KR102672275B1 - Apparatus for predicting low back torque using ground reaction information and method of the same - Google Patents

Abstract

One embodiment of the present invention provides a ground reaction force sensor capable of measuring ground reaction force data and COP data, receives the ground reaction force data and COP data, and uses the received ground reaction force data as input to an artificial intelligence model to determine lumbar muscle strength. Provided is a lumbar muscle force estimation device using ground reaction force information that includes a data processing device that generates an estimated value.

Description

Lumbar muscle force estimation device and method using ground reaction force information {APPARATUS FOR PREDICTING LOW BACK TORQUE USING GROUND REACTION INFORMATION AND METHOD OF THE SAME}

The present invention relates to a device for estimating lumbar muscle force, and more specifically, to a device and method for estimating lumbar muscle force applied to the waist using ground reaction force information during a golf swing.

Recently, golf has become a popular sport and its popularity is increasing. Nowadays, golf is not only enjoyed as a hobby and social tool for middle-aged people, but the trend is changing to a hobby sport enjoyed by teenagers and young adults as well. In addition, golf is no longer only possible on the field, but can also be played at golf driving ranges such as indoor golf courses and screen golf courses.

Golf may be thought to have fewer injuries because it is a walking sport, unlike other fast-paced sports, but the load applied during the golf swing movement can cause serious injuries. Among golf players who visited the 'KPGA Physio Service Center' operated by the Korea Professional Golf Association (KPGA) in 2015 and received treatment for injuries, back injuries were the most common. In 2015, 20.6% received treatment for back injuries, and in 2016, the rate of back injuries increased to 26.5%. Based on these results, it can be confirmed that back injuries are an important problem for golfers. Therefore, it is necessary to analyze the causes of back injuries and explore solutions.

Conventionally, in order to analyze the cause of back injuries that occur during a golf swing, there was a method of measuring electromyography and torque applied to the waist using a 3D camera. Using eight cameras, the user's golf swing motion can be filmed, and ground reaction force data can be collected using a ground reaction force device. Using shooting data and ground reaction force data, electromyography or load on body parts can be measured during swing, and this can be used as data to predict back injuries in golf players. However, because it requires the use of numerous cameras and ground reaction machines, there is a problem in that measurement cannot be performed outdoors and can only be performed indoors.

In addition, there is a limitation that it is difficult to accurately obtain data because the infrared marker attached to the user's body may be hidden or fall off during fast and asymmetric swing movements.

US Patent No. 8696450

The technical problem to be achieved by the present invention is to provide a lumbar muscle force estimation device using ground reaction force information that can estimate lumbar muscle force using data from a ground reaction force sensor.

In addition, it provides a technology that can estimate lumbar muscle strength using only a ground reaction force sensor without using expensive infrared cameras and infrared markers.

Additionally, it provides technology that can measure the degree of back injury that occurs during a golf swing, regardless of whether indoors or outdoors.

The technical problem to be achieved by the present invention is not limited to the technical problem mentioned above, and other technical problems not mentioned can be clearly understood by those skilled in the art from the description below. There will be.

In order to achieve the above technical problem, a lumbar muscle force estimation device using ground reaction force information according to an embodiment of the present invention is capable of measuring ground reaction force data and COP (Center of pressure) data according to the user's motion. It includes a ground reaction force sensor and a data processing device that receives the ground reaction force data and the COP data, and generates a lumbar muscle force estimate using the received ground reaction force data and the COP data as input to an artificial intelligence model. Here, the artificial intelligence model can be built using three-dimensional motion data generated by collecting three-dimensional spatial position data, ground reaction force data, and COP data of markers attached to the human body.

Meanwhile, the user's motion is a golf swing motion, and the lumbar muscle strength estimate is the estimated value of L5/S1 torque during each motion of the user's backswing, backswing top, downswing, impact, follow-throw, and finish, which are performed in time series. This can be.

The artificial intelligence setup model configures a Stacked LSTM (Long Short Term Memory) with a first-layer input layer and a third-layer LSTM layer, and the epoch can be set to 200 and K-fold to 10.

The ground reaction force data may be 6 pieces of data according to each axis of both feet, and the COP data may be 4 pieces of data measured from the front/back and inside/outside of both feet.

The three-dimensional spatial location data can be obtained from the user's anatomical boundary points and a marker attached to the golf club.

