KR20210128943A - Apparatus and method for comparing and correcting sports posture using neural network - Google Patents

Abstract

In a method for correcting a sports posture using an artificial neural network, performed by a server according to an embodiment of the present invention, (a) extracting posture information of a user through a motion recognition model from an exercise video received from a user terminal to do; (b) retrieving pre-stored correction information corresponding to the motion video, and calculating an error value for each time by comparing the posture information and the correction information; and (c) generating correction information for an error value for each time and providing the correction information to the user terminal, wherein the posture information calculates the angle and length of at least one specific body part from the user included in the exercise video for each time. information, the correction information is posture information calculated by an expert associated with an exercise video, and the correction information refers to information for correcting a user's posture.

Description

Apparatus and method for comparing and correcting sports posture using artificial neural network {APPARATUS AND METHOD FOR COMPARING AND CORRECTING SPORTS POSTURE USING NEURAL NETWORK}

The present invention analyzes or corrects the exercise posture of an athlete participating in sports using a pose estimation model, which is one of the artificial neural network models, and further calculates the result for correction after comparing with the exercise posture of professional players It is about a system for

Conventionally, various artificial intelligence services that help users exercise are provided. However, most of the services are related to home training, and rather than correcting and analyzing the user's exercise posture, they are only serving as a helper who helps simple counting of the number of times.

For example, golf is a representative field that actually needs to analyze or correct a user's posture other than home training. Barriers to entry are high.

On the other hand, pose estimation, a field of computer vision, is an artificial intelligence field that estimates and detects the position and orientation of an object, and a person who becomes a key point It is a technology that analyzes movement by detecting the joint position of However, since there is no exercise assistance program to which the pose estimation technology is applied, it is not being used by the general public.

An object of the present invention is to lower the entry barrier by expanding the applications that can correct the posture of the user performing the exercise by using the pose estimate model, which is an artificial neural network technology, in order to solve the above problems.

In addition, it aims to help users take one step closer to the athlete they admire by comparing and analyzing the posture of the athlete they want to refer to.

In a method for correcting a sports posture using an artificial neural network, performed by a server, according to an embodiment of the present invention for achieving the above technical problem, (a) motion from an exercise video received from a user terminal extracting the user's posture information through the recognition model; (b) retrieving pre-stored correction information corresponding to the motion video, and calculating an error value for each time by comparing the posture information and the correction information; and (c) generating correction information for an error value for each time and providing the correction information to the user terminal, wherein the posture information calculates the angle and length of at least one specific body part from the user included in the exercise video for each time. information, the correction information may be posture information calculated by an expert associated with an exercise video, and the correction information may refer to information for correcting a user's posture.

Also, a pose estimation model may be applied to the motion recognition model.

In addition, the motion recognition model comprises the steps of (a1) recognizing a user from a motion video, and calculating a key point corresponding to a specific body part of the user; (a2) generating a coordinate value for each key point; and (a3) calculating changes in angles and lengths with respect to the plurality of coordinate values in units of time to generate the user's posture information.

In addition, in step (a1), when two or more people are recognized in the exercise video, different identifiers are assigned to each person, and in step (a3), posture information of the person corresponding to the different identifiers may be calculated.

In addition, in step (a), the motion recognition model recognizes the user included in the exercise video, generates a bounding box, and estimates the user's pose in the bounding box in a preset time unit, and posture information can be calculated.

In addition, before step (b), a list of experts associated with the exercise video is delivered to the user terminal, and when the user terminal selects a specific expert, step (b) is mapped with a specific expert and compares pre-stored correction information with posture information. can

In addition, the step (b) comprises the steps of: (b1) dividing the posture information by a preset time unit, and mapping the correction information corresponding to the exercise posture including each section; (b2) matching key points for the same part of a person included in each of the posture information and the correction information; and (b3) calculating an error value based on the coordinate values of each of the key points matched with each other in the same region.

In addition, if the posture information consists of information about a plurality of people, step (b2) selects a person with the highest similarity to the location and direction of the key point of the expert included in the long-term correction information to match the key point. .

In addition, in step (b3), the coordinate value of the user's key point matched in the same way as the expert may be substituted into a preset formula and designated as an error value.

