WO2022097795A1

WO2022097795A1 - Smart insole device and system for estimating user's weight

Info

Publication number: WO2022097795A1
Application number: PCT/KR2020/015610
Authority: WO
Inventors: 전상훈; 양재완
Original assignee: 주식회사 길온
Priority date: 2020-11-06
Filing date: 2020-11-09
Publication date: 2022-05-12
Also published as: KR102421699B1; KR20220061473A

Abstract

A weight estimation system according to one embodiment may measure inertia information including acceleration and angular velocities of a user's foot, extract section data related to the user's weight from the inertia information, generate input data from the extracted section data, and estimate a weight result by inputting the input data into a neural network model stored in a weight estimation server.

Description

Smart guide device and system for estimating user's weight

Hereinafter, a technique for estimating a user's weight using a smart insole device is provided.

Existing weight measurement methods include a method of using a scale equipped with a physical load cell in a fixed place, or a method of measuring weight by mounting a sensor that detects pressure and load on the lower part of a shoe. However, since the above-described methods use professional measuring equipment equipped with many pressure sensors, they require high cost or a large form factor, making it practically impossible to commercialize them.

There is a need for a technique for estimating a user's weight even during an activity, for example, walking.

The weight estimation system may extract inertia information of a section related to the user's weight from the smart insole terminal.

The weight estimation system can improve the structure of the neural network model and the retraining of the neural network model by continuously updating the gait-related factors related to the weight change.

The weight estimation system may monitor a sudden change in the user's weight.

However, the technical tasks are not limited to the above-described technical tasks, and other technical tasks may exist.

A smart insole device according to an embodiment includes an inertial measurement unit (IMU) that measures inertial information including acceleration and angular velocity of a user's foot; a processor for extracting section data related to the weight of the user from the inertia information and generating input data from the extracted section data; and a communication unit for requesting weight estimation while transmitting the input data to be input to the neural network model to the weight estimation server storing the neural network model.

The processor extracts, from the inertia information, at least one of heel strike data of a section including a heel landing time and toe off data of a section including a heel easy time point from the inertia information as the section data can do.

The smart insole device further includes a pressure sensor for sensing the pressure applied to the ground by the foot at a plurality of points while the bottom of the shoe equipped with the smart insole device touches the ground, and the processor is configured by the pressure sensor. At least one of the heel landing time and the heel easy time may be determined based on the momentum calculated from the pressure sensed at the plurality of points and the inertia information.

The processor may calculate an acceleration along a gravity direction from at least one of the heel landing data and the heel easy data, and generate input data including the calculated acceleration.

The processor may generate the input data by using section data extracted from the inertia information during the user's constant speed walking state.

The processor calculates the movement speed of the user in a direction parallel to the ground based on the inertia information, and when the calculated movement speed is maintained over a predetermined threshold time, the user is in the constant-speed walking state it can be decided that

The processor may generate the input data including an amount of exercise and an amount of impulse calculated based on the current weight of the user, together with the section data.

The processor may calculate an acceleration and momentum in the direction of gravity for every sampling point of the section data, and calculate an impact amount that is a sum of momentum amounts within a section of the section data.

The communication unit receives a weight result estimated by applying the neural network model to the input data from the weight estimation server, and the processor, after receiving the weight result, uses the weight result and the inertia information You can create input data.

A method of operating a smart insole device according to an embodiment includes measuring inertial information including acceleration and angular velocity of a user's foot; extracting section data related to the weight of the user from the inertia information; generating input data from the extracted section data; and requesting weight estimation while transmitting the input data to be input to the neural network model to a weight estimation server storing the neural network model.

A weight estimation server according to an embodiment includes: a memory for storing a neural network model; a communication unit configured to receive input data generated from section data related to a user's weight among inertia information measured by the smart insole device from the smart insole device; and a processor for estimating a weight result of the user based on the neural network model from the input data.

The communication unit may receive the input data generated for at least one of a section including a heel landing time point and a section including a heel easy time point among the inertia information.

The communication unit may receive the input data including the acceleration along the direction of gravity, calculated from at least one of heel landing data and heel easy data.

The communication unit may receive the input data generated for the user's constant speed walking state from the inertia information.

The communication unit may receive input data including the amount of exercise and impact calculated according to the section data and the current weight of the user.

The processor may estimate the weight result of the user by applying an operation according to the neural network model to the input data, and the communication unit may transmit the estimated weight result to the smart insole device.

The processor additionally collects population data in which an error between the weight estimated by the neural network model and the actual weight exceeds a threshold value for each weight group, and further trains the neural network model based on the additionally collected population data can do it

The communication unit may receive the population data collected on the user's constant speed walking state in the smart insole device when a connection between the smart insole device and the weight estimation server is established.

The processor may add a new input layer capable of receiving new input data defined to include, as an input variable, a target gait factor whose weight relevance exceeds a threshold score among a plurality of gait factors to the neural network model.

The weight estimation system can minimize noise and calculations that may occur in the weight estimation server by accurately extracting inertia information of a section related to the user's weight from the smart insole terminal.

The weight estimation system may provide a more accurate weight estimation result by retraining the neural network model and improving the structure of the neural network model.

The weight estimation system may monitor an abnormal condition such as a case in which a user's weight changes rapidly without omission.

1 illustrates a weight estimation system according to an embodiment.

2 is a flowchart illustrating a weight estimation method according to an exemplary embodiment.

3 is a block diagram illustrating a configuration of a smart insole device according to an embodiment.

4 is a flowchart illustrating a method of operating a smart insole device according to an embodiment.

5 is a view for explaining a physical quantity in the direction of gravity according to an embodiment.

6 is a diagram illustrating a configuration of a sensor included in a smart insole device according to an embodiment.

7 is a view for explaining a heel landing event and a heel easy event according to an embodiment.

8 illustrates raw data sensed by a smart insole device according to an exemplary embodiment.

9 and 10 are diagrams for explaining section data extraction based on the heel landing event and the heel easy event according to an embodiment.

11 is a block diagram illustrating a configuration of a weight estimation server according to an embodiment.

12 is a flowchart illustrating a training method according to an embodiment.

13 is a diagram for explaining training and improvement of a neural network model according to an embodiment.

Specific structural or functional descriptions of the embodiments are disclosed for purposes of illustration only, and may be changed and implemented in various forms. Accordingly, the actual implementation form is not limited to the specific embodiments disclosed, and the scope of the present specification includes changes, equivalents, or substitutes included in the technical spirit described in the embodiments.