Here, the artificial intelligence model includes a Long Short Term Memory (LSTM) architecture and can be learned in the form of dropout normalization at a drop out rate of 0.2 to 0.5.

In addition, the method of estimating lumbar muscle strength using ground reaction force information according to an embodiment of the present invention includes three-dimensional motion data generated by collecting three-dimensional spatial position data, ground reaction force data, and COP data of markers attached to the human body. constructing an artificial intelligence model by outputting the extracted lumbar muscle force, receiving ground reaction force data and COP data according to the user's motion from a ground reaction force sensor, and receiving the received ground reaction force data and COP data. It may include inputting into a constructed artificial intelligence model and generating a lumbar muscle force estimate corresponding to the input ground reaction force data and the COP data.

Here, the user's motion is a golf swing motion, and the lumbar muscle strength estimate is the estimated value of L5/S1 torque during each of the user's backswing, backswing top, downswing, impact, follow throw, and finish, which are performed in time series. It can be.

The ground reaction force data may be 6 pieces of data according to each axis of both feet, and the COP data may be 4 pieces of data measured from the front/back and inside/outside of both feet.

Meanwhile, the artificial intelligence model includes a Long Short Term Memory (LSTM) architecture and can be learned in the form of dropout normalization at a drop out rate of 0.2 to 0.5.

According to an embodiment of the present invention, the lumbar muscle force generated during the user's golf movement can be estimated using only a ground reaction force sensor, so complex equipment is not required.

In addition, since the lumbar muscle strength can be estimated using only a ground reaction force sensor without using an expensive infrared camera, the degree of the user's back injury can be predicted regardless of whether it is indoors or outdoors.

Additionally, since the user performs golf movements without wearing an infrared marker, it is possible to prevent accurate data from being measured due to the infrared marker being hidden or falling off during golf movements.

The effects of the present invention are not limited to the effects described above, and should be understood to include all effects that can be inferred from the configuration of the invention described in the detailed description or claims of the present invention.

Figure 1 is a diagram illustrating an apparatus for estimating lumbar muscle strength according to an embodiment of the present invention.
Figure 2 is a diagram illustrating obtaining three-dimensional motion data according to an embodiment of the present invention.
Figure 3 is a flowchart showing a method of building an artificial intelligence model according to an embodiment of the present invention.
Figure 4 is a diagram illustrating a method of setting an artificial intelligence model according to an embodiment of the present invention.
Figure 5 is a diagram illustrating an artificial intelligence model according to an embodiment of the present invention.
Figure 6 is a flowchart showing a method of estimating lumbar muscle strength using an artificial intelligence model according to an embodiment of the present invention.
FIG. 7 is a diagram illustrating actual measured lumbar muscle strength values and estimated values estimated using an artificial intelligence model for each axis according to an embodiment of the present invention.
FIG. 8 is a diagram illustrating a scatter plot and linear correlation coefficient of estimated values for actual measured values of lumbar muscle strength of each axis according to an embodiment of the present invention.
Figure 9 is a diagram showing r-RMSE for each axis of an amateur player and a professional player according to an embodiment of the present invention.
FIG. 10 is a diagram illustrating an estimated lumbar muscle strength value estimated using an artificial intelligence model according to an embodiment of the present invention.

Hereinafter, the present invention will be described with reference to the attached drawings. However, the present invention may be implemented in various different forms and, therefore, is not limited to the embodiments described herein. In order to clearly explain the present invention in the drawings, parts unrelated to the description are omitted, and similar parts are given similar reference numerals throughout the specification.

Throughout the specification, when a part is said to be "connected (connected, contacted, combined)" with another part, this means not only "directly connected" but also "indirectly connected" with another member in between. "Includes cases where it is. Additionally, when a part is said to “include” a certain component, this does not mean that other components are excluded, but that other components can be added, unless specifically stated to the contrary.

The terms used in this specification are merely used to describe specific embodiments and are not intended to limit the invention. Singular expressions include plural expressions unless the context clearly dictates otherwise. In this specification, terms such as “comprise” or “have” are intended to designate the presence of features, numbers, steps, operations, components, parts, or combinations thereof described in the specification, but are not intended to indicate the presence of one or more other features. It should be understood that this does not exclude in advance the possibility of the existence or addition of elements, numbers, steps, operations, components, parts, or combinations thereof.