In addition, the preset formula calculates the inclination and angle of a straight line generated by connecting key points of the user and the expert, and calculates as an error value whether the postures of the user and the expert match.

In addition, in step (c), based on the error value, correction information including at least one of text, image, and sound for correcting the posture of the user to be the same as that of the expert in units of time may be generated.

In addition, the correction information includes the score for each exercise posture calculated through the error value, image information comparing the previously analyzed exercise video with the exercise video received in step (a), and advice information for correcting the user's exercise posture It may consist of at least one of

In addition, after step (c), it is possible to update the deep learning of the motion recognition model based on the posture information and the correction information.

In addition, in the apparatus for correcting a sports posture using an artificial neural network, a memory in which a program for correcting a sports posture using an artificial neural network is stored and a processor for correcting a sports posture using an artificial neural network by executing the program stored in the memory Including, the processor extracts the user's posture information through the motion recognition model from the exercise video received from the user terminal, searches for pre-stored correction information corresponding to the exercise video, and compares the posture information with the correction information to obtain an error value for each time calculates, generates correction information for the error value for each time, and provides it to the user terminal, and the posture information is information obtained by calculating the angle and length of at least one specific body part from the user included in the exercise video by time, and correction The information may be posture information calculated by an expert associated with an exercise video, and the correction information may be information for correcting a user's posture.

In addition, it may be a computer-readable storage medium in which a program for performing a method of correcting a sports posture using an artificial neural network according to claim 1 is recorded.

By using the pose estimate model, which is an artificial neural network technology, according to an embodiment of the present invention, it is possible to lower the entry barrier by expanding the applications that can correct the posture of the user performing the exercise.

Also, by comparing and analyzing the postures of athletes that users want to reference, it is possible to help them take one step closer to the athletes they admire.

1 is a diagram showing the configuration of a system for correcting a sports posture using an artificial neural network, according to an embodiment of the present invention.
2 is a diagram showing the configuration of a server for correcting a sports posture using an artificial neural network, according to an embodiment of the present invention.
3 is an operation flowchart illustrating a method of correcting a sports posture using an artificial neural network, according to an embodiment of the present invention.
4 is an operation flowchart illustrating a method of recognizing a user's posture through an artificial neural network, according to an embodiment of the present invention.
5A to 5C are exemplary diagrams for explaining a process of recognizing a user's posture through an artificial neural network, according to an embodiment of the present invention.
6 is an operation flowchart illustrating a method for calculating an error value according to an embodiment of the present invention.
7A to 7B are exemplary views for explaining a process of calculating an error value according to an embodiment of the present invention.
8A to 8C are diagrams illustrating a method of providing correction information according to an embodiment of the present invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those of ordinary skill in the art can easily implement them. However, the present invention may be embodied in many different forms and is not limited to the embodiments described herein. And in order to clearly explain the present invention in the drawings, parts irrelevant to the description are omitted, and similar reference numerals are attached to similar parts throughout the specification.

Throughout the specification, when a part is "connected" with another part, this includes not only the case of being "directly connected" but also the case of being "electrically connected" with another element interposed therebetween. . Also, when a part "includes" a component, it means that other components may be further included, rather than excluding other components, unless otherwise stated, and one or more other features However, it is to be understood that the existence or addition of numbers, steps, operations, components, parts, or combinations thereof is not precluded in advance.

The following examples are detailed descriptions to help the understanding of the present invention, and do not limit the scope of the present invention. Accordingly, an invention of the same scope performing the same function as the present invention will also fall within the scope of the present invention.

1 is a diagram showing the configuration of a system for correcting a sports posture using an artificial neural network, according to an embodiment of the present invention.

Referring to FIG. 1 , the system includes a server 100 and a user terminal 200 . In this case, each device may be interconnected by wire or wirelessly through a communication network.

According to an embodiment of the present invention, the server 100 receives the exercise video from the user terminal 200 and calculates the posture information of the user in the video using an artificial neural network.

In this case, the artificial neural network may be, for example, a pose estimation model, which is a type of a motion recognition model. Also, the posture information may correspond to information in which at least one specific body part of the user included in the exercise video is recognized, and the angle and length (or speed and direction, etc.) of the specific body part are calculated in units of time. In this case, the pose estimation model may use HPll Human pose, COCO, and Leeds Sports Poses data as posture information.