Although terms such as first or second may be used to describe various elements, these terms should be interpreted only for the purpose of distinguishing one element from another. For example, a first component may be termed a second component, and similarly, a second component may also be termed a first component.

When a component is referred to as being “connected to” another component, it may be directly connected or connected to the other component, but it should be understood that another component may exist in between.

The singular expression includes the plural expression unless the context clearly dictates otherwise. In this specification, terms such as "comprise" or "have" are intended to designate that the described feature, number, step, operation, component, part, or combination thereof exists, and includes one or more other features or numbers, It should be understood that the possibility of the presence or addition of steps, operations, components, parts or combinations thereof is not precluded in advance.

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art. Terms such as those defined in a commonly used dictionary should be interpreted as having a meaning consistent with the meaning in the context of the related art, and should not be interpreted in an ideal or excessively formal meaning unless explicitly defined in the present specification. does not

Hereinafter, embodiments will be described in detail with reference to the accompanying drawings. In the description with reference to the accompanying drawings, the same components are assigned the same reference numerals regardless of the reference numerals, and overlapping descriptions thereof will be omitted.

1 illustrates a weight estimation system according to an embodiment.

The weight estimation system 100 according to an embodiment may include a smart guide device 110 , a weight estimation server 120 , and a mobile terminal 130 . The weight estimation system 100 may estimate the user's weight using the neural network model of the weight estimation server 120 from information collected by the smart insole device 110 .

The smart insole device 110 is a device in the form of an insole mounted in a shoe, and may collect information related to the movement of a user's foot while wearing the shoe. For example, the smart insole device 110 may measure inertia information of the foot. The smart insole device 110 may calculate various physical quantity data correlated with body weight, such as a user's walking speed, driving speed, amount of exercise on the ground, and amount of impact, from the inertia information.

The weight estimation server 120 may receive data measured or calculated by the smart insole device 110 , and estimate the user's weight using the received data. The weight estimation server 120 may return the estimated weight to the smart insole device 110 and/or the mobile terminal 130 . The weight estimation server 120 may store a neural network model for estimating the user's weight from various physical quantity data that can be measured and calculated by the smart insole device 110 .

The mobile terminal 130 relays the connection between the smart insole device 110 and the weight estimation server 120 , and at least a portion of the measurement data collected by the smart insole device 110 and/or the estimated weight result can also be stored and managed. However, the connection between the smart insole device 110 and the weight estimation server 120 is not established only through the mobile terminal 130 , but the smart insole device 110 and the weight are established through a gateway device other than the mobile terminal 130 . A connection may be established between the estimation servers 120 .

First, in step 210 , the smart insole device 110 may generate input data from inertia information. The smart insole device 110 may extract section data of a section related to the user's weight from among the inertia information, and calculate weight-related physical quantity data from the extracted section data. For example, the smart insole device 110 may capture the moment the foot lands on the ground and the moment the foot falls off the ground, and pre-process the inertia information sensed at the moment to calculate the physical quantity data. The physical quantity data related to body weight will be described in FIG. 5, and the weight related section will be described in FIG. 7 below. The smart insol device 110 may generate gait data including section data and physical quantity data to be input to the neural network model as input data.

Then, in step 220 , the smart insol device 110 may transmit input data to the weight estimation server 120 . In this case, the smart insole device 110 may transmit input data to the weight estimation server 120 via the mobile terminal 130 (eg, a smart phone, etc.). For example, the smart insole device 110 may transmit input data to the mobile terminal 130 , and the mobile terminal 130 may transmit data collected from the smart insole device 110 to the weight estimation server 120 . .

Subsequently, in step 230 , the weight estimation server 230 may perform weight estimation using the neural network model. The weight estimation server 120 according to an embodiment may store a neural network model designed to estimate weight. The neural network model is a model designed to output weight from input data, and learns the correlation between weight-related physical quantity data (eg, inertia information, momentum, impulse, and acceleration, etc.) and weight included in the previously described input data. It can be one model. Due to external factors such as a user's gait pattern, mathematical modeling such as polynomials for directly calculating weight from momentum, impulse, and acceleration may be difficult. The weight estimation server 230 may estimate the user's weight from the input data using the neural network model trained for the above-described correlation without direct mathematical modeling. Weight estimation will be described with reference to FIG. 11 below.

And in step 240 , the weight estimation server 120 may transmit the estimated weight result. The mobile terminal 130 may receive and store the inference result of the weight estimation server 120 , or may transmit it to the smart insole device 110 . The communication unit of the smart insole device 110 may receive a weight result estimated by applying a neural network model to input data from the weight estimation server 120 .

Thereafter, after receiving the weight result, the processor of the smart insole device may generate input data using the weight result and inertia information. As will be described later, the user's weight must be given in order to calculate the amount of exercise and impulse included in the input data, and the smart insole device 110 may update the weight with the estimated weight result each time.

The weight estimation system 100 according to an embodiment can minimize noise and calculation amount that may occur in the weight estimation server 120 by accurately extracting inertia information of a section related to the user's weight from the smart insol terminal. Also, as will be described later, the weight estimation system 100 may provide a more accurate weight estimation result by retraining the neural network model and improving the structure of the neural network model.

The weight estimation system 100 according to an embodiment may determine a medical abnormality when the weight changes more than a predetermined threshold ratio (eg, 10%) for a predetermined period (eg, 6 months to 12 months). can For example, a change in body weight above a predetermined threshold rate is related to various medical factors such as aging, dementia, depression, anorexia, tumors, endocrine diseases (eg diabetes), and neurological diseases (eg dementia). can be caused by The weight estimation system 100 may estimate the weight by simply walking while wearing the shoes in which the smart insole device 110 is mounted, without a conscious action of the user climbing on the scale. Accordingly, the weight estimation system 100 may monitor an abnormal state such as a case in which the user's weight is rapidly changed without omission. The weight estimation system 100 may detect an abrupt change in body weight compared to the normal weight change through continuous weight estimation in daily life. The weight estimation system 100 may provide a warning, a notification, and a guide regarding the weight abnormality change rate to the user. The weight estimation system 100 may induce a user to be alert to a disease or an external factor, and may induce an early response.

The weight estimation system 100 may monitor and provide real-time weight changes during external activities and exercise to users who need weight management. The weight estimation system 100 may provide the user with a weight that changes in real time as described above as an additional indicator for controlling and managing the amount of exercise.

