KR102578113B1 - 3-dimention object shape acquisition system and method - Google Patents

Abstract

The 3D object shape acquisition system includes an upper sensor located on the top of the 3D object and generating a top image by photographing the top of the 3D object, located on the left side of the 3D object, and the left side of the 3D object. A left sensor that captures and generates a left image, a right sensor located on the right side of the three-dimensional object and creates a right image by photographing the right side of the three-dimensional object, the position and overlap of the three-dimensional object from the upper image A control unit that calculates and sequentially generates a plurality of image acquisition signals from a starting point based on the degree of overlap, and is triggered according to the plurality of image acquisition signals to generate a plurality of image acquisition signals from the upper image, the left image, and the right image. It includes a restoration server that stores target frames and generates a 3D biometric shape for the 3D object from the plurality of target frames.

Description

3D object shape acquisition system and method {3-DIMENTION OBJECT SHAPE ACQUISITION SYSTEM AND METHOD}

The present invention relates to a system and method for obtaining a three-dimensional object shape.

In smart livestock farming or companion animal management, there is a need for technology to determine the 3D biometric shape of livestock.

In order to restore the 3D biomorphic shape of livestock, the biomorphic shape of livestock is acquired by confining the livestock in a frame or weighbridge and obtaining biometric images from a plurality of fixed multi-view points, for example, 4 viewpoints, and thereby restoring the biomorphic shape, thereby reducing the unnatural posture of the livestock. Due to the insufficient number of images, the completeness of the 3D biological shape was lacking.

In addition, since filming takes place while livestock are confined, improvements are needed in terms of animal welfare for livestock.

The purpose of the present invention is to solve the above problems, and is a three-dimensional object shape acquisition system that allows acquiring a three-dimensional biological shape of livestock based on image data generated by the livestock passing through a plurality of sensors. and method are provided.

In order to achieve the above object, according to an embodiment of the present invention, a three-dimensional object shape acquisition system includes an upper sensor located on the upper part of a three-dimensional object and generating an upper image by photographing the upper part of the three-dimensional object; A left sensor located on the left side of the 3D object and generating a left image by photographing the left side of the 3D object. A left sensor located on the right side of the 3D object and generating a right image by photographing the right side of the 3D object. A right sensor, a control unit that calculates the position and overlap of the three-dimensional object from the upper image, and sequentially generates a plurality of image acquisition signals from a starting point based on the overlap, and according to the plurality of image acquisition signals It includes a restoration server that is triggered to store a plurality of target frames from the upper image, the left image, and the right image, and generates a 3D biometric shape for the 3D object from the plurality of target frames.

The height of the left sensor from the ground is the same as the height of the right sensor from the ground, and the sensor central axis of the upper sensor may be perpendicular to each of the sensor central axis of the left sensor and the sensor central axis of the right sensor.

The control unit detects the 3D object from image frames included in the upper image, detects a bounding box from the frame in which the 3D object is detected, and uses a learned livestock detection neural network to detect the 3D object. The center point of one end of the 3D object can be calculated from the bounding box.

The control unit sets the time point at which one end of the 3D object is first detected from the time it starts detecting the 3D object from the upper image as the start point, and generates a first image acquisition signal corresponding to the start point. It can be transmitted to the restoration server.

The control unit, after the start time, generates the nth image acquisition signal at a time when the degree of overlap between the nth image frame of the upper image and the n-1th image frame corresponds to a predetermined reference overlap degree or more. It is transmitted to the restoration server, and n may be a natural number of 2 or more.

The restoration server receives a plurality of image acquisition signals from the control unit, and the upper target frame and left target corresponding to the time of receiving each of the plurality of image acquisition signals from the upper image, the left image, and the right image. and a storage unit for extracting and storing a frame and a right target frame, wherein the plurality of target frames include the upper target frame, the left target frame, and the right target frame corresponding to each of the plurality of image acquisition signals. can do.