In addition, the server 100 compares the calculated posture information with the pre-stored correction information, and generates correction information for the error and transmits it to the user terminal 200 .

In this case, the correction information refers to posture information calculated by an expert associated with the moving picture, and the correction information refers to information for correcting a user's posture.

For example, when a user wants to correct a sub posture of volleyball, an image of the sub posture may correspond to an exercise video. Here, the server 100 recognizes the user, and calculates information on the movements of joints and bones recognized in the process of the user performing a serve as posture information. In addition, the posture information calculated from the sub-motion of the actual volleyball player corresponds to the correction information, and information generated so that the user's sub-motion can lower the error value from the sub-motion of the player (this is text, sound, image, etc. various methods ) corresponds to the correction information.

According to an embodiment of the present invention, the server 100 calculates posture information using a motion recognition model from an exercise video obtained from the user terminal 200, and compares it with pre-stored correction information to calculate correction information. characterized.

As described above, a pose estimation model may be used to recognize a user's posture in a video. Specifically, the pose estimation technology finds the human joint in the image (image) by detecting the part affinity fields and the part confidence map in the image. Thereafter, the pose estimation model plays a role of finding and matching the relationship between each joint. In this case, the input value may be an image, and the churyeo value may correspond to the coordinates constituting the joint.

For example, it can be assumed that an image is selected as an input value, the output value is set as a key point for 18 recognized user's joints (temporary number), and applied to the CNN model (pose estimation model). This model recognizes keypoints for the user and his joints on the specific image by inputting a specific image.

In this recognition process, the pose estimation model performs a convolution operation on 18 keypoints to extract relationship information between joints. In addition, based on the calculated relationship information between the joints, each key point with high correlation is combined to generate a skeleton. Finally, by using the joint information between the joints and the combined information between the key points, as shown in FIG. 5B, the skeleton is displayed on the image by combining it. Based on this, information on the position and movement of each joint of the user in the image can be calculated.

However, when an advanced recognition model is developed according to the development of technology, it may be replaced with the corresponding model and used.

According to an embodiment of the present invention, the user terminal 200 transmits the user's exercise video to the server 100 and receives correction information therefor.

In this case, according to an optional embodiment, the user terminal 200 may select a comparison group (expert) for a desired exercise video or additionally provide a comparison video to the server 100 .

Also, as an additional embodiment, when the user terminal is driven in a mobile environment, the exercise video may be used as an image captured in real time through a camera included in the mobile terminal.

In addition, the application for correcting sports posture using an artificial neural network as the user terminal 200 may be an application built into the user terminal 200 or an application downloaded from an application distribution server and installed in the user terminal 200 .

Meanwhile, the user terminal 200 refers to a communication terminal capable of using a terminal application in a wired/wireless communication environment. Here, the user terminal 200 may be a user's portable terminal. In FIG. 1 , the user terminal 200 is illustrated as a smart phone, which is a type of portable terminal, but the inventive concept is not limited thereto. can be borrowed

On the other hand, the communication network means a communication network that provides an access path so that the user terminal 200 can transmit and receive data after accessing the server 100 (this can be borrowed in the opposite case). The communication network is, for example, a wired network such as LANs (Local Area Networks), WANs (Wide Area Networks), MANs (Metropolitan Area Networks), ISDNs (Integrated Service Digital Networks), etc., or wireless networks such as wireless LANs, CDMA, Bluetooth, and satellite communication. However, the scope of the present invention is not limited thereto.

2 is a diagram showing the configuration of a server for correcting a sports posture using an artificial neural network, according to an embodiment of the present invention.

Referring to FIG. 2 , the server 100 according to an embodiment of the present invention includes a communication module 110 , a memory 120 , a processor 130 , a database 140 , a posture recognition module 150 , and a posture comparison module. 160 and a correction information generating module 170 .

In detail, the communication module 110 provides a communication interface necessary to provide a transmission/reception signal between the server 100 and the user terminal 200 in the form of packet data by interworking with a communication network. Furthermore, the communication module 110 may perform a role of receiving a data request from the user terminal 200 and transmitting data in response thereto.