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

Figure 1 is a diagram showing an apparatus for estimating lumbar muscle strength according to an embodiment of the present invention.

Referring to FIG. 1, the lumbar muscle force estimation device 100 includes a ground reaction force sensor 110, a battery 120, a communication module 130, a microcontroller (MCU: Micro Controller Unit) 140, and a data processing device ( 150) may be included.

First, the ground reaction force sensor 110 is located at the bottom of the user's plantar surface and can detect ground reaction force data and COP (Center of pressure) data distributed to the left and right feet during a swing motion. The user can perform the backswing, which tilts the golf club back, the top of the backswing, which is tilted to the maximum, the downswing, which is the starting motion of the swing, the impact, which is the motion at the moment of hitting the golf ball, and the follow-throw motion, which pushes the golf ball, etc., from above the ground reaction force sensor 110. It can be done.

The measured ground reaction force data may include coordinate values for three axes and may be measured based on each of the three axes. Additionally, data from each of the three axes of both feet of the user can be measured. For example, based on the left foot, the ground reaction force data includes first ground reaction force data measured based on a first axis extending in the anterior/posterior direction, and first ground reaction force data measured based on the first axis extending in the left/right (medial/lateral) direction. It may include second ground reaction force data measured based on the second axis, and third ground reaction force data measured based on the third axis extending in the up/down (proximal/distal) directions.

In addition, based on the right foot, the ground reaction force data includes the fourth ground reaction force data measured based on the first axis extending in the anterior/posterior direction, and the fourth axis extending in the left/right (medial/lateral) direction. It may include fifth ground reaction force data measured based on two axes, and sixth ground reaction force data measured based on a third axis extending in the up/down (proximal/distal) directions.

COP data may be the location where the resultant force applied to the ground through the foot is applied among the entire contact surface where the user's foot touches the ground reaction force sensor 110. COP data may include the sum of the components of each of the three axes, and data from both feet of the user can be measured.

For example, based on the left foot, COP data may include first COP data measured on the anterior/posterior side and second COP data measured on the inside/external side.

Additionally, based on the right foot, COP data may include third COP data measured on the anterior/posterior side and fourth COP data measured on the inside/external side.

According to this, it can include 6 ground reaction force data and 4 COP data.

The battery 120 may supply power to components provided in the lumbar muscle strength device 100.

The communication module 130 can transmit and receive lumbar muscle strength data to devices such as a laptop (T) and a smartphone (P). Devices such as laptops (T) and smartphones (P) can display lumbar muscle strength data and the degree of back injury.

The microcontroller can control the ground reaction force sensor 110 and the communication module 130. As an example, the microcontroller controls the ground reaction force sensor 110 using communication methods such as Inter-Integrated Circuit (I2C), Serial Peripheral Interface (SPI), Serial Communication Interface (SCI), and Enhanced Controller Area Network (ECAN). It is possible to control transmission and reception between the communication module 130 and devices such as a laptop (T) and a smartphone (P).

The data processing device 150 may receive ground reaction force data and COP data from the ground reaction force sensor 110, and may estimate lumbar muscle force using the received ground reaction force data and COP data.

Specifically, the artificial intelligence model is stored in the data processing device 150, and the artificial intelligence model can estimate lumbar muscle force using at least some of the ground reaction force data and COP data as input. The artificial intelligence model can be implemented as an artificial intelligence model LSTM (Long Short Term Memory) that learns characteristic values on its own over time and makes predictions based on correlated time series data.

The data processing device 150 may be implemented in a smart device such as a laptop (T) or a smartphone (P), or may be implemented in a remote server to perform the same function through data communication.

Figure 2 is a diagram illustrating obtaining three-dimensional motion data according to an embodiment of the present invention.

Referring to FIG. 2, the user's three-source motion data can be obtained using the ground reaction force sensor 110 and the infrared camera 200.

Ground reaction force data and COP data according to the user's movement can be collected using the ground reaction force sensor 110. For example, the data collected from the ground reaction force sensor 110 may be composed of a combination of six ground reaction force data along each axis of both feet and four COP data measured on the front/back and inner/outer sides of both feet.

