that captured the image into a weight estimation model to output the weight of the animal. The above characteristic point information includes the end position of the head of the animal (nose), the neck position (neck), the first back position (back1) close to the head divided into three parts, the second back position (back2) in the middle, the third back position (back3) close to the tail, the right shoulder position (right shoulder), the left shoulder position (left shoulder), the front armpit position (front armpit), or the end position of the torso (hip). Information about the above bounding box includes the center point coordinates (Xc, Yc) of the bounding box, the width (W) of the bounding box, the length (H) of the bounding box, and the angle (theta) by which the bounding box is rotated about the reference axis. An image-based animal weight measurement device, wherein the weight estimation model is learned to estimate the weight of the animal based on the horizontal length of the animal, the vertical length of the animal, information on the plurality of feature points, information on the bounding box, and the installation height of the camera that captured the image.

15. In paragraph 14, The above weight estimation model is, For each of a plurality of animal objects, sequence data grouped into a predetermined sequence length based on a plurality of frames including the horizontal length of the animal object, the vertical length of the animal object, information on a plurality of feature points, information on a bounding box, and camera installation height, and An image-based animal weight measurement device constructed using learning data including average weight information obtained from a movement record report as the correct weight value of each animal individual corresponding to the above sequence data.

Images