Here, the communication module 110 may be a device including hardware and software necessary for transmitting and receiving signals such as control signals or data signals through wired/wireless connection with other network devices.

In the memory 120, a program for correcting a sports posture using an artificial neural network is recorded. In addition, the processor 130 performs a function of temporarily or permanently storing the processed data. Here, the memory 120 may include magnetic storage media or flash storage media, but the scope of the present invention is not limited thereto.

The processor 130 controls the entire process of correcting the sports posture by using an artificial neural network as a kind of central processing unit. Each step performed by the processor 130 will be described later with reference to FIG. 3 .

Here, the processor 130 may include all kinds of devices capable of processing data, such as a processor. Here, the 'processor' may refer to a data processing device embedded in hardware, for example, having a physically structured circuit to perform a function expressed as a code or an instruction included in a program. As an example of the data processing apparatus embedded in the hardware as described above, a microprocessor, a central processing unit (CPU), a processor core, a multiprocessor, an application-specific integrated (ASIC) circuit) and a processing device such as a field programmable gate array (FPGA), but the scope of the present invention is not limited thereto.

The database 140 stores data on the correction information to be compared with the posture information extracted from the exercise video. In addition, according to an optional embodiment, data for generating user information about the user terminal 200 , posture information extracted from an exercise video, and correction information may be further stored.

Although not shown in FIG. 2 , some of the data for the calibration information may be stored in a database (not shown) physically or conceptually separated from the database 140 .

The posture recognition module 150 calculates posture information from the exercise video received from the user terminal 200 . Specifically, the posture recognition module 150 recognizes a person in a moving picture, and converts the movement (speed, direction, etc.) of joints and bones into data according to the passage of time and generates it as posture information.

The posture comparison module 160 compares the posture information generated by the posture recognition module 150 with the correction information previously stored in the database 140 to generate an error value. For example, the error value may be information about a difference in movement (speed, direction, etc.) of the same joint and bone of a person included in the posture information and the correction information.

The correction information generating module 170 uses the error value calculated by the posture comparison module 160 to generate a guideline through which the user included in the posture information can match the correction information. In this case, the guideline may be composed of text, sound, image, and video, and may be generated by receiving data from an external server if necessary.

3 is an operation flowchart illustrating a method of correcting a sports posture using an artificial neural network, according to an embodiment of the present invention.

Referring to FIG. 3 , the server 100 extracts posture information from the exercise video ( S110 ).

Specifically, the server 100 extracts the user's posture information through the motion recognition model from the exercise video received from the user terminal 200 .

In this case, as described above, a pose estimation model may be applied to the motion recognition model, and a detailed method of extracting posture information will be described with reference to FIG. 4 to be described later.

Next, the server 100 searches for correction information corresponding to the posture information and calculates an error value for each time ( S120 ).

Specifically, the server 100 searches for pre-stored correction information corresponding to the exercise video, compares the posture information with the correction information, and calculates an error value for each time.

In this case, a detailed method of calculating the error value will be described with reference to FIG. 6 to be described later.

Meanwhile, before step S120 , the server 100 may transmit a list of experts associated with the exercise video to the user terminal 200 . In addition, the user of the user terminal 200 selects a specific expert desired by the user from the received expert list and informs the server 100 . For example, when a user wants to correct his or her posture for soccer, a soccer image is provided to the server 100 , and the server 100 delivers a list of past or currently active players to the user.

Thereafter, the server 100 is mapped with a specific expert and compares the pre-stored correction information with the posture information. In addition, as an optional embodiment, in step (S110), the user terminal 200 transmits the video of the expert together with the exercise video to the server 100, and in step (S120), the server 100 transmits the video of the exercise and the video of the expert. You can also compare.

Finally, the server 100 generates correction information for each time error value, and transmits it to the user terminal 200 ( S130 ).

Specifically, the server 100 generates correction information for an error value for each time and provides it to the user terminal 200 .

At this time, the server 100 generates the correction information composed of at least one of text, image, and sound for correcting the posture of the user to be the same as that of the expert in units of time, based on the error value. For example, through a video that recognizes joints and bones and displays them as images, you can indicate the positions of joints or bones by time with text, or create correction information by superimposing the user's posture and the posture of a specific expert as an image. have.