The restoration server converts and registers a plurality of overlapping first target frames among the plurality of target frames into a point cloud based on the plurality of target frames, and registers feature points from each of the plurality of first target frames. It may further include a shape restoration unit that generates the three-dimensional biological shape by matching the relationships between the two.

According to another embodiment of the present invention, the method for obtaining the shape of a 3D object includes a control unit located on top of a 3D object and receiving the upper image from an upper sensor that captures the upper part of the 3D object and generates the upper image. A step of calculating, by the control unit, the position and degree of overlap of the three-dimensional object from the upper image, by the control part, sequentially generating a plurality of image acquisition signals from a starting point based on the degree of overlap, A restoration server that receives the plurality of image acquisition signals from the control unit is triggered according to the plurality of image acquisition signals to generate the upper image, the left image generated from the left sensor located on the left side of the three-dimensional object, and the 3 Storing a plurality of target frames from a right image generated from a right sensor located on the right side of the 3D object, and generating, by the restoration server, a 3D biometric shape for the 3D object from the plurality of target frames. Includes.

Calculating the position and degree of overlap of the 3D object includes detecting the 3D object from image frames included in the upper image, and creating a bounding box from the frame in which the 3D object is detected. It may include a step of detecting, and a step of calculating a central point of a set of the three-dimensional object from the bounding box using a learned livestock detection neural network.

In the step of sequentially generating the plurality of image acquisition signals, the start time is the time when one end of the three-dimensional object is first detected from the time when the three-dimensional object is started to be detected from the upper image, and the start time is It may include generating a first image acquisition signal corresponding to and transmitting it to the restoration server.

The step of sequentially generating the plurality of image acquisition signals includes, after the starting point, a point in time when the degree of overlap between the nth image frame of the upper image and the n-1th image frame corresponds to a predetermined standard overlap degree or more. and generating an nth image acquisition signal and transmitting it to the restoration server, where n may be a natural number of 2 or more.

The step of storing the plurality of target frames includes receiving a plurality of image acquisition signals from the control unit, and corresponding to a time point of receiving each of the plurality of image acquisition signals from the upper image, the left image, and the right image. Extracting and storing an upper target frame, a left target frame, and a right target frame, wherein the plurality of target frames include the upper target frame, the left target frame, and It may include the right target frame.

The step of generating a 3D biometric shape for the 3D object includes converting a plurality of overlapping first target frames among the plurality of target frames into a point cloud based on the plurality of target frames and registering them. It may include generating the three-dimensional biometric shape by matching relationships between feature points from each of the plurality of first target frames.

The present invention acquires continuous biometric images generated as an object moves between sensors, and restores a three-dimensional biometric shape of the object based on the acquired images, thereby reducing defects in biometric shape restoration and providing high-quality, natural-looking images. It is possible to obtain the shape of a living body.

According to the present invention, precision and real-time can be improved by calculating the acquisition time points of multiple biological images that overlap between images by more than a predetermined standard, even without using a different type of sensor.

According to the present invention, it can be effectively used in smart livestock farming or companion animal management, such as managing precise specifications of livestock, determining shipment timing, and monitoring companion animal health and welfare.

According to the present invention, a three-dimensional living body shape can be obtained while allowing an object to move naturally, so the posture of the livestock is natural, the three-dimensional shape is natural by using continuous image extraction, and animal welfare for livestock can be improved. .

Figure 1 is a block diagram schematically showing a 3D livestock biometric image acquisition system according to an embodiment of the present invention.
Figure 2 is an example diagram for explaining the location where a pair of image sensors is installed.
Figure 3 is an example diagram for explaining the distance between an image sensor pair and a 3D object 10.
Figure 4 is an example diagram for explaining the bounding box of the n-1th frame of the upper image and the center point of the bounding box.
Figure 5 is an example diagram for explaining the bounding box of the nth frame of the upper image and the center point of the bounding box.
Figure 6 is a flowchart of a method for acquiring a 3D object biometric shape according to an embodiment of the present invention.