Additionally, the infrared camera 200 can be used to collect location data in three-dimensional space of a marker attached to the user's body. As an example, the first infrared camera 201 and the second infrared camera 202 are used to collect rear data of the user and the golf club, and the third infrared camera 203 and the fourth infrared camera 204 are used to collect side data. Data can be collected, and front data can be collected using the fifth infrared camera 205 and the sixth infrared camera 206. The marker attached to the user's body can be a full-body marker set and can be attached to 35 important anatomical boundary points of the human body and 4 golf clubs.

The motion analysis system 300 can receive ground reaction force data, COP data, and position data. Additionally, the three-dimensional motion corresponding to the user's motion can be analyzed using the ground reaction force data and COP data collected from the ground reaction force sensor 110 and the position data collected by the infrared camera 200.

The artificial intelligence model building unit 400 can use three-dimensional motion data according to the user's movements analyzed by the motion analysis system 300 as learning and verification data for the artificial intelligence model.

The data processing device 150 may output an estimated value of lumbar muscle strength according to the user's movement through the artificial intelligence model building unit 400.

Figure 3 is a flowchart showing a method of building an artificial intelligence model according to an embodiment of the present invention.

In step S100, position data, ground reaction force data, and COP data to be used as input in the artificial intelligence model can be collected. Position data, ground reaction force data, and COP data measured in time series can be collected during golf movements such as the user's backswing, backswing top, downswing, impact, follow throw, and finish. For example, six infrared cameras can collect data by photographing the front, side, and back of the user through a marker attached to the user. The location data may include location data in three-dimensional space taken from the back, side, and front of the user. The ground reaction force sensor is located under both feet of the user and can collect ground reaction force data and COP data. Additionally, the ground reaction force data may be 6 pieces of data according to each axis of both feet, and the COP data may be 4 pieces of data measured from the front/back and inside/outside of both feet.

The data collected in step S110 can be preprocessed. As an example, marker data attached to the user's body and golf club can be smoothly preprocessed from location data. Because the first infrared camera and the second infrared camera measure the back of the user, the marker on the front may be obscured and not visible. Therefore, during the preprocessing process, the hidden portion of the marker data can be stitched or modified. Likewise, marker data that is hidden and invisible from the third and fourth infrared cameras and marker data that is hidden and invisible from the fifth and sixth infrared cameras can be stitched or modified through preprocessing.

Lumbar muscle strength may be extracted from the data preprocessed in step S120 by analysis software. That is, the lumbar muscle force corresponding to the combination of preprocessed position data, ground reaction force data, and COP data can be extracted.

As an example, the user can calculate the L5/S1 torque corresponding to a combination of position data, ground reaction force data, and COP data measured in time series during golf movements such as backswing, backswing top, downswing, impact, follow throw, and finish. there is. L5 is one of the parts of the lumbar spine, the spine of the lower back, and S1 is one of the sacrum, the bones that make up the pelvis. Therefore, lumbar muscle strength can be extracted based on the L5/S torque corresponding to each user's movement.

An artificial intelligence model can be set up using the three-dimensional spatial position data, ground reaction force data, COP data, and extracted lumbar muscle force collected in step S130. The detailed configuration of the artificial intelligence model settings will be described in detail with reference to FIGS. 4 and 5.

As an example, the LSTM artificial intelligence model used may consist of a Stacked LSTM with a first-layer input layer and a third-layer LSTM layer. In order to optimize the algorithm of the artificial intelligence model, epoch is set to 200, K-fold is set to 10, and 80% of the total data can be randomly selected as the training data set. Additionally, the data can be composed of 10% verification data set and 10% test data set.

Additionally, an artificial intelligence model that can extract lumbar muscle force can be set up by inputting each or a combination of ground reaction force data and COP data as input values.

In step S140, an estimated value of lumbar muscle force can be output using the ground reaction force data and COP data using the set artificial intelligence model. The artificial intelligence model does not use infrared camera position data, but only ground reaction force data and COP data can be used as input values for estimating lumbar muscle strength. That is, the lumbar muscle force corresponding to the input ground reaction force data and COP data can be estimated.

The estimated value of lumbar muscle strength output in step S150 can be verified. Modification of the artificial intelligence model may be performed according to the verification results. At this time, the artificial intelligence model learned with the learning set can be evaluated for actual and measured values using the validation set. Depending on the evaluation results, it can be decided whether to modify the artificial intelligence model.

