KR20180097091A - Apparatus and method for estimating fall risk based on machine learning - Google Patents

Abstract

The present invention relates to an apparatus for estimating the degree of risk of falling based on machine learning and a method thereof and, more specifically, to an apparatus for estimating the degree of risk of falling based on machine learning, which estimates the degree of risk of falling from signals obtained by measuring X, Y, and Z triaxial acceleration values of a body. The apparatus for estimating the degree of risk of falling based on machine learning comprises: an acceleration sensor capable of measuring acceleration information and tilt information of a body of a user; a signal processing unit capable of extracting and processing the acceleration and tilt information of the user, sensed by the acceleration sensor; and a control unit capable of determining the degree of risk of falling of the user, wherein the control unit determines the degree of risk of falling from signals extracted from the X, Y, and Z triaxial acceleration values of the body according to movement of the body of the user. Therefore, it is possible to determine the degree of risk of falling of the user and enable the user to predict risk of falling.

Description

[0001] The present invention relates to an apparatus and method for estimating fall risk based on machine learning,

More particularly, the present invention relates to a machine learning-based fall risk estimation apparatus for estimating a fall risk from a signal obtained by measuring X, Y and Z three-axis acceleration values of a body .

As the society rapidly progresses into an aging society, a variety of healthcare services are emerging that help elderly people and pedestrians who are uncomfortable to run their daily life with peace of mind.

Especially, the elderly who live alone like the elderly living alone often have accidents caused by falls because they have more time to live alone like elderly living alone. Therefore, various fall relief services Has been in the limelight, and research on this has been actively proceeding.

Especially, the technique of using 3 - axis accelerator to recognize fall of elderly people is being studied. These studies show that body movements caused by natural steps in daily behavior can be expressed as the body X, Y, Z 3-axis acceleration values, and through observations and analysis of acceleration signals, non-ideal body movements such as falls I start from the science that I can grasp.

However, the conventional method only describes a general method and notification function related to general fall-related signal detection, and has a problem that the implementation possibility is unclear due to a lack of specific fall estimation algorithm or specific methodology for securing accuracy.

The machine learning based fall risk estimation apparatus and method according to the present invention provides an apparatus for determining a fall risk of a user so as to enable a user to predict the risk of falling.

According to an aspect of the present invention, there is provided a machine learning based fall risk estimation apparatus comprising: an acceleration sensor capable of measuring acceleration information and tilt information of a user's body; a user sensed by an acceleration sensor; And a controller capable of determining a risk of falling of the user, wherein the controller controls the X, Y, Z three-axis acceleration of the body according to the movement of the user's body, Based fall risk estimation apparatus for determining a fall risk from a signal extracted from a value of a machine learning-based fall risk.

According to another aspect of the present invention, there is provided a method of estimating fall risk based on machine learning, comprising: measuring acceleration information and tilt information of a user's body; Processing an acceleration signal; Selecting a characteristic point; A step of performing machine learning training, and a step of applying a fall risk model to determine a fall risk.

The apparatus and method for estimating fall risk based on a machine learning method according to the present invention provide an apparatus for determining a fall risk of a user so as to enable a user to predict the risk of falling.

1 is a block diagram of a machine learning based fall risk estimation apparatus according to an embodiment of the present invention.
2 is a flowchart of a machine learning based fall risk estimation method in accordance with an embodiment of the present invention.
FIG. 3 is a conceptual diagram of a fall risk test used as a standard in rehabilitation medicine, a timed-up and go test (TUGT), a five times sit-to-stand test (FTSS), and an alternate step test (AST).
4 is a conceptual diagram for explaining a signal of the TUGT test according to an embodiment of the present invention.
5A and 5B are views for explaining a characteristic point according to an embodiment of the present invention.
FIGS. 6 and 7 are diagrams for explaining the correlation of the fall risk estimates according to an embodiment of the present invention.

Hereinafter, specific embodiments of the present invention will be described in detail with reference to the drawings. However, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims. It will be understood by those skilled in the art that the present invention may be embodied in many other specific forms without departing from the spirit or essential characteristics thereof.

While the invention is susceptible to various modifications and alternative forms, specific embodiments thereof are shown by way of example in the drawings and will herein be described in detail to the concrete inventive concept. It should be understood, however, that the invention is not intended to be limited to the particular embodiments, but includes all modifications, equivalents, and alternatives falling within the spirit and scope of the invention. Like reference numerals are used for like elements in describing each drawing.