The present invention can be implemented with various changes without departing from the spirit, and can have one or more embodiments. In addition, in the present invention, the embodiments described in “specific details for carrying out the invention” and “drawings” are examples for specifically explaining the present invention, and do not limit or limit the scope of the present invention.

Accordingly, what a person skilled in the art of the present invention can easily infer from the “specific details for carrying out the invention” and “drawings” of the present invention shall be interpreted as falling within the scope of the present invention. can do.

In addition, the size and shape of each component shown in the drawings may be exaggerated for description of the embodiment, and do not limit the size and shape of the invention in actual practice.

Unless the terms used in the specification of the present invention are specifically defined, they may have the same meaning as those generally understood by those skilled in the art to which the present invention pertains.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

Figure 1 is a block diagram schematically showing a 3D livestock biometric image acquisition system according to an embodiment of the present invention.

Referring to FIG. 1, a 3D object biometric shape acquisition system (hereinafter, “system”) 1 for a 3D object 10 includes a sensor unit 100, a control unit 200, and a restoration server 300. It can be included.

In this specification, the three-dimensional object 10 is described assuming that it is an animal (e.g., livestock) that can move autonomously, but the invention is not limited thereto, and the three-dimensional object 10 is a movable object or It could be a living thing.

The three-dimensional object 10 can pass between the sensor units 100, and the system 1 can obtain a three-dimensional body shape based on an image generated by photographing the three-dimensional object 10.

The sensor unit 100 may include a plurality of image sensors that generate image data by photographing one side of the external shape of the three-dimensional object 10 .

The sensor unit 100 may include an upper sensor 110, a left sensor 120, and a right sensor 130. Each of the upper sensor 110, left sensor 120, and right sensor 130 may be a camera module that captures RGB and/or RGB-D images. The upper sensor 110, left sensor 120, and right sensor 130 may generate a two-dimensional image and/or a three-dimensional image for the three-dimensional object 10.

The upper sensor 110 is located at the top of the 3D object 10 and may be installed in a direction looking at the 3D object 10 from the top of the 3D object 10. The upper sensor 110 may capture the upper part of the three-dimensional object 10 and generate two-dimensional and/or three-dimensional image data (hereinafter, “upper image”).

The left sensor 120 is located on the left side of the 3D object 10 and may be installed in a direction looking at the 3D object 10 from the left side of the 3D object 10. The left sensor 120 may generate two-dimensional and/or three-dimensional image data (hereinafter, “left image”) by photographing the upper left side of the three-dimensional object 10.

The right sensor 130 is located on the right side of the 3D object 10 and may be installed in a direction looking at the 3D object 10 from the right side of the 3D object 10. The right sensor 130 may generate two-dimensional and/or three-dimensional image data (hereinafter, “right image”) by photographing the upper right side of the three-dimensional object 10.

The control unit 200 may be included in an embedded module, and the restoration server 300 may be a cloud server.

The control unit 200 may include an object detection unit 210, an object position and overlap calculation unit 220, and an image acquisition signal generator 230.

The object detector 210 may detect the three-dimensional object 10 from the upper image received from the upper sensor 110. When the object detector 210 detects at least one of one end and the other end of the three-dimensional object 10 from the image frames included in the upper image, a bounding box containing the detected part from each of the detected image frames is generated. ) can be detected.

Here, when the 3D object 10 is a cow, one end of the 3D object 10 may represent the head of the livestock, and the other end of the 3D object 10 may represent the buttocks of the livestock. The method by which the object detector 210 detects the position and size of the bounding box from each of the image frames included in the upper image may be performed according to a known object recognition method.

The object position and overlap calculation unit 220 may calculate the center point and overlap of a set of three-dimensional objects 10 from each of the image frames included in the upper image using a learned livestock detection neural network. The object position and degree of overlap calculation unit 220 may store the center point and degree of overlap of a set of three-dimensional objects 10 in a database. For example, the object position and overlap calculation unit 220 may extract a position representing the center point of the head of the livestock from the bounding box. Hereinafter, for convenience of explanation, one end of the three-dimensional object 10 will be described as a head of livestock.