In step S160, it can be determined whether modification of the artificial intelligence model is necessary. For example, a method of modifying the weight of the hidden layer of the artificial intelligence model can be used to reduce the error between the calculated value and the estimated value. If it is determined that the artificial intelligence model needs to be modified, the weight of the artificial intelligence model hidden layer can be modified in the artificial intelligence model setting step (S130). The artificial intelligence model may repeat steps S130 to S160 to reduce the error between the actual measured value and the estimated value.

If it is determined that modification of the artificial intelligence model is not necessary, the final artificial intelligence model can be determined in the artificial intelligence model construction step (S170), and the lumbar muscle strength can be estimated from this. In other words, it is possible to build an artificial intelligence model that can estimate lumbar muscle force using only ground reaction force data and COP data individually or in combination.

Figure 4 is a diagram illustrating a method of setting an artificial intelligence model according to an embodiment of the present invention.

Referring to FIG. 4, the image data, ground reaction force data, and COP data collected in step S131 can be preprocessed. The collected data can be preprocessed kinematically. The preprocessed data can be processed with an analysis program to estimate lumbar muscle strength. The estimated lumbar muscle forces can be kinematically preprocessed.

In step S132, an artificial intelligence model can be set up using only each or a combination of ground reaction force data and COP data as input variables and lumbar muscle force as an output variable.

As an example, an artificial intelligence model can learn in the form of drop out regularization. Dropout can be used when an artificial intelligence model is learning, excluding some rather than learning from all layers. As an example, the dropout rate can be set to 0.3, and a typical dropout rate of 0.2 to 0.5 can be used. If the artificial intelligence model is complex, it can be adjusted by increasing the dropout rate, and if it is simple, it can be adjusted by lowering the dropout rate.

In step S133, noise can be removed by filtering using a low-pass filter.

Figure 5 is a diagram showing the LSTM architecture of an artificial intelligence model according to an embodiment of the present invention.

Referring to Figure 5, the artificial intelligence model can use a deep learning model that learns data that changes over time to estimate lumbar muscle strength. In one embodiment, the artificial intelligence model may be implemented with a Long Short Term Memory (LSTM) architecture. LSTM architecture is an artificial intelligence model that predicts correlated time series data and is a type of recurrent neural network model that processes data sequentially. There is a short-term memory layer (short term, hidden state) and a layer responsible for long-term memory (long term, cell state), resulting in high accuracy in time series data prediction.

According to this, lumbar muscle strength can be estimated more accurately. This is because the position data, ground reaction force data, and COP data input to the artificial intelligence model are data measured in time series when the user operates. In other words, the artificial intelligence model can more accurately estimate lumbar muscle force by adopting an LSTM architecture suitable for prediction of time series data to analyze time series measured position data, ground reaction force data, and COP data.

The input vectors of the LSTM architecture can be each or a combination of position data on the x, y, and z axes, ground reaction force data, and COP data. The output vector can be lumbar muscle force values in the x, y, and z axes.

Figure 6 is a flowchart showing a method of estimating lumbar muscle strength using an artificial intelligence model according to an embodiment of the present invention.

Referring to FIG. 6, ground reaction force data and COP data can be input in step S200. Six ground reaction force data for each axis of the user's both feet and four COP data for both feet can be used as input values for the artificial intelligence model.

Lumbar muscle strength can be estimated using the data input in step S210. An artificial intelligence model is created by using only the ground reaction force data and COP data of the ground reaction force sensor without using the position data according to the user's movement from the infrared camera. The lumbar muscle strength can be estimated. As an example, the L5/S1 torque calculated during the user's backswing, backswing top, downswing, impact, follow throw, and finish, which are performed in time series, can be estimated using only ground reaction force data and COP data. Therefore, lumbar muscle strength can be estimated based on the L5/S torque corresponding to each movement of the user.

In step S220, the estimated lumbar muscle force can be output using the artificial intelligence model.

FIG. 7 is a diagram illustrating actual measured lumbar muscle strength values and estimated values estimated using an artificial intelligence model for each axis according to an embodiment of the present invention.