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

1 is a block diagram of a machine learning based fall risk estimation apparatus according to an embodiment of the present invention.

1, the machine learning based fall risk estimation apparatus 100 includes at least one of an acceleration sensor 110, a signal processing unit 120, a control unit 130, a storage unit 140, and a communication unit 150 can do. , And the machine learning based fall risk estimation apparatus 100 will be described in detail below.

The acceleration sensor 110 may be a variety of acceleration sensors capable of sensing acceleration information and tilt information. In the embodiment of the present invention, a three-axis acceleration sensor can be used. The three-axis acceleration sensor senses the three-axis acceleration information and the tilt information according to the user's motion.

In one embodiment of the present invention, the machine learning-based fall risk estimation apparatus 100 includes a fall-through test (TUGT), a five-times sit-to-stand test FTSS, and Alternate Step Test (AST) are used to measure the X, Y and Z acceleration values of the body according to the body movements of the user. Such a signal can be measured by the acceleration sensor 110. [

The signal processing unit 120 may extract and process the acceleration and tilt information of the user sensed by the acceleration sensor 110. Generally, a three-axis acceleration sensor can measure a raw signal including both a motion acceleration component due to acceleration / deceleration during walking or motion and a gravitational acceleration component due to a tilt.

The control unit 130 can recognize the state of the user by moving the acceleration information and the slope information acquired by the acceleration sensor 110. [ In addition, the controller 130 may extract and process acceleration and tilt information of the user obtained in the signal processor 120. In an embodiment of the present invention, the controller 130 may determine whether a fall event has occurred using the acceleration information and the slope information detected from the acceleration 110. [ In addition, the user is finally confirmed that the fall has occurred, and the fall of the user is notified to the external server through the communication unit 150, and finally the emergency, such as a family, doctor, emergency center, or the like, can be transmitted.

The storage unit 140 may store information generated or received by the acceleration sensor 110, the signal processing unit 120, the control unit 130, and the communication unit 150. In one embodiment of the present invention, the past information of the user can be stored using the information stored in the storage unit 140. [ Therefore, the machine learning based fall risk estimation apparatus 100 can provide information optimized for a user.

The communication unit 150 may include one or more modules between the fall risk estimation apparatus 100 and the wireless communication system to enable communication between the fall risk estimation apparatus 100 and the external server. In addition, the communication unit 150 may include one or more modules that connect the fall risk estimation apparatus 100 to one or more networks.

The communication unit 150 may include at least one of a broadcast receiving module, a mobile communication module, a wireless Internet module, a short distance communication module, and a location information module.

2 is a flowchart of a machine learning based fall risk estimation method in accordance with an embodiment of the present invention.

The machine learning based fall risk estimation method according to an embodiment of the present invention includes a step S110 of obtaining an acceleration signal, a step S120 of processing an acceleration signal, a step of selecting a characteristic point S130, (S140), and applying a fall risk model (S150).

In the step of acquiring the acceleration signal (S110), the acceleration sensor 110 can continuously measure the acceleration and tilt information of the user.

The step of processing the acceleration signal (S120) may process the signal measured by the acceleration sensor 110 in the signal processing unit 120 or the control unit 130.

Specifically, when the acceleration sensor 110 senses the acceleration and the tilt, the signal processing unit 120 may extract acceleration and tilt information at predetermined time intervals and store the acceleration and tilt information in the storage unit 140. Here, the acceleration information can be extracted and stored as a motion acceleration component and a gravity acceleration component.

In one embodiment of the present invention, movement of the body, which occurs according to a natural step in the user's daily behavior, can be measured in the acceleration sensor 110, and the acceleration sensor 110 can measure X, Y, And the signal processing unit 120 and the control unit 130 can perceive motion through observation and analysis of the acceleration signal, so that non-ideal body motion such as a fall can be recognized.

In one embodiment of the present invention, the fall risk estimation apparatus 100 includes a fall-risk and a fall-threshold test (TUGT), a five-times sit-to-stand test (FTSS) A signal measuring the X, Y, Z 3-axis acceleration values of the body according to the body motion of the user performing the Alternate Step Test (AST) can be used.

FIG. 3 is a conceptual diagram of a fall risk test used as a standard in rehabilitation medicine, a timed-up and go test (TUGT), a five times sit-to-stand test (FTSS), and an alternate step test (AST).