The learned livestock detection neural network may be a neural network that has learned a method of extracting the position of the livestock's head from the upper image of the livestock.

The image acquisition signal generator 230 may generate a plurality of image acquisition signals based on the center point and degree of overlap of the bounding box and transmit them to the restoration server 300.

The restoration server 300 may include a storage unit 310 and a 3D biomorphic shape restoration unit 320.

The storage unit 310 uses each of the plurality of image acquisition signals received from the image acquisition signal generator 230 as a trigger signal to extract and store a plurality of target frames from the upper image, the left image, and the right image, 3 It can be transmitted to the dimensional biomorphic shape restoration unit 320.

The 3D biological shape restoration unit 320 may receive a plurality of target frames and generate a 3D biological shape for the 3D object 10.

Hereinafter, for convenience of explanation, the upper sensor 110, the left sensor 120, and the right sensor 130 are defined as one “image sensor pair.”

Hereinafter, the image sensor pair will be described with reference to FIG. 2.

Figure 2 is an example diagram for explaining the location where a pair of image sensors is installed.

Referring to FIG. 2, a pair of image sensors may be installed on a frame installed on the ground. The height at which the left sensor 120 is installed from the ground is the same as the height at which the right sensor 130 is installed from the ground. The three-dimensional object 10 can move through the frame. The image sensor pair can photograph a 3D object 10 while the 3D object 10 moves through the frame.

The control unit 200 may store data representing the size and position of a virtual minimum cuboid (hereinafter, “minimum cuboid”) including the shape of the three-dimensional object 10 in a database. The minimum rectangular parallelepiped represents the living body size for determining the size of the three-dimensional object 10, and is predetermined as initial information based on the average body posture of livestock raised on the farm where the system 1 is used or the average body posture data for each species. It can happen.

Although FIG. 2 shows one image sensor pair installed on a frame installed on the ground, the invention is not limited thereto, and the sensor unit 100 may include at least one image sensor pair. For example, when the size of the three-dimensional object 10 is less than a predetermined standard, the sensor unit 100 includes one image sensor pair, and when the size of the three-dimensional object 10 is greater than a predetermined standard, the , the sensor unit 100 may include two or more image sensor pairs.

Referring to FIG. 3, the relationship between the image sensor pair and the minimum cuboid of the three-dimensional object 10 will be described.

Figure 3 is an example diagram for explaining the distance between an image sensor pair and a 3D object 10.

Referring to FIG. 3 , when the 3D object 10 is a cow, the smallest cuboid (CUB) corresponding to the cow may include the shape of the 3D object 10. Each pair of image sensors may be positioned at the same vertical distance from a corresponding surface of a minimal cuboid (CUB). For example, the left sensor 120 is installed at a vertical distance (R) away from one side of the minimum cuboid (CUB), and the right sensor 130 is installed perpendicularly from the other side facing one side of the minimum cuboid (CUB). It can be installed at a distance (R) away.

In addition, the sensor central axis of the upper sensor 110 is perpendicular to the sensor central axis of the left sensor 120 and the sensor central axis of the right sensor 130, respectively.

Hereinafter, the operation of the control unit 200 will be described with reference to FIGS. 4 and 5.

Figure 4 is an example diagram for explaining the bounding box of the n-1th frame of the upper image and the center point of the bounding box.

Hereinafter, n is a natural number of 2 or more.

Referring to FIG. 4, the object detector 210 may detect the bounding box BB_n-1 from the n-1th frame 111_n-1 of the upper image. The object detector 210 may recognize the location and size of the bounding box BB_n-1.

Since the n-1th frame (111_n-1) does not include the entire shape of the three-dimensional object 10, the object detector 210 detects the size of the bounding box from the n-1th frame (111_n-1). The length (L ₀ ) cannot be detected.