Additionally, the weight estimation system 100 may be linked with a body scale that is fixed in an existing room. The weight estimation system 100 may provide weight management indoors and outdoors, and in particular, may provide daily life weight management for a group of patients with diseases or diseases in which weight management is essential. The real-time weight estimation result monitored by the weight estimation system 100 may be utilized as a new basic body information index in rehabilitation and medical care.

The smart insole device 300 according to an embodiment may include a sensor 310 , a processor 320 , a communication unit 330 , and a memory 340 .

The sensor 310 may sense information related to the user's weight. The sensor 310 according to an embodiment may include a pressure sensor 311 and an inertial measurement unit (IMU) 312 . The inertia measurement unit 312 may measure inertia information including acceleration and angular velocity of the user's foot. The inertial measurement unit 312 may include an acceleration sensor and a gyro sensor. The acceleration sensor may detect acceleration in three directions, and the gyro sensor may detect angular velocity based on three axes. The smart insole device 300 may obtain complex sensing data from the pressure sensor 311 and the inertia measurement unit 312 and measure the user's weight using only the inertia information. The complex sensing data may include pressure information and inertia information. The pressure sensor 311 will be described with reference to FIG. 6 , and the inertia measuring unit 312 will be described with reference to FIG. 5 .

The processor 320 may extract section data related to the user's weight from the inertia information. The processor 320 may generate input data from the extracted section data. For reference, the processor 320 of the smart insole device mounted on a shoe worn on one foot of the user generates input data that pairs the first input data for one foot with the second input data for the other foot. may be

The communication unit 330 may request weight estimation while transmitting input data to be input to the neural network model to the weight estimation server storing the neural network model. The weight estimation server of the weight estimation system may estimate the user's weight based on a neural network model previously trained from input data in response to a request from the smart guide device 300 .

The memory 340 may temporarily or semi-permanently store inertial information and pressure information measured by the sensor 310 . For example, the memory 340 may include a buffer for storing section data related to the user's weight among the inertia information.

The smart insole device 300 has a minimum device space and low power consumption, and can be easily commercialized. The smart insole device 300 may provide a wearable device-based weight management service. A weight estimation system including the smart insole device 300 may estimate the user's weight while walking and/or driving.

First, in step 411 , the sensor of the smart insole device may measure inertial information. For example, the inertia measurement unit may measure an acceleration with respect to three axes and an angular velocity based on the three axes as inertia information for one foot of the user. The inertial information measured by the inertial measurement unit may also be referred to as inertial raw data.

And in step 412, the processor of the smart insole device may extract section data related to the user's weight from the inertia information. As will be described later with reference to FIG. 7 , the weight-related section data may be section data including the moment the user's foot touches the ground and section data including the moment the user's foot falls off the ground. The generation of input data will be described with reference to FIGS. 5 to 10 below.

Subsequently, in step 413 , the processor may generate input data based on the extracted section data. For example, the section data includes raw data of a section related to weight, and the processor calculates weight-related physical quantity data (eg, acceleration in the direction of gravity, impulse, and momentum, etc.) from the raw data of the section related to weight. can do. The processor may include the calculated weight-related physical quantity data together with the raw data and additional information of the weight-related section in the input data. Additional information is auxiliary information related to weight, for example, the user's weight, movement speed (eg, walking speed or running speed), age, gender, height, and movement time ( For example, information such as a walking time or a driving time) may be included. The input data will be described in detail with reference to FIG. 10 below.

Theoretically, knowing the force F and the acceleration a at F = ma, the mass m can be calculated. However, since the force is applied in various directions during the user's walking due to the user's gait pattern and other various external factors, it is difficult to accurately measure the force applied to the total mass. In addition, in order to accurately measure the force applied to the total mass, more sophisticated and expensive equipment than the smart insole device 500 may be required. Since the user belongs to the gravimeter on the Earth, a force by the user's total mass is always applied in the direction of gravity 511 . While the user wearing the shoes walks or runs, a force related to the user's total mass may always be generated in the direction of gravity 511 in the user's foot portion. In other words, as the amount of change in the amount of momentum in the direction of gravity 511 by this force while the user wearing the shoes walks, an impact amount may be generated. Since the impulse is the integration of a force over time, physical quantities such as momentum, impulse, and acceleration (eg, acceleration in the direction of gravity 511 ) are correlated with the user's total mass. Since the user's body weight is a value obtained by multiplying the mass by the gravitational acceleration in the Earth's gravimeter, the body weight also has a correlation with physical quantities such as momentum, impulse, and acceleration.

The inertia measurement unit 510 of the smart insole device 500 according to an exemplary embodiment may measure acceleration with respect to three axes and angular velocities of individual axes. For example, the inertia measurement unit 510 may include a longitudinal axis of the user's foot, a lateral axis perpendicular to the longitudinal axis, and a vertical axis perpendicular to the longitudinal axis and the lateral axis. ) can be measured. The inertia measuring unit 510 may measure a total of three angular velocities based on each of the longitudinal axis, the horizontal axis, and the vertical axis. The smart insole device 500 may calculate the acceleration in the direction of gravity 511 as a physical quantity correlated with body weight as described above by using three accelerations and three angular velocities with respect to three axes. The smart insol device 500 may also calculate the momentum and impulse in the direction of gravity 511 by using the weight given at the time (eg, weight updated at every estimation) and the acceleration in the direction of gravity 511 . . A physical quantity that is correlated with body weight may be interpreted as affecting a change in the user's weight.

Accordingly, the weight estimation system may estimate the weight of the current section changed in response to the physical quantity collected and/or calculated in the current section compared to the previous weight by the smart insole device 500 . The weight estimation system may estimate the user's weight while walking on the user's ground 590 using information collected and/or calculated by the smart insole device 500 .

The smart insole device 600 according to an embodiment may include a pressure sensor 611 and an inertia measurement unit as described above. The smart insole device 600 may be mounted in the shoe 690 .

The inertia measurement unit 612 may measure inertia information about the user's foot, and the inertia information may include acceleration of three axes and angular velocity based on three axes, as described with reference to FIG. 5 . The inertia information can be used to calculate a weight-related physical quantity with respect to the direction of gravity. In FIG. 6 , the inertia measuring unit 612 is illustrated as being disposed at a portion where an arch of the user's foot is positioned, but the present invention is not limited thereto.