4 is a conceptual diagram for explaining a signal of the TUGT test according to an embodiment of the present invention.

Referring to FIGS. 3 and 4, it is possible to describe a Timed-Up and Go Test (TUGT), a Five Times Sit-to-Stand test (FTSS), and an Alternate Step Test (AST).

Timed-Up and Go Test (TUGT) is an evaluation tool for evaluating dynamic balance capability. The TUGT allows the subject to be comfortably seated on the chair, walking with the signal "departure" from the chair to the point 3 meters, returning to the chair, and measuring the signal from the falling risk estimation apparatus 100 until the sitting position .

For example, as shown in FIG. 4, the fall risk estimation apparatus 100 can measure the acceleration of the user until the user wakes up, walks 3 meters, performs a turn operation, and again takes a walking and sitting operation.

The Five Times Sit-to-Stand test (FTSS) measures the time and condition required for the fall risk assessment device 100 to perform five sit-up and standing-up tests to assess mainly leg strength and balance ability.

Like the TUGT, the fall risk estimation apparatus 100 during the FTSS can measure the acceleration of the user.

The Alternate Step Test (AST) is a test that measures a table with a step, such as a stair, by repeating the steps of raising a pair, raising another foot on the table, and lowering the legs.

The fall risk estimating apparatus 100 measures the acceleration of the user while performing the above-mentioned Timed-Up and Go Test (TUGT), Five Times Sit-to-Stand test (FTSS) and Alternate Step Test can do.

Fig. 2 will be described again.

In the step of selecting characteristic points (S130), a predetermined number of characteristic points are selected from the predetermined characteristic point candidate groups.

5A and 5B are views for explaining a characteristic point according to an embodiment of the present invention.

The step S130 will be described with FIGS. 5A and 5B.

The predetermined characteristic points candidate group may be as shown in FIG. 5A.

In one embodiment of the present invention, the falling risk estimator 100 is configured to include 132 characteristic point candidates that can be referred to as characteristics of signals in terms of time, space, and frequency energy with respect to a measurement signal that is the result of an inspection test The county can be selected first.

The feature candidates are a set of candidates that are expected to be efficient among the feature points to be measured in order to efficiently analyze the state of the user, according to an embodiment of the present invention.

Therefore, the characteristic candidate group is not limited to the one included in FIG. 5A, and the number thereof may be changed.

In one embodiment of the present invention, the controller 130 of the fall risk estimator 100 uses the Sequential Forward Floating Search (SFFS) algorithm to find the last characteristic point of a valid number among the characteristic point candidates. It is possible to determine the final characteristic point that can represent the fall risk factor.

In one embodiment of the present invention, the SFFS algorithm can be implemented until a feature point that starts with an empty set without a selected feature point among the feature point candidates and provides the best prediction performance over other feature points is selected.

In an embodiment of the present invention, the controller 130 of the fall risk estimator 100 can adjust the number of characteristic points according to the estimation complexity. For example, when nine characteristic points are extracted, .

In the step S140 for performing the machine learning training and the step S150 for applying the fall risk model, a machine learning is performed using a model corresponding to the selected characteristic points, Can be determined.

In one embodiment of the invention, the fall risk model is a machine learning exercise in accordance with the rehabilitation medicine standard fall risk test, and a user-specific multiple variable regression-based fall estimator can be constructed as a trained result.

In one embodiment of the present invention, the machine learning based fall risk estimation apparatus 100 uses a linear least squares (LLS) model to compare the selected feature X with the target of the fall risk score r of the Berg Balance Scale (BBS) You can map to a value. The LLS model has the advantages of simplicity and efficiency that provide good results in small datasets, as in other similar studies.

In addition, it is easy to interpret the parameters appropriate to the model, so it can provide the relative importance of each feature, unlike the nonlinear model. The linear prediction function for the i < th > subject's fall risk can be defined by Equation (1) and Equation (2).

In the equations (1) and (2), the column vector W means a selective regression weight of each characteristic point, and X may be a measured result of each characteristic point.

That is, the linear prediction function for the risk of falling of the i-th subject can be obtained by multiplying the measured value of each characteristic point by the selective regression weight of each characteristic point.

In one embodiment of the present invention, the machine learning based fall risk estimation apparatus 100 may measure the likelihood that a user will fall.