In addition, after step S130, the server 100 updates the deep learning of the motion recognition model based on the posture information and the correction information. Through this, the precision of the posture information in the exercise video is increased, or the accuracy of the error value in the process of comparing the posture information and the correction information is increased.

4 is an operation flowchart illustrating a method of recognizing a user's posture through an artificial neural network, according to an embodiment of the present invention.

Referring to FIG. 4 , the server 100 recognizes the user from the exercise video and calculates a key point ( S111 ).

Specifically, by recognizing the user from the exercise video, a key point corresponding to a specific body part of the user is calculated.

Referring to FIG. 5A , a specific body part of a user may correspond to a fixed reference part, such as an eye, nose, and mouth, and a joint such as a wrist, elbow, shoulder, or knee. Accordingly, the server 100 designates the recognized user as the key point 210 in order to smoothly process the recognized body movement. At this time, the server 100 assigns a number to each key point 210 to distinguish it from each other, and connects the physically adjacent key points 210 to distinguish the skeleton 220 . At this time, the recognized skeleton 220 is used to calculate the user's movement.

In this case, the motion recognition model recognizes the user based on the following two methods.

First, when a top-dwon method is used, a person is first detected and the user's pose can be estimated within the bounding box. Although this has high accuracy, it has a disadvantage that the recognition speed may be slowed down.

Next, when using the bottom-up method, all key points (key points, in the present invention, the joints 210 are designated as key points) are estimated, and the pose is estimated by analyzing each correlation. will do This can be applied in real time because it has a fast recognition speed, but the accuracy may be lower than that of the top-down method.

Accordingly, a model suitable for the processing speed of the operating server 100 to be implemented is applied.

On the other hand, briefly describing the method of generating the skeleton 220, by performing a convolution operation on the key point 210, the relationship information between each joint is extracted.

Next, the pose estimation model generates the skeleton 220 by combining the key points 210 based on the calculated relationship information between the joints.

Next, the server 100 generates a coordinate value for each key point 210 (S112).

For example, referring to FIG. 5B , a key point 210 is designated by recognizing the user's joints playing golf, and a coordinate value 230 between each key point is calculated.

Finally, the server 100 generates the user's posture information by calculating changes in angles and lengths of the plurality of coordinate values in units of time (S113).

In this case, the server 100 divides the exercise video based on a preset playback time (or frame) to design a scenario for each situation. For example, as shown in FIG. 7A , a scene in which a baseball player swings a bat is divided into scenes (1) to (5) and each situation is designated. This is for comparison with the calibration information to be carried out later. For example, the posture (1) is compared with the correction information corresponding to the stance of the correction information, and the posture (4) is compared with the correction information corresponding to the bat acceletation. Through this method, it is possible to aim for accurate matching between the posture information and the correction information.

In an optional embodiment, when two or more people are recognized from the exercise video in step S111, the server 100 assigns a different identifier to each person. Through this, the server 100 calculates each person's posture information corresponding to different identifiers in step S113. In this case, in an optional embodiment, the server 100 may provide the posture information to which each identifier is assigned to the user terminal 200 , and may receive a selection of posture information to be compared with the correction information.

Meanwhile, the method described in steps S111 to S113 may correspond to a bottom-up method. In this way, all key points 210 are estimated from the user, and the pose is estimated by analyzing each correlation. This has lower accuracy than the top-down method, which will be described later, but has a fast recognition speed, so it can be applied in real time, so it has the advantage of being able to be used in environments such as mobile. Another method may use a top-down (top-dwon) method. This is, as shown in FIG. 5c, the motion recognition model recognizes the user included in the motion video, generates a bounding box 240, and sets the user's pose inside the bounding box 240 in a preset time unit. Posture information is calculated by estimating . The top-down method has high accuracy, but has a disadvantage that the recognition speed may be slow. Therefore, the service provider selects and applies either the bottom-up method or the top-down method as needed.

6 is an operation flowchart illustrating a method for calculating an error value according to an embodiment of the present invention.

Referring to FIG. 6 , the server 100 divides the posture information into sections by preset time units, and maps the exercise postures included in each section and the corresponding correction information ( S121 ).

As described above, the posture information is divided into units of time or frames as shown in FIG. 7A.