Therefore, when the entire shape of the 3D object 10 is not included, the object detection unit 210 determines the length (L _a ) of the bounding box based on the size of the minimum rectangular parallelepiped previously stored in the database for the 3D object 10. It can be decided based on initial information. For example, the object detector 210 may use the length (L _a ) of the bounding box in the n-1th frame (111_n-1) as the side (L _a ) indicating the length of the object among the minimum cuboids (CUB) in FIG. 3 ) can be determined by the length.

The object position and overlap calculation unit 220 calculates the position (L ^H _n-1 ) of the livestock head corresponding to the n-1th frame from the bounding box (BB_n-1) using the learned livestock detection neural network. You can.

Figure 5 is an example diagram for explaining the bounding box of the nth frame of the upper image and the center point of the bounding box.

Referring to FIG. 5 , the object detector 210 may detect the bounding box BB_n from the nth frame 111_n of the upper image. The object detector 210 may recognize the location and size of the bounding box BB_n.

Since the nth frame (111_n) includes the entire shape of the three-dimensional object 10, the object detector 210 determines the length (L ₀ ) of the bounding box among the sizes of the bounding box (BB_n) from the nth frame (111_n). It can be detected.

The object position and overlap calculation unit 220 may calculate the position (L ^H _n ) of the livestock head corresponding to the nth frame from the bounding box (BB_n) using the learned livestock detection neural network.

Hereinafter, for convenience of explanation, both the length of the bounding box (L _a ) determined based on the size of the minimum cuboid (CUB) and the length of the bounding box (L ₀ ) detected from the upper image are expressed as the length of the bounding box (L ₀ ). Explain.

Referring to Figures 4 and 5, the object position and overlap calculation unit 220 calculates the position of the livestock head (L ^H _n-1 ), the position of the livestock head (L ^H _n ), and the length of the bounding box (L ₀ ) Based on this, the degree of overlap corresponding to the nth frame can be calculated.

The object position and overlap calculation unit 220 may calculate the overlap degree corresponding to the nth frame according to [Equation 1] below.

[Equation 1]

here is the degree of overlap, represents the length of the bounding box of the nth frame. for example, may be a length representing the length from the head to the buttocks of the livestock among the bounding boxes of the nth frame.

The image acquisition signal generator 230 may generate an image acquisition signal based on the degree of overlap of image frames included in the upper image.

According to the above-mentioned [Equation 1], the distance moved by the 3D object 10 included in the corresponding bounding box during the time interval of the image frames included in the upper image can be expressed as [Equation 2] below. there is.

[Equation 2]

here may represent the distance that the 3D object 10 moves from the previous image frame to the current image frame.

Based on the above-described [Equation 1] and [Equation 2], the distance moved by the three-dimensional object 10 can be expressed as [Equation 3] below.

From the point in time when the object detector 210 starts detecting the 3D object from the frames of the upper image, the point in time when one end of the 3D object 10 is first detected can be set as the starting point. The image acquisition signal generator 230 may generate a first image acquisition signal corresponding to the starting point and transmit it to the restoration server 300.

The image acquisition signal generator 230 calculates the distance that the three-dimensional object 10 has moved from the start point to the previous point among the frames of the upper image ( ) The point in time that satisfies [Equation 4] below is set as the n-th point in time, and the n-th image acquisition signal corresponding to the n-th point in time can be generated and transmitted to the restoration server 300.

[Equation 4]

here is the standard overlap degree, which is the standard for the degree of overlap. Standard overlap ( ) The value may be previously stored in the control unit 200 as initial information.

For example, the image acquisition signal generator 230 determines the difference between the position of the livestock head (L ^H ₁ ) corresponding to the first image acquisition signal and the position of the livestock head (L ^H ₂ ) corresponding to the second image acquisition signal. the length of the bounding box of the second frame ( The value divided by ) is the standard overlap ( ), a second image acquisition signal corresponding to the second time point may be generated.

The control unit 200 may repeat the above-described process until the object detection unit 210 fails to detect the 3D object 10 from the upper image.