For example, the user may use the machine learning based fall risk estimation apparatus 100 to measure the degree of fall risk on that day. Therefore, the user can avoid the risk of falling by restraining external activity on a high risk of falls.

FIGS. 6 and 7 are diagrams for explaining the correlation of the fall risk estimates according to an embodiment of the present invention.

In one embodiment of the present invention, the fall risk estimated by the rehabilitation specialist according to the Berg Balance Scale (BBS) method and the fall risk using the proposed instrument and algorithm to evaluate the performance of the constructed machine learning based fall estimator 100 The results of analyzing the correlation between the estimates are shown in FIGS. 6 and 7.

Referring to FIG. 6, the finally selected characteristic points are 20 kinds, and the results are shown as estimates estimate 1 and estimate 2, respectively, as a result of two experiments.

Referring to FIG. 7, it can be seen that the final selected characteristic point is the result of two experiments on the 20 kinds of estimates, respectively, and the results of the fall estimation experiments showing a correlation of 95% or more with the expert's evaluation .

That is, it can be seen that the constructed machine learning based fall estimator 100 estimates of the present invention show very similar results to the fall risk scores of the Berg Balance Scale (BBS) as measured by the specialist.

Therefore, according to one embodiment of the present invention, it is possible to obtain the effect of preliminarily recognizing the risk of falling using only a simple measurement without diagnosis of a specialist.

As described above, the present invention has been described with reference to particular embodiments, such as specific elements, and limited embodiments and drawings. However, it should be understood that the present invention is not limited to the above- And various modifications and changes may be made thereto without departing from the scope of the present invention. Accordingly, the spirit of the present invention should not be construed as being limited to the embodiments described, and all of the equivalents or equivalents of the claims, as well as the following claims, fall within the scope of the spirit of the present invention .

100: Machine Learning Based Fall Risk Estimator
110: acceleration sensor
120: Signal processor
130:
140:
150:

Claims (12)

A machine learning based fall risk estimation apparatus,
An acceleration sensor capable of measuring acceleration information and tilt information of the user's body,
A signal processing unit that can extract and process acceleration and tilt information of the user detected by the acceleration sensor, and
And a control section capable of determining a risk of falling of the user,
Wherein the controller determines a fall risk from a signal extracted from X, Y and Z three-axis acceleration values of a body according to a movement of the user's body. The method according to claim 1,
The control unit may be configured to perform the X, Y, and Z axes of the user's body according to the motion of the user performing the Timed-Up and Go Test (TUGT), the Five Times Sit-to-Stand test (FTSS), and the Alternate Step Test A machine learning based fall risk estimator that determines fall risk based on measured acceleration signals. 3. The method of claim 2,
Wherein the controller determines a fall risk based on a final characteristic point selected from a plurality of previously selected characteristic point candidates. The method of claim 3,
Wherein the controller determines a fall risk of selecting a final characteristic point using a Sequential Forward Floating Search (SFFS) algorithm. The method according to claim 1,
Wherein the controller is a machine learning based fall risk estimator for estimating a fall risk according to a Berg Balance Scale (BBS) method. The method according to claim 1,
Wherein the controller estimates a fall risk by a different method for each user through user-specific training. In a machine learning based fall risk estimation method,
Measuring acceleration information and tilt information of the user's body;
Processing an acceleration signal;
Selecting a characteristic point;
A step of conducting machine learning training and
A method for estimating fall risk based on machine learning comprising the step of determining a fall risk by applying a fall risk model. 8. The method of claim 7,
The three-axis acceleration values of the body according to the body movements of the user performing the Timed-Up and Go Test (TUGT), the Five Times Sit-to-Stand test (FTSS) and the Alternate Step Test A Machine Learning Based Fall Risk Estimation Method to Determine Fall Risk Based on Measured Signals. 9. The method of claim 8,
A method of estimating falling risk based on machine learning based on the selected final characteristic points among the pre-selected characteristic point candidates 10. The method of claim 9,
A Machine Learning Based Fall Risk Estimation Method to Determine the Fall Risk of Selecting the Final Characteristics by Using the Sequential Forward Floating Search (SFFS) Algorithm. 8. The method of claim 7,
A method of estimating fall risk based on machine learning based on Berg Balance Scale (BBS) method. 8. The method of claim 7,
A Machine Learning Based Fall Risk Estimation Method that Estimates Falling Risk by Different Methods for Each User through Training by User.

Images