Referring to Figure 7, (A) is a graph showing the actual and estimated values of lumbar muscle strength for each axis of an amateur athlete. (B) is a graph showing the actual and estimated values of lumbar muscle strength for each axis of professional athletes. The backswing, which tilts the golf club back, the top of the backswing, which is tilted back to its maximum, the downswing, which is the starting motion of the swing, the impact, which is the motion at the moment of hitting the golf ball, and the follow-throw motion, which pushes the golf ball, were set as one cycle. Figures 7 (A) and (B) continuously show actual and estimated values for one cycle of athletes.

)이고, (B)에서 프로 선수의 RMSE의 값은 각 축에 대해 0.20 ~ 0.38(

)인 것을 확인할 수 있다. RMSE는 추정값과 실측값의 차이를 확인하기 위하여 사용했다. As shown in Figure 7, it can be seen that most of the estimated value graphs for each axis have a similar pattern to the actual measured value graph depending on the frame. Additionally, in (A), the values of RMSE (Root Mean Square Deviation) of amateur players range from 0.23 to 0.42 for each axis (

), and in (B), the value of the professional player's RMSE is 0.20 to 0.38 for each axis (

) can be confirmed. RMSE was used to check the difference between estimated and actual values.

FIG. 8 is a diagram illustrating a scatter plot and linear correlation coefficient of estimated values for actual measured values of lumbar muscle strength of each axis according to an embodiment of the present invention.

As an example, the correlation between the estimated value and the actual value can be expressed with PCC (Pearson's correlation coefficient), which is an indicator that compares the linear correlation of graphs made from two different data. Expresses the linear correlation between X and Y, and can have values between +1 and -1. A positive linear graph with a slope of +1 means that the positive correlation matches perfectly, a slope of -1 means that the negative correlation matches perfectly, and a slope of 0 means that there is no correlation at all. It can be expressed.

The solid line shows y=x, and the dotted line shows the correlation coefficient, so it is a graph showing y=rx for the correlation coefficient r.

In (A), it can be seen that the amateur player's X-axis correlation coefficient is 0.96±0.03, Y-axis correlation coefficient is 0.95±0.03, and Z-axis correlation coefficient is 0.92±0.06. Additionally, in (B), it can be seen that the professional player's X-axis correlation coefficient is 0.95±0.03, Y-axis correlation coefficient is 0.96±0.03, and Z-axis correlation coefficient is 0.96±0.03.

Figure 9 is a diagram showing r-RMSE for each axis of an amateur player and a professional player according to an embodiment of the present invention.

RMSE can react more sensitively when the error due to the difference between the estimated value and the actual value increases. Therefore, r-RMSE was used to quantify and compare RMSE without being affected by the size of the data value. r-RMSE is a normalized graph in which the RMSE value is divided by the difference between the maximum and minimum values and then multiplied by 100 and expressed as a ratio, allowing you to check the ratio of errors.

Referring to Figure 9, the r-RMSE value of the amateur player's It can be seen that the r-RMSE value of the professional player's

FIG. 10 is a diagram illustrating an estimated lumbar muscle strength value estimated using an artificial intelligence model according to an embodiment of the present invention.

Lumbar muscle strength can be estimated using only data from the ground reaction force sensor as input to the artificial intelligence model. As an example, data from the ground reaction force sensor can estimate lumbar muscle strength using six types of ground reaction force data and four types of COP data.

)이며, r-RMSE는 5.23 ~ 15.71 (%)이고, 상관 계수는 0.914~0.963으로 나타났다. 인공 지능 모델을 이용한 생체 역학 분야에서 r-RMSE가 15% 미만, PCC가 0.9 수준이면 예측 수행을 잘하는 것으로 판단되고 있다. 가장 큰 값을 가지는 z축의 r-RMSE는 15.71%로 15% 수준이며, 가장 낮은 값을 가지는 z축의 PCC는 0.914이므로 0.9 수준인 것을 확인할 수 있다. Referring to Figure 10, the scatter plot and linear correlation between the estimated value and the actual measured value for the lumbar muscle strength of the artificial intelligence model can be confirmed. It can be seen that artificial intelligence models generally have small errors and a linear correlation close to +1. In addition, in the case of r-RMSE, the z-axis was the highest at 10.17~15.71, and the correlation coefficient was also the lowest at the z-axis at 0.914~0.933. Overall, RMSE ranged from 0.13 to 0.69 (

), r-RMSE was 5.23 ~ 15.71 (%), and the correlation coefficient was 0.914 ~ 0.963. In the field of biomechanics using artificial intelligence models, prediction performance is considered good if r-RMSE is less than 15% and PCC is at the level of 0.9. The r-RMSE of the z-axis with the highest value is 15.71%, which is at the 15% level, and the PCC of the z-axis with the lowest value is 0.914, so it can be confirmed that it is at the 0.9 level.