The pressure sensor 611 may sense pressures of a plurality of

points

611a, 611b, 611c, and 611d. For example, the pressure sensor 611 may sense the pressure applied by the foot to the ground at a plurality of

points

611a, 611b, 611c, and 611d while the bottom of the shoe equipped with the smart insole device touches the ground. The pressure sensor 611 according to an embodiment may include force sensitive resistors (FSRs) disposed at a plurality of

points

611a, 611b, 611c, and 611d. The pressure sensor 611 may detect a force applied to a corresponding point based on a change in resistance at each point. When the force applied to each point increases, the resistance value at that point decreases, and when the force decreases, the resistance value may increase. The force sensing resistor is a pressure indicating the strength of the force corresponding to the resistance value at the point value can be printed. For example, the pressure sensor 611 may include a forefoot point 611a, a medial point 611b (eg, an arch medial part), a lateral point 611c (eg, an arch external part) on the sole of the user's foot; And it is possible to sense the pressure of the heel point (611d). In this case, force sensing resistors may be disposed at the heel point 611a, the medial point 611b, the lateral point 611c, and the heel point 611d. However, this is an example for better understanding, and the implementation method of the pressure sensor 611 may vary depending on the design.

According to an embodiment, the pressure information sensed by the pressure sensor 611 may be used as auxiliary information for determining a section related to weight. For example, the pressure sensor 611 may sense the pressure applied to the ground by the foot at a plurality of

points

611a, 611b, 611c, and 611d while the bottom of the shoe equipped with the smart insole device 600 touches the ground. there is. At this time, the processor of the smart insole device 600 determines an effective section related to the weight based on the physical quantity calculated from the inertia information measured by the inertia measurement unit 612 and the pressure information sensed by the pressure sensor 611 . can decide Determination of the effective interval will be described with reference to FIG. 9 below.

If the sensing data of all sections during walking is used for weight estimation, in order to remove environmental noise (for example, forward energy with little relation to the direction of gravity, gait maneuver pattern due to joint angle when walking by each wearer) Too many filtering techniques must be applied unnecessarily. Furthermore, impulse extraction can be inaccurate despite many filtering techniques. The time section in which the energy in the direction of gravity is highest when walking, related to the body weight, is a section including a moment when the user's foot touches the ground 790 and a section including a moment when the user's foot falls from the ground 790 . When users with different body weights walk or run at the same speed, the amount of impact and the repulsion force at the moment the foot touches the ground 790 and the moment the foot falls from the ground 790 appear differently for each user's weight. Therefore, the smart insole device may extract only the data of the section related to the moment the foot touches the ground 790 and the moment the foot falls from the ground 790, instead of data of all sections collected during walking, and can be used to generate input data.

In the present specification, a time point at which the foot touches the ground 790 may be referred to as a heel strike time point, and a time point at which the foot is released from the ground 790 may be referred to as a toe off time point. The heel easy event 720 (TO event, toe off event) indicates an event in which the forefoot of the user's foot falls from the ground 790 and the heel landing event 710 (HS event, heel strike event) is the heel of the user's foot on the ground. A landing event at 790 may be indicated. The heel easy section may represent a section including a time point at which the heel easy event 720 occurs, and the heel landing section may represent a section including a time point at which the heel landing event 710 occurs. In the heel easy event 720, the effect mainly due to the ground 790 repulsion force appears, and in the heel landing event 710, the effect mainly by the ground 790 impact amount may appear. The smart insole device can accurately extract a landing time and an easy time by using an inertia measurement unit and a pressure sensor, which will be described with reference to FIG. 9 below.

According to an embodiment, the smart insole device may generate input data by using section data extracted from the inertia information during the user's constant speed walking state. For example, the smart insole device may determine, among inertia information collected while the user walks at a constant speed, a section including the above-described heel easy time and/or a section including the heel easy time point as the weight-related section.

For example, the smart insole device may determine in real time whether the user is in a constant speed driving state while the user is walking while wearing shoes for real-time weight estimation. For example, the processor of the smart insole device may calculate a movement speed of the user in a direction parallel to the ground 790 based on the inertia information. The processor may determine that the user is in a constant speed walking state when the calculated moving speed is maintained over a predetermined threshold time. The processor of the smart insole device determines the reference speed at an arbitrary reference point, and when the speed changes only within a predetermined speed increase/decrease range compared to the reference speed after the reference point exceeds the threshold time, it can be determined to be in a constant speed walking state. . Illustratively, if the threshold time is 5 minutes, the reference speed at the reference point is 5 km/h, and the predetermined speed increase/decrease range is ±0.2 km/h, the smart insole device increases the user's walking speed for more than 5 minutes When it changes only within the range of 4.8 km/h to 5.2 km/h, it may be determined that the user is in a constant speed walking state. The reference time point may be a time point repeated every predetermined period during walking, but is not limited thereto.

The smart insole device may store inertia information in the constant speed driving section. The smart insole device may create an n-th circular buffer in the memory and store inertia information during the walking state in the circular buffer. The smart insole device may determine an effective section used to generate input data based on pressure information sensed by a pressure sensor among weight-related sections, which will be described with reference to FIG. 9 below.

The smart insole device corresponding to one foot of the user may be paired with another smart insole device corresponding to the other foot. The paired smart insole devices may operate independently of each other or may operate in synchronization with each other. 8 shows sensing data 891 collected by the smart insole device worn on the left foot and sensing data 892 collected by the smart insole device worn on the right foot during walking as shown in FIG. 7 .

The sensing data 891 collected by the smart insole device may include inertial information 810 including acceleration for 3 axes and angular velocity based on 3 axes and pressure information 820 sensed at a plurality of points. there is. The smart insole device may accurately extract the heel landing event 851 and the heel easy event 852 based on the inertia information 810 and the pressure information 820 . For example, the smart insole device may calculate the acceleration of the foot in the direction of gravity from the inertia information 810 , and integrate the calculated acceleration to calculate the speed of the foot in the direction of gravity. The smart insole device may calculate the amount of exercise based on the speed with respect to the direction of gravity and the weight of the user previously estimated. The smart insole device may extract a heel landing event 851 and a heel easy event 852 based on the amount of exercise. Event extraction based on the amount of exercise will be described with reference to FIGS. 9 and 10 below. For reference, the smart insole device may determine an effective section among the heel landing section and the heel easy section based on the pressure information.

The smart insole device according to an embodiment may determine at least one of a heel landing time and a heel easy time based on the momentum 900 calculated from pressure and inertia information sensed at a plurality of points by the pressure sensor.