The storage unit 310 may extract a plurality of target frames from the upper image, the left image, and the right image by using each of the plurality of image acquisition signals as a trigger signal. Here, the target frame may include an upper target frame, a left target frame, and a right target frame. For example, based on the time of receiving the second image acquisition signal, the storage unit 310 extracts the upper target frame from the upper image, extracts the left target frame from the left image, and extracts the right target frame from the right image. You can save it.

Hereinafter, a plurality of target frames corresponding to one image acquisition signal are referred to as a “target frame pair.” The storage unit 310 may store pairs of target frames corresponding to each of the plurality of image acquisition signals and transmit them to the 3D biomorphic shape restoration unit 320. The point in time at which the storage unit 310 receives each of the plurality of image acquisition signals may be at any point in the period from the time in which each of the plurality of image acquisition signals is received to the time in which the target frame pair is extracted.

The 3D biometric shape restoration unit 320 may restore the 3D biometric shape of the 3D object 10 by inputting pairs of overlapping target frames into a Structure from Motion (SfM) algorithm or another known algorithm.

For example, the 3D biomorphic shape restoration unit 320 converts and registers overlapping image frames among target frame pairs corresponding to each of a plurality of image acquisition signals into a point cloud, and registers each of the overlapping image frames. A 3D biometric shape can be created by matching the relationships between feature points.

The system 1 generates a plurality of image acquisition signals and repeats the above-described process to improve the quality of the three-dimensional biomorphic shape and determines the reference overlap ( ) can be optimized.

As described above, the restoration server 300 receives image data for a series of overlapping three-dimensional objects from the control unit 200 included in the embedded system. Additionally, the 3D biometric shape restoration unit 320 can be implemented in a separate computing device and/or server, such as a cloud server or an external computer, to perform 3D shape restoration in real time or non-real time.

Hereinafter, a method for acquiring a 3D object biometric shape using the system 1 will be described with reference to FIG. 6.

Figure 6 is a flowchart of a method for acquiring a 3D object biometric shape according to an embodiment of the present invention.

Hereinafter, descriptions that overlap with the above description may be omitted.

The control unit 200 may predetermine n=0 as the initial value (S100).

The object detector 210 may receive an upper image from the upper sensor 110 (S200).

The object detection unit 210 may determine whether the 3D object 10 is detected from the upper image (S300).

In step S300, when the 3D object 10 is detected, the object detection unit 210 detects a bounding box from the frame in which the 3D object 10 is detected, and the object position and overlap calculation unit 220 detects the livestock The position of the head (L ^H _n ) can be calculated (S400).

The object detection unit 210 may determine whether n=0 (S500).

If n=0 in step S500, the object detection unit 210 changes n=1 (S600) and performs the process again from step S200. If n = 0 in step S500, the detected frame is the first image frame corresponding to the starting point, so the object position and overlap calculation unit 220 cannot calculate the overlap degree because there is no previous frame.

If n = 0 in step S500, the object position and overlap calculation unit 220 calculates the distance that the three-dimensional object 10 has moved ( ) can be derived as in [Equation 2] described above (S700)

The image acquisition signal generator 230 may determine whether the above-described [Equation 4] is met (S800).

In step S800, if [Equation 4] is satisfied, the image acquisition signal generator 230 generates an image acquisition signal and transmits it to the restoration server 300, and the storage unit 310 generates the upper image, the left image, And the target frame pair can be extracted and stored from the right image (S900), and the n value can be increased by one (S1000).

In step S800, if [Equation 4] is not satisfied, or following step S1000, the object detection unit 210 may perform the operation again from step S200.

In step S300, if the 3D object 10 is not detected and n=0 (S1100), the 3D object 10 has not yet entered the frame of the upper image, so the object detector 210 starts again from step S200. It can be done.

In step S300, if the 3D object 10 is not detected and n=0 (S1100), the 3D object 10 has passed through the frame of the upper image and is out of the frame, and the step can be ended.