As a further embodiment, when it is difficult for the user included in the exercise video to compare with the expert's correction information, the user terminal 200 may request another exercise video. For example, if the user's posture in the exercise video cannot be recognized or the pose (or composition) created by the expert of the pre-stored correction information is too different, the exercise re-recorded by the user terminal 200 for smooth comparison You will be asked for a video. In this case, the server 100 may provide a video of a specific expert to be compared to the user terminal 200 to present an appropriate composition for comparison.

Next, the server 100 matches the key point 210 for the same part of the person included in each of the posture information and the correction information (S122).

In an optional embodiment, the posture information (or a video of an expert) may be composed of information about a plurality of people. At this time, the server 100 selects a person with the highest similarity to the location and direction of the expert key point 210 included in the calibration information, matches the key point 210, and deletes unnecessary data.

Finally, the server 100 calculates an error value based on the coordinate value 230 of the key point 210 (S123).

Specifically, the server 100 calculates an error value based on the coordinate values 230 of each of the key points matched with each other as the same part.

For example, the server 100 substitutes the coordinate value 230 of the key point 210 of the user matched in the same way as the expert into a preset formula to calculate an error value.

At this time, the preset formula calculates the inclination and angle of a straight line (skeleton 220) generated by connecting the key points 210 of the user and the expert as shown in FIG. 7B, and how much the postures of the user and the expert match. Whether or not is calculated as the error value.

For example, in FIG. 7B , while moving in the user's Pose1 and Pose2, the skeleton 220 corresponding to the arm changes to a straight line 1 and a straight line 2, respectively.

At this time, the straight line 1 and the straight line 2 have the following formula.

In addition, in order to determine the movement, the change in inclination and angle and each score are calculated as an error value in the process of changing the straight line 1 to the straight line 2.

First, the slope difference follows the formula below.

The matching score according to the slope follows the following three formulas.

Next, the following three formulas for angles are followed.

Through the above formula, a matching score between a user and an expert for each key point 210, a difference in arm angle, and the like are calculated. In this case, the preset formula presented above is an example that can be applied to the position or angle of the skeleton or joint. If necessary, a formula related to speed or direction may be further applied, and a formula covering all of them may be applied.

8A to 8C are diagrams illustrating a method of providing correction information according to an embodiment of the present invention.

At this time, the correction information includes the score for each exercise posture calculated through the error value, image information comparing the previously analyzed exercise video with the exercise video received in step S110, and advice information for correcting the user's exercise posture. It may consist of at least one of

First, referring to FIG. 8A , an example of correction information provided to a user through the user terminal 200 is shown.

In the first area 311 of the display, analysis information on the exercise video may be displayed as an image. For example, the scene in which the batter swings the bat, the key point 210 and the skeleton 220 are synthesized and displayed.

In the second area 313 of the display, an error value obtained by calculating whether or not the expert matches each key point 210 may be displayed.

And, their overall score 312 is displayed in one area of the display.

Next, FIG. 8B is correction information that may be provided when a user periodically updates an exercise video for the same exercise posture.

For example, the total score 312 obtained for each number may be displayed as a graph 321 , and previously uploaded exercise images may be displayed on the third area 322 of the screen in the order in which they were recently uploaded.

As an optional embodiment, as shown in the third area 322 of FIG. 8B , an environment for real-time comparison may be provided by playing an exercise video and a video including correction information of an expert together.

Finally, FIG. 8C is a subdivision of the synthesis graph 321 , and various information may be provided.

The above description of the present invention is for illustration, and those of ordinary skill in the art to which the present invention pertains can understand that it can be easily modified into other specific forms without changing the technical spirit or essential features of the present invention. will be. Therefore, it should be understood that the embodiments described above are illustrative in all respects and not restrictive. For example, each component described as a single type may be implemented in a dispersed form, and likewise components described as distributed may be implemented in a combined form.

The scope of the present invention is indicated by the following claims rather than the above detailed description, 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. do.