For example, as shown in FIG. 9 , the smart insole device may calculate the amount of exercise 900 over time calculated for one foot for each sampling point. The smart insole device may determine a sampling point that is a local extremum point in the momentum 900 as an event time point. For example, the smart insole device may determine a local minimum point as a heel landing time and a local maximum point as a heel landing time in the momentum 900 . As shown in FIG. 9, when the heel touches the ground, the speed of the point where the inertia measuring unit is disposed on the foot is momentarily reduced, so the momentum 900 is minimized, and when the heel starts kicking the ground, inertia is measured at the foot Since the speed of the point at which the auxiliary is disposed increases instantaneously, the momentum 900 may be maximized. For reference, as described above, the inertia measuring unit may be disposed in the arch portion of the foot between the forefoot point and the heel point.

In this case, the maximum value 952 and the minimum value 951 may appear in the momentum 900 before the user's foot touches the ground or leaves the ground only with inertia information. Therefore, since it is difficult to accurately detect an event using only inertia information, the smart insole device can detect a heel landing event and a heel landing event using pressure information as well. According to an embodiment, the smart insole device may determine an effective pole among the poles based on the pressure information. For example, the smart insole device may determine a sampling point corresponding to the minimum value 951 in the momentum 900 and at the same time a sampling point in which the pressure value of the heel point is equal to or greater than the threshold value as a valid heel landing time. In addition, the smart insole device may determine a sampling point corresponding to the maximum value 952 in the momentum 900 as well as a sampling point in which the pressure value of the forefoot point is equal to or greater than the threshold value as an effective forefoot easy time. Accordingly, the smart insole device can accurately extract the moment the heel actually lands on the ground and the moment the forefoot actually leaves the ground.

According to an embodiment, the smart insol device may extract a valid section based on each event time point. The smart insol device may extract a section including the event time point as a valid section. For example, the smart insole device may determine a section including a predetermined number of sampling centered on the heel landing time point as the heel landing section 910 . The smart insole device may determine a section including a predetermined number of sampling centered on the forefoot easy time point as the forefoot easy section 920 . 9, the heel landing section 910 is shown as a section from t0 to t0+∝, and the heel easy section 920 is shown as a section from t1 to t1+∝.

According to an embodiment, the smart insole device includes at least one of the heel strike data of the section including the heel landing time 1051 and the toe off data of the section including the heel easy time from the inertia information. One can be extracted as interval data. Figure 10 is an enlarged view of the heel landing section shown in Figure 9, illustratively describes the heel landing data. The smart insole device may sample the inertia information at a sampling rate. For reference, although the sampling rate is illustrated as 200 Hz by way of example in FIG. 10 , the present invention is not limited thereto.

The section data may include raw values of inertia information for each sampling point 1001 belonging to a valid section. For example, the heel landing data may include inertia raw values corresponding to a predetermined number of sampling points 1001 within an effective period determined based on the heel landing time 1051 . The heel easy data may include inertia raw values corresponding to a predetermined number of sampling points 1001 within an effective period determined based on the forefoot easy time. An inertia raw value at an arbitrary sampling point 1001 may include an acceleration with respect to three axes and an angular velocity based on the three axes measured at the corresponding sampling point 1001 .

According to an embodiment, the processor of the smart insole device may calculate the acceleration along the direction of gravity from at least one of the heel landing data and the heel easy data. In addition, the smart insol device may calculate the acceleration and momentum in the direction of gravity for every sampling point 1001 of the section data, and calculate the amount of impact that is the sum of the momentums within the section of the section data. The smart insole device may generate input data including an acceleration calculated with respect to a direction of gravity, an amount of exercise and an amount of impact calculated based on the user's current weight, along with the section data.

The input data includes section data including raw acceleration values and raw angular velocity values for each sampling point 1001 within the effective section, along with acceleration and gravity with respect to the direction of gravity calculated from section data. It can include momentum with respect to direction and impulse with respect to direction of gravity. The impulse may correspond to a change in momentum during the corresponding effective period. The input data may be exemplarily represented as shown in Table 1 below.

In Table 1 above, N may be an integer greater than or equal to 1 as the number of samplings within the effective interval. ax1 to axN are accelerations along the x-axis for each sampling point 1001, ay1 to ayN are accelerations to the y-axis, az1 to azN are accelerations to the z-axis, wr1 to wrN are rolling angular velocities, wp1 to wpN are pitch angular velocities, and wy1 to wyN may represent a yaw angular velocity. ag1 to agN may represent accelerations with respect to the direction of gravity for each sampling point 1001 calculated from raw data within an effective period. P1 to PN may represent momentum for each sampling point 1001 . As described above, the momentum may be calculated based on the speed and the body weight at which the acceleration is integrated. I is the amount of change in the amount of exercise within the effective period, and may exemplarily correspond to the sum of P1 to PN.

However, the input data is not limited thereto, and additional information such as height, age, gender, date and time (eg, date and time, etc.) may be additionally included in the input data. In the data used for training, the speed may be the speed of the treadmill.

However, in FIGS. 9 and 10 , the calculation of the momentum and the detection of the event in one foot are described for convenience of explanation, but it is not limited thereto. can Input data may be generated for both feet and transmitted to a weight estimation server.

The weight estimation server 1100 according to an embodiment may estimate the weight result 1109 by using the input data 1101 received from the smart insole device. The weight estimation server 1100 may include a communication unit 1110 , a processor 1120 , and a memory 1130 .

The communication unit 1110 may receive input data 1101 generated from section data related to the user's weight among inertia information measured by the smart insole device from the smart insole device. For example, the communication unit 1110 may collect information from a pair of smart insole devices worn on both feet of the user.

As described above, the communication unit 1110 may receive the input data 1101 generated for the user's constant speed gait state from the inertia information, for example, including a heel landing time among inertia information during the constant speed gait state. It is possible to receive the input data 1101 generated for at least one of the section and the section including the forefoot easy time. The communication unit 1110 receives the input data 1101 including section data along with the acceleration along the direction of gravity calculated from at least one of the heel landing data and the heel easy data, and the momentum and impulse calculated according to the user's current weight. can do. The communication unit 1110 may receive a weight estimation request together with the input data 1101 . The communication unit 1110 may periodically receive input data 1101 from the smart insole device. Also, the communication unit 1110 may transmit the weight result 1109 estimated using the neural network model 1131 to the smart insole device, as will be described later.