Meanwhile, the disclosed embodiments may be implemented in the form of a recording medium that stores instructions executable by a computer. Instructions may be stored in the form of program code, and when executed by a processor, may create program modules to perform operations of the disclosed embodiments. The recording medium may be implemented as a computer-readable recording medium.

Computer-readable recording media include all types of recording media storing instructions that can be decoded by a computer. For example, there may be read only memory (ROM), random access memory (RAM), magnetic tape, magnetic disk, flash memory, optical data storage device, etc.

Although the embodiments of the present invention have been described above, the present invention is not limited to the above embodiments, and may be varied within the scope of the detailed description and accompanying drawings, as long as it does not deviate from the spirit of the present invention and does not impair the effect. It can be implemented by changing it accordingly. It is also natural that such embodiments fall within the scope of the present invention.

1: 3D object biometric shape acquisition system
10: 3D object
100: sensor unit
110: upper sensor
120: Left sensor
130: Right sensor
200: control unit
210: object detection unit
220: Object position and overlap calculation unit
230: Image acquisition signal generator
300: Restore server
310: storage unit
320: 3D biomorphic shape restoration unit

Claims (13)

An upper sensor located on top of a three-dimensional object and generating an upper image by photographing the upper part of the three-dimensional object;
A left sensor located on the left side of the 3D object and generating a left image by photographing the left side of the 3D object;
A right sensor located on the right side of the 3D object and generating a right image by photographing the right side of the 3D object;
a control unit that calculates the position and degree of overlap of the three-dimensional object from the upper image and sequentially generates a plurality of image acquisition signals from a starting point based on the degree of overlap; and
Triggered according to the plurality of image acquisition signals to store a plurality of target frames from the upper image, the left image, and the right image, and to generate a 3D biometric shape for the 3D object from the plurality of target frames Includes a restoration server;
The object position and overlap calculation unit included in the control unit,
Calculate the degree of overlap corresponding to the nth frame according to Equation 1 below,
[Equation 1]

here is the overlap degree, L ^H _n-1 represents the position of the livestock head in the n-1th frame, L ^H _n represents the position of the livestock head in the n-th frame, L ₀ represents the length of the bounding box of the n-th frame,
The image acquisition signal generator included in the control unit,
Generating an image acquisition signal based on the degree of overlap of image frames included in the upper image,
Calculate the distance moved by the 3D object included in the corresponding bounding box during the time interval of the image frames included in the upper image using Equation 2 below,
[Equation 2]

Here, ΔL represents the distance the 3D object moved from the previous image frame to the current image frame, L ^H _n-1 is the location of the livestock head in the n-1th frame, and L ^H _n is the livestock head in the nth frame. is the location of
The control unit,
After the start time, at the point when the degree of overlap of the nth video frame with the n-1th video frame of the upper video corresponds to a predetermined standard overlap degree or more, the nth image acquisition signal is generated and transmitted to the restoration server. And
A three-dimensional object shape acquisition system, wherein n is a natural number of 2 or more. According to paragraph 1,
The height of the left sensor from the ground is the same as the height of the right sensor from the ground, and the sensor central axis of the upper sensor is orthogonal to each of the sensor central axis of the left sensor and the sensor central axis of the right sensor.
3D object shape acquisition system. According to paragraph 1,
The control unit,
The three-dimensional object is detected from image frames included in the upper image, a bounding box is detected from the frame in which the three-dimensional object is detected, and the bounding box is detected from the bounding box using a learned livestock detection neural network. Calculating the center point of a set of three-dimensional objects,
3D object shape acquisition system. According to paragraph 3,
The control unit,
Starting from the time when detecting the 3D object from the upper image, the time when one end of the 3D object is first detected is set as the start time, and the first image acquisition signal corresponding to the start time is generated to generate the restoration server. transmitted to,
3D object shape acquisition system. delete According to paragraph 4,
The restoration server is,
Receive a plurality of image acquisition signals from the control unit, and receive an upper target frame, a left target frame, and a right target corresponding to a point in time at which each of the plurality of image acquisition signals is received from the upper image, the left image, and the right image. Includes a storage unit for extracting and storing frames,
The plurality of target frames are:
Containing the upper target frame, the left target frame, and the right target frame corresponding to each of the plurality of image acquisition signals,
3D object shape acquisition system. According to clause 6,
The restoration server is,
Based on the plurality of target frames, a plurality of overlapping first target frames among the plurality of target frames are converted into point clouds and registered, and the relationships between feature points from each of the plurality of first target frames are matched to each other. Further comprising a shape restoration unit that generates the three-dimensional biological shape,
3D object shape acquisition system. Receiving, by a control unit, the upper image from an upper sensor located on the upper part of the three-dimensional object and generating the upper image by photographing the upper part of the three-dimensional object;
Calculating, by the control unit, the position and overlap of the three-dimensional object from the upper image;
The control unit sequentially generating a plurality of image acquisition signals from a starting point based on the degree of overlap;
A restoration server that receives the plurality of image acquisition signals from the control unit is triggered according to the plurality of image acquisition signals to generate the upper image, the left image generated from the left sensor located on the left side of the three-dimensional object, and the 3 Storing a plurality of target frames from a right image generated from a right sensor located on the right side of the dimensional object; and
A step of generating, by the restoration server, a 3D biometric shape for the 3D object from the plurality of target frames,
The object position and overlap calculation unit included in the control unit,
Calculate the degree of overlap corresponding to the nth frame according to Equation 1 below,
[Equation 1]