100: server
200: user terminal

Claims (15)

In a method of correcting a sports posture using an artificial neural network, performed by a server,
(a) extracting the user's posture information through the motion recognition model from the exercise video received from the user terminal;
(b) retrieving pre-stored correction information corresponding to the exercise video, and calculating an error value for each time by comparing the posture information with the correction information; and
(c) generating correction information for the error value for each time and providing it to the user terminal;
The posture information is information obtained by calculating the angle and length of at least one specific body part from the user included in the exercise video for each time period,
The correction information is the posture information calculated by an expert associated with the exercise video,
The correction information means information for correcting the user's posture, a method of correcting a sports posture using an artificial neural network. The method of claim 1,
The method for correcting a sports posture using an artificial neural network, wherein the motion recognition model is to which a pose estimation model is applied. 3. The method of claim 2,
The motion recognition model is
(a1) recognizing the user from the exercise video, calculating a key point corresponding to a specific body part of the user;
(a2) generating a coordinate value for each of the key points; and
(a3) calculating changes in angles and lengths for a plurality of the coordinate values in units of time, and generating the user's posture information; 4. The method of claim 3,
In the step (a1), when two or more people are recognized in the exercise video, a different identifier is given to each person, and the step (a3) is to calculate the posture information of the person corresponding to the different identifiers, respectively A method of correcting sports posture using artificial neural networks. The method of claim 1,
The step (a) is
The motion recognition model recognizes the user included in the motion video, generates a bounding box, and estimates the user's pose in the bounding box in a preset time unit to calculate the posture information A method of correcting sports posture using an artificial neural network. The method of claim 1,
before step (b)
When a list of experts associated with the exercise video is transmitted to the user terminal, and the user terminal selects a specific expert,
Step (b) is
A method of correcting a sports posture using an artificial neural network, which is to compare the pre-stored correction information mapped with the specific expert with the posture information. 4. The method of claim 3,
Step (b) is
(b1) dividing the posture information by a preset time unit, and mapping the corrective information corresponding to the posture included in each section;
(b2) matching the keypoint to the same part of the person included in each of the posture information and the correction information; and
(b3) calculating the error value based on the coordinate values of each of the key points matched with each other in the same part; a method of correcting a sports posture using an artificial neural network, including a. 8. The method of claim 7,
When the posture information consists of information about a plurality of people,
The step (b2) is
A method of correcting a sports posture using an artificial neural network, which matches the key point by selecting a person having the highest similarity to the position and direction of the expert's key point included in the long-term correction information. 8. The method of claim 7,
The step (b3) is
A method of correcting a sports posture using an artificial neural network, which calculates the error value by substituting the coordinate value of the key point of the user matched identically with the expert into a preset formula. 8. The method of claim 7,
The preset formula calculates the slope and angle of a straight line created by connecting the key points of the user and the expert, and calculates how much the postures of the user and the expert match as the error value, using an artificial neural network How to correct sports posture. The method of claim 1,
Step (c) is
Sports posture using an artificial neural network to generate the correction information consisting of at least one of text, image, and sound for correcting the posture of the user to be the same as the posture of the expert in units of time based on the error value how to correct it. The method of claim 1,
The correction information includes a score for each exercise posture calculated through the error value, image information comparing the previously analyzed exercise video with the exercise video received in step (a), and correcting the user's exercise posture A method of correcting a sports posture using an artificial neural network, which consists of at least one of the advice information for doing. The method of claim 1,
After step (c)
A method of correcting a sports posture using an artificial neural network, which updates the deep learning of the motion recognition model based on the posture information and the correction information. In the apparatus for correcting sports posture using an artificial neural network,
a memory in which a program for correcting sports posture using the artificial neural network is stored; and
A processor for correcting a sports posture using the artificial neural network by executing the program stored in the memory,
The processor extracts the user's posture information through a motion recognition model from the exercise video received from the user terminal, searches for pre-stored correction information corresponding to the exercise video, and compares the posture information with the correction information to obtain an error value for each time Calculating, generating correction information for the error value for each time, and providing it to the user terminal, wherein the posture information is calculated by calculating the angle and length of at least one specific body part from the user included in the exercise video by time information, the correction information is the posture information calculated by an expert associated with the exercise video, and the correction information is information for correcting the posture of the user, an apparatus for correcting a sports posture using an artificial neural network A computer-readable storage medium in which a program for performing a method of correcting a sports posture using an artificial neural network according to claim 1 is recorded.

Images