The processor 1120 may estimate the weight result 1109 of the user based on the neural network model 1131 from the input data 1101 . For example, in response to receiving a weight estimation request together with the input data 1101 from the smart insole device, the processor 1120 applies an operation according to the neural network model 1131 to the input data 1101 by A weight result 1109 of the user may be estimated. The processor 1120 may partially change the format of the input data 1101 received from the smart insole device and input it to the neural network model 1131 . For example, the processor 1120 may add additional information related to the user's weight to the input data 1101 .

The memory 1130 may store the neural network model 1131 . The neural network model 1131 may be a model of a machine learning structure designed to estimate weight from a plurality of input data 1101 . The neural network model 1131 may correspond to an example of a deep neural network (DNN). The DNN may include a fully connected network, a deep convolutional network, and a recurrent neural network. The neural network model 1131 may perform various tasks (eg, weight estimation, etc.) by mapping input data 1101 and output data in a non-linear relationship to each other based on deep learning. Deep learning is a machine learning technique from a big data set, and can map input data 1101 and output data to each other through supervised or unsupervised learning. Referring to FIG. 11 , the neural network model 1131 includes an input layer, a hidden layer, and an output layer. The input layer, the hidden layer, and the output layer each include a plurality of artificial nodes. Weighted inputs transmitted from the previous layer may be input to each artificial node included in each layer. Each node may output a value obtained by applying an activation function to the weighted inputs. The weighted input is a weight multiplied by the outputs of artificial nodes included in the previous layer. A weight (eg, a connection weight of a connection line) may be referred to as a parameter of the neural network model 1131 . The activation function may include a sigmoid, a hyperbolic tangent (tanh), and a rectified linear unit (ReLU), in which nonlinearity is formed in the neural network model 1131 by the activation function. can According to an embodiment, when input data 1101 is given, the neural network model 1131 may output a result value from an output layer through a hidden layer. For example, the neural network model 1131 shown in FIG. 11 may output the weight result 1109 from the output layer.

If the width and depth of the neural network model 1131 are sufficiently large, the neural network model 1131 may have enough capacity to implement an arbitrary function. When the neural network model 1131 learns enough training data through an appropriate training process, an optimal recognition performance may be achieved. Parameters (eg, connection weights) of the neural network model 1131 included in the neural network model 1131 may be trained in advance. For example, in supervised learning, a parameter of the neural network model 1131 may be updated based on paired training data of a training input and a training output (eg, a ground truth). The neural network model 1131 during training may be referred to as an ad hoc network. The temporary network may propagate the training input to each layer to produce a temporary output, and parameters of the temporary network may be updated to reduce the loss between the temporary output and the training output. When the loss reaches the target by repetition of the above-described training, the training may be terminated. However, although supervised learning has been mainly described for convenience of explanation, the present invention is not limited thereto, and the neural network model 1131 may be learned unsupervised.

For reference, the input data 1101 of one foot has been mainly described in FIG. 11 for convenience of explanation, but the present invention is not limited thereto. The weight estimation server 1100 may receive input data 1101 for both feet from a pair of smart insole devices, respectively, and apply the pair of input data 1101 for both feet to the neural network model 1131 . may be This is because the input data 1101 for both feet is data about the same user. However, the present invention is not limited thereto, and the operation of pairing the input data 1101 of both feet may be performed by the smart insole device or the weight estimation server 1100 .

Also, the weight estimation server 1100 may exclude the currently estimated weight as an error when the currently estimated weight with respect to the previous weight (eg, the previous weight) increases by more than a threshold error ratio. The weight estimation server 1100 may monitor only the weight change due to the user's activity by excluding a section in which an instantaneous or temporary rapid weight change occurs.

12 is a flowchart illustrating a training method according to an embodiment.

Previously, training of the neural network model was briefly described in FIG. 11 , and FIG. 12 describes a specific flow of training.

First, in step 1210, a population by weight may be secured. For example, the training device 1204 may build a population by weight. The training device 1204 may record the weight of each subject, and may classify the weight group by weight group of the population. For example, the training device 1204 may classify a 65 kg subject into a 60 kg weight group. The training device 1204 is a computing device, and may be, for example, a personal computer (PC), but is not limited thereto, and may be a weight estimation server. In this case, the smart insole device 110 and the training device 1204 may be connected to each other by wire, and the training device 1204 may be configured to collect sensing data from the smart insole device 110 in real time.

And in step 1220 , the smart insole device 110 may sense data while walking. For example, a subject belonging to the population by weight may start walking on a treadmill while wearing shoes equipped with the smart insole device 110 . The speed of the treadmill may be determined according to the training speed group to be collected. The smart insole device 110 may transmit sensing data (eg, inertia information) according to the subject's walking for a certain time at a certain speed to the training apparatus 1204 . The subject's gait may be an action performed on a treadmill. The training apparatus 1204 may store inertia information of a walking section among sensing data by classifying it by weight of a subject. At this time, the smart insole device 110 may extract the heel landing event and the heel easy event similar to those described above with reference to FIGS. 7 to 10 .

Subsequently, in step 1230 , the smart insole device 110 may perform pre-processing. For example, the smart insole device 110 may additionally calculate a physical quantity related to the direction of gravity in addition to the raw data measured by the inertia measurement unit. For example, the smart insol device 110 may calculate and store the acceleration value and momentum in the direction of gravity at every sampling point in each valid section for each walking speed of each subject, similar to the above-described description. Also, the smart insole device 110 may calculate the amount of impact for the entire effective section. The training device 1204 may generate training data consisting of pairs of a training input and a training output. The training input may include section data that is inertia information of an effective section and input data having a physical quantity related to a direction of gravity. The training output may include the actual weight of the subject. The smart insole device 110 may deliver pre-processed training data.

And in step 1240, the training device 1204 may collect the training data by dividing the training data by weight group and speed range. As described above, the training data may be composed of input data for each weight group and each speed range. Table 2 below describes exemplary training data. In Table 2 below, the HS event may represent a heel landing event, and the TO event may represent a heel easy event.

Then, in step 1250, the training device 1204 may train the neural network model. For example, the training device 1204 may generate a temporary output (eg, a temporary weight result) by inputting and propagating a training input to the neural network model. The training device 1204 may calculate a loss from the temporary output. The training apparatus 1204 may update a parameter (eg, a connection weight) of the neural network model based on back-propagation so that the loss is minimized. The training device 1204 may perform primary cross validation on the neural network model and additionally secure a test group for each weight for secondary validation. The training apparatus 1204 may evaluate the performance of the trained neural network model by extracting input data from the test group and inputting it to the neural network model.