here is the overlap degree, L ^H _n-1 represents the position of the livestock head in the n-1th frame, L ^H _n represents the position of the livestock head in the n-th frame, L ₀ represents the length of the bounding box of the n-th frame,
The image acquisition signal generator included in the control unit,
Generating an image acquisition signal based on the degree of overlap of image frames included in the upper image,
Calculate the distance moved by the 3D object included in the corresponding bounding box during the time interval of the image frames included in the upper image using Equation 2 below,
[Equation 2]

Here, ΔL represents the distance the 3D object moved from the previous image frame to the current image frame, L ^H _n-1 is the location of the livestock head in the n-1th frame, and L ^H _n is the livestock head in the nth frame. is the location of
After the start time, at the point when the degree of overlap of the nth video frame with the n-1th video frame of the upper video corresponds to a predetermined standard overlap degree or more, the nth image acquisition signal is generated and transmitted to the restoration server. It further includes steps of:
where n is a natural number of 2 or more,
Method for acquiring 3D object shape. According to clause 8,
The step of calculating the position and overlap of the 3D object is,
detecting the three-dimensional object from image frames included in the upper image;
detecting a bounding box from the frame in which the 3D object is detected; and
Comprising the step of calculating a set of center points of the three-dimensional object from the bounding box using a learned livestock detection neural network,
Method for acquiring 3D object shape. According to clause 8,
The step of sequentially generating the plurality of image acquisition signals is,
Starting from the time when detecting the 3D object from the upper image, the time when one end of the 3D object is first detected is set as the start time, and the first image acquisition signal corresponding to the start time is generated to generate the restoration server. Including the step of transmitting to,
Method for acquiring 3D object shape. delete According to clause 8,
The step of storing the plurality of target frames includes:
Receive a plurality of image acquisition signals from the control unit, and receive an upper target frame, a left target frame, and a right target corresponding to a point in time at which each of the plurality of image acquisition signals is received from the upper image, the left image, and the right image. Including the step of extracting and storing a frame,
The plurality of target frames are:
Containing the upper target frame, the left target frame, and the right target frame corresponding to each of the plurality of image acquisition signals,
Method for acquiring 3D object shape. According to clause 8,
The step of generating a 3D biometric shape for the 3D object is:
Converting and registering a plurality of overlapping first target frames among the plurality of target frames into a point cloud based on the plurality of target frames; and
Comprising the step of generating the three-dimensional biometric shape by matching relationships between feature points from each of the plurality of first target frames.
Method for acquiring 3D object shape.