Based on these results, it can be determined that the prediction accuracy in estimating lumbar muscle strength using data from the ground reaction force sensor is high. Therefore, it can be confirmed that the LSTM artificial intelligence model can estimate lumbar muscle strength using only the ground reaction force data and COP data from the ground reaction force sensor.

The description of the present invention described above is for illustrative purposes, and those skilled in the art will understand that the present invention can be easily modified into other specific forms without changing the technical idea or essential features of the present invention. will be. Therefore, the embodiments described above should be understood in all respects as illustrative and not restrictive. For example, each component described as single may be implemented in a distributed manner, and similarly, components described as distributed may also be implemented in a combined form.

The scope of the present invention is indicated by the claims described below, and all changes or modified forms derived from the meaning and scope of the claims and their equivalent concepts should be construed as being included in the scope of the present invention.

100. Lumbar muscle force estimation device
120. Battery
130. Communication module
140.MCU
150. Data processing device

Claims (10)

In a device for estimating lumbar muscle force using ground reaction force information,
Ground reaction force sensor that can measure ground reaction force data and COP (Center of pressure) data according to the user's motion; and
A data processing device that receives the ground reaction force data and the COP data and generates a lumbar muscle force estimate using the received ground reaction force data and the COP data as input to an artificial intelligence model,
The artificial intelligence model is constructed using three-dimensional motion data generated by collecting three-dimensional spatial position data, ground reaction force data, and COP data of markers attached to the human body,
The artificial intelligence model is implemented with a Long Short Term Memory (LSTM) architecture that predicts correlated time series data,
The user's motion is a golf swing motion,
The artificial intelligence model uses the ground reaction force data and the COP data measured during the time-series actions of backswing, backswing top, downswing, impact, follow-throw, and finish as input values, and is estimated in response to the time-series actions. A lumbar muscle force estimation device for estimating the lumbar muscle force based on the L5/S1 torque. delete According to paragraph 1,
The artificial intelligence setting model consists of a stacked LSTM (Long Short Term Memory) with a first-layer input layer and a third-layer LSTM layer, and the epoch is set to 200 and K-fold is set to 10. A lumbar muscle force estimation device . In paragraph 1
The ground reaction force data is 6 pieces of data according to each axis of both feet, and the COP data is 4 pieces of data measured from the front/back and inside/outside of both feet. Lumbar muscle force estimation device. According to paragraph 1,
The lumbar muscle strength estimation device wherein the three-dimensional spatial location data is obtained from the user's anatomical boundary points and a marker attached to a golf club. According to clause 3,
The LSTM (Long Short Term Memory) architecture is a lumbar muscle force estimation device that is learned in the form of dropout normalization at a drop out rate of 0.2 to 0.5. In a method of estimating lumbar muscle force using ground reaction force information,
Constructing an artificial intelligence model by outputting lumbar muscle force extracted using three-dimensional motion data generated by collecting three-dimensional spatial position data, ground reaction force data, and COP data of markers attached to the human body;
Receiving ground reaction force data and COP data according to the user's motion from a ground reaction force sensor;
Inputting the received ground reaction force data and COP data into the constructed artificial intelligence model; and
Generating a lumbar muscle force estimate corresponding to the input ground reaction force data and the COP data,
The artificial intelligence model is implemented with a Long Short Term Memory (LSTM) architecture that predicts correlated time series data,
The user's motion is a golf swing motion,
The artificial intelligence model uses the ground reaction force data and the COP data measured during the time-series actions of backswing, backswing top, downswing, impact, follow-throw, and finish as input values, and is estimated in response to the time-series actions. A method for estimating lumbar muscle force based on the L5/S1 torque. delete In clause 7,
The ground reaction force data is 6 pieces of data according to each axis of both feet, and the COP data is 4 pieces of data measured from the front/back and inside/outside of both feet. A method of estimating lumbar muscle strength. In clause 7,
The LSTM (Long Short Term Memory) architecture is learned in the form of dropout normalization at a drop out rate of 0.2 to 0.5.

Images