The weight estimation server according to an embodiment may perform weight estimation using the neural network model trained as described above. In addition, the weight estimation server may continuously improve the performance of the trained neural network model. The weight estimation server may receive an improved neural network model from the training device 1204 , or the weight estimation server may improve the neural network model itself.

For example, in step 1260 , the smart insole device 110 may transmit additional training data. The training apparatus 1204 may additionally collect population data in which an error between the weight estimated by the neural network model and the actual weight exceeds a threshold value for each weight group. The training device 1204 and/or the communication unit of the weight estimation server transmits the population data collected about the user's constant speed gait state in the smart insole device 110 to the smart insole 110 device and the training device 1204 and/or the body weight. It can be received when a connection between the estimation servers is established. The smart insole device 110 may collect data for retraining separately from weight estimation and transmit it to the weight estimation server and/or the training device 1204 . For example, the smart insole device 110 may transmit the input data and the weight result to the weight estimation server and/or the training device 1204 whenever an effective period is secured. In addition, the transmission of additional training data is not limited thereto, and whenever the smart insol device 110 and/or the mobile terminal accumulates and stores the gait analysis record and accesses the weight estimation server and/or the training device 1204 , can also be transmitted. The transmitted data may be used for retraining in step 1270 to be described later.

In operation 1270 , the training apparatus 1204 according to an embodiment may perform retraining. For example, the training device 1204 may further train the neural network model based on the additionally collected population data. The training device 1204 may perform retraining similarly to the above-described steps 1210 to 1250 after a certain period (eg, a month or a quarter, etc.) has elapsed after completion of training in order to improve the error rate. . In this case, the training device 1204 may select an additional subject having an actual weight change (eg, weight loss, etc.). The training device 1204 may compare the error of the weight result estimated based on the actual data of the additional subject and the existing trained neural network model. The training device 1204 may retrain the neural network model by using the subject's data exceeding the threshold error as additional training data. At this time, subjects who have undergone physical changes due to disease, disease, or accident during the period, or subjects who have experienced external circumstances that have caused rapid weight loss or weight gain (e.g., diet, binge eating, etc.) may be excluded from further training data. External factors may be collected through interviewing additional subjects. The training apparatus 1204 may exclude data of a subject indicating a change in weight due to an external factor other than the exercise factor from the additional training data.

In an environment in which the accuracy of the neural network model is guaranteed, the smart guide device 110 and/or the weight estimation server detects an abrupt change in body weight (eg, a change in weight exceeding a threshold change amount), input at that time Rather than using the data as training data, the user may be alerted of an abnormal weight change. If external factors for abnormal weight change can be specified through online questionnaire, it can be used to derive a correlation between weight change and disease in the future healthcare field. However, in an environment before the accuracy of the neural network model is guaranteed, the training apparatus 1204 may collect questionnaire information about the estimation error from the user and analyze the error of the neural network model.

Also, in step 1280 , the training device 1204 may reflect the additional gait factor to the neural network model. The reflection of additional gait factors is described below in FIG. 13 .

The weight estimation server according to an embodiment includes a new input layer 1331 capable of receiving new input data defined to include, as an input variable, a target gait factor whose weight relevance exceeds a threshold score among a plurality of gait factors in a neural network. can be added to model 1330 .

Illustratively, gait-related factors that can be analyzed from information collectible by the smart insole device are shown in Tables 3 to 6 below.

The weight estimation server according to an embodiment may select a target gait factor that affects weight change from among the gait-related factors described above. The weight estimation server may define a format of new input data including the target gait factor as an additional input variable 1312 . In this case, the weight estimation server may further include another neural network model for estimating the weight relevance of each gait-related factor separately from the neural network model 1330 for weight estimation. Another neural network model may be a model in which a correlation between activity amount data and weight change is learned. The training data 1310 used in another neural network model may be data in which activity amount data and period data of valid periods are paired.

The weight estimating server transmits section data of valid sections and user activity data (for example, values recorded and/or calculated for each gait-related factor in Tables 3 to 6 described above) in the valid section to another neural network model. can be entered. A user's activity amount (eg, number of steps, distance, speed, stance/swing ratio, etc.) may affect the user's weight change. For example, the weight estimation server may calculate the weight relevance while monitoring the day on which the change in weight increases and the day on which the change in weight is decreased, mainly on the day of the daily weight change, monitoring what kind of action or activity is performed. Activities that contributed to weight loss may have increased weight relevance, and activities that contributed to weight gain may have lower weight relevance. For example, among walking-related factors, a patient with a neurological disease Parkinson's disease is accompanied by a gait disorder, and in this case, a characteristic of a significantly narrower stride and a reduced speed, and a decrease in weight at the same time may be observed from the patient.

The weight estimation server may estimate the weight relevance of the gait-related factor by applying a different neural network to the data collected for each gait-related factor. For example, the weight estimation server may estimate the weight relevance for each gait related factor as shown in Table 7 below. Accordingly, the weight estimation server may quantify the effect of the gait-related factors on the body weight.

As described above, the weight estimation server may select a target gait factor exceeding a threshold score among gait factors as the additional input variable 1312 . The weight estimation server may define new input data obtained by adding an additional input variable 1312 to the existing input data 1311 . In the example described above in Table 7, the threshold score may be 60%, and the weight estimation server may select run time and walk time as additional input variables 1312 and add them to the input data. . The weight estimation server may upgrade the neural network model 1330 by using the training data 1310 including new input data.

In this case, the weight estimation server may connect the new input layer 1331 capable of receiving the newly defined input data format to the front end of the existing input layer of the neural network model 1330 . Since parameters of layers other than the new input layer 1331 have been previously trained, the time and cost required to train the entire neural network model 1330 may be reduced. Training of the upgraded neural network model 1330 may be performed similarly to that described above with reference to FIG. 12 . For example, the weight estimation server applies the training data 1310 to the neural network model 1330 to calculate a temporary output 1390, and the neural network model ( 1330) may be updated. Here, the weight estimation server may update only parameters related to the new input layer 1331, but is not limited thereto, and may update parameters of the entire network.

Also, in response to a case in which new input data is defined, the weight estimation server may share information related to the format of the new input data to the smart insole device. For example, the weight estimation server may request the smart insole device to additionally collect additional input variables 1312 to generate input data.

The embodiments described above may be implemented by a hardware component, a software component, and/or a combination of the hardware component and the software component. For example, the apparatus, methods and components described in the embodiments may include, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable gate (FPGA) array), a programmable logic unit (PLU), a microprocessor, or any other device capable of executing and responding to instructions, may be implemented using a general purpose computer or special purpose computer. The processing device may execute an operating system (OS) and a software application running on the operating system. A processing device may also access, store, manipulate, process, and generate data in response to execution of the software. For convenience of understanding, although one processing device is sometimes described as being used, one of ordinary skill in the art will recognize that the processing device includes a plurality of processing elements and/or a plurality of types of processing elements. It can be seen that can include For example, the processing device may include a plurality of processors or one processor and one controller. Other processing configurations are also possible, such as parallel processors.

Software may comprise a computer program, code, instructions, or a combination of one or more thereof, which configures a processing device to operate as desired or is independently or collectively processed You can command the device. The software and/or data may be any kind of machine, component, physical device, virtual equipment, computer storage medium or apparatus, to be interpreted by or to provide instructions or data to the processing device. , or may be permanently or temporarily embody in a transmitted signal wave. The software may be distributed over networked computer systems and stored or executed in a distributed manner. Software and data may be stored in a computer-readable recording medium.

The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded in a computer-readable medium. The computer readable medium may include program instructions, data files, data structures, etc. alone or in combination, and the program instructions recorded on the medium are specially designed and configured for the embodiment, or are known and available to those skilled in the art of computer software. may be Examples of the computer-readable recording medium include magnetic media such as hard disks, floppy disks and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic media such as floppy disks. - includes magneto-optical media, and hardware devices specially configured to store and execute program instructions, such as ROM, RAM, flash memory, and the like. Examples of program instructions include not only machine language codes such as those generated by a compiler, but also high-level language codes that can be executed by a computer using an interpreter or the like.

The hardware devices described above may be configured to operate as one or a plurality of software modules to perform the operations of the embodiments, and vice versa.

As described above, although the embodiments have been described with reference to the limited drawings, those of ordinary skill in the art may apply various technical modifications and variations based thereon. For example, the described techniques are performed in an order different from the described method, and/or the described components of the system, structure, apparatus, circuit, etc. are combined or combined in a different form than the described method, or other components Or substituted or substituted by equivalents may achieve an appropriate result.

Therefore, other implementations, other embodiments, and equivalents to the claims are also within the scope of the following claims.

Claims

In the smart insole device,

an inertial measurement unit (IMU) for measuring inertial information including acceleration and angular velocity of a user's foot;

a processor for extracting section data related to the weight of the user from the inertia information and generating input data from the extracted section data; and

A communication unit for requesting weight estimation while transmitting the input data to be input to the neural network model to a weight estimation server storing the neural network model

A smart insole device comprising a.
According to claim 1,

The processor is

Extracting at least one of the heel strike data of the section including the heel landing time and the toe off data of the section including the heel easy time from the inertia information as the section data,

Smart insole device.
According to claim 1,

A pressure sensor that senses the pressure applied by the foot to the ground at a plurality of points while the bottom of the shoe equipped with the smart insole device touches the ground

further comprising,

The processor is

Determining at least one of the heel landing time and the heel easy time based on the momentum calculated from the pressure and the inertia information sensed at the plurality of points by the pressure sensor,

Smart insole device.
According to claim 1,

The processor is

Calculating an acceleration along the direction of gravity from at least one of the heel landing data and the heel easy data, and generating input data including the calculated acceleration,

Smart insole device.
According to claim 1,

The processor is

Generating the input data using the section data extracted during the user's constant speed walking state from the inertia information,

Smart insole device.
6. The method of claim 5,

The processor is

Calculating the moving speed of the user in a direction parallel to the ground based on the inertia information, and determining that the user is in the constant speed walking state when the calculated moving speed is maintained over a predetermined threshold time,

Smart insole device.
According to claim 1,

The processor is

generating the input data including the amount of exercise and the amount of impulse calculated based on the current weight of the user along with the section data,

Smart insole device.
8. The method of claim 7,

The processor is

Calculating the acceleration and momentum in the direction of gravity for every sampling point of the section data, and calculating the amount of impulse that is the sum of the momentum within the section of the section data,

Smart insole device.
According to claim 1,

The communication unit,

receiving a weight result estimated by applying the neural network model to the input data from the weight estimation server;

The processor is

After receiving the weight result, input data is generated using the weight result and the inertia information;

Smart insole device.
In the method of operating a smart insole device,

measuring inertial information including acceleration and angular velocity of the user's foot;

extracting section data related to the weight of the user from the inertia information;

generating input data from the extracted section data; and

requesting weight estimation while transmitting the input data to be input to the neural network model to a weight estimation server storing the neural network model;

How to include.
A computer program stored in a computer-readable recording medium in combination with hardware to execute the method of claim 10.
In the weight estimation server,

a memory for storing the neural network model;

a communication unit configured to receive input data generated from section data related to a user's weight among inertia information measured by the smart insole device from the smart insole device; and

A processor for estimating a weight result of a user based on the neural network model from the input data

A weight estimation server comprising a.
13. The method of claim 12,

The communication unit,

Receiving the input data generated for at least one of a section including a heel landing time and a section including a heel easy time of the inertia information,

Weight Estimation Server.
13. The method of claim 12,

The communication unit,

Receiving the input data including the acceleration along the direction of gravity, calculated from at least one of heel landing data and heel easy data,

Weight Estimation Server.
13. The method of claim 12,

The communication unit,

Receiving the input data generated for the user's constant speed walking state from the inertia information,

Weight Estimation Server.
13. The method of claim 12,

The communication unit,

receiving input data including an amount of exercise and an amount of impulse calculated according to the section data and the current weight of the user;

Weight Estimation Server.
13. The method of claim 12,

The processor is

estimating the weight result of the user by applying an operation according to the neural network model to the input data;

The communication unit,

transmitting the estimated weight result to the smart insole device,

Weight Estimation Server.
13. The method of claim 12,

The processor is

additionally collecting population data in which the error between the weight estimated by the neural network model and the actual weight exceeds a threshold value for each weight group, and additionally training the neural network model based on the additionally collected population data;

Weight Estimation Server.
19. The method of claim 18,

The communication unit,

receiving the population data collected on the user's constant speed walking state in the smart insole device when a connection between the smart insole device and the weight estimation server is established;

Weight Estimation Server.
13. The method of claim 12,

The processor is

adding, to the neural network model, a new input layer capable of receiving new input data defined to include, as an input variable, a target gait factor whose weight relevance exceeds a threshold score among a plurality of gait factors;

Weight Estimation Server.