KR20190119879A - System and the method for fall-down prediction - Google Patents

Abstract

According to the present invention, provided is a falling prediction method, which comprises the steps of: calculating a feature vector by extracting feature points from user motion acceleration data measured by a measuring device of a user; labeling the calculated feature vector; applying each of the labeled feature vectors to a plurality of falling prediction models of a machine learning algorithm; and comparing a data value calculated by the plurality of falling prediction model with a previously stored falling data value to select the falling prediction model most similar to the falling data value, and then providing an advance notification of the falling occurrence prediction using the selected prediction result value. According to the present invention, the falling occurrence can be predicted by applying a falling prediction model with respect to various types of falls and injured parts, and injuries due to falling can be prevented in advance by notifying the user of the falling risk before the falling occurs.

Description

Fall prediction system and its method {System and the method for fall-down prediction}

The present invention relates to a fall prediction system and a method thereof, and in particular, by applying a fall prediction model to various types of falls and injured areas to predict a fall occurrence and informing a user of the fall risk before falling, the injury caused by the fall. A fall prediction system and method thereof that can be prevented in advance.

As the aging society progresses rapidly, various health care services are appearing to help elderly people or pedestrians who are inconvenient to live their daily lives with confidence.

In particular, elderly people living alone, such as elderly people living alone, are more frequent accidents due to falls than ordinary young people, so various fall rescue services are available to prepare for falls for elderly people who live alone alone. Is in the spotlight, and research on this is being actively conducted.

However, in the prior art for the fall structure, since the sensor and the mobile phone are separated and transmit data wirelessly, problems such as an increase in battery consumption, an increase in fall detection errors, and a decrease in filtration efficiency for error detection have occurred. Falls are detected depending on the amount of instantaneous change of data to detect falls, so if the user has a sensor value similar to a fall (user wears or removes the sensor, shakes it, runs, or bumps). This is judged as a fall.

In addition, since falls are different in intensity and direction depending on the type of falls (falling, slipping, falling forward, etc.), the injury area and the degree of danger of the human body are changed, and post-fall detection technology cannot minimize or prevent falls. .

Korean Patent Publication No. 10-2013-0112158

An object of the present invention is to provide a fall prediction system and method for predicting a fall in advance by applying fall data to a fall prediction model of a plurality of machine learning algorithms for various types of falls and injuries.

In addition, an object of the present invention is to provide a fall prediction system and method that can prevent the injury caused by the fall in advance by informing the user of the fall risk before the fall occurs.

The fall prediction method according to the present invention for achieving the above object comprises the steps of extracting a feature point from the user motion acceleration data measured from the user's measuring device to calculate a feature vector; Labeling the calculated feature vector; Applying each of the labeled feature vectors to a plurality of fall prediction models of a machine learning algorithm; And comparing the data values calculated from the plurality of falling prediction models with previously stored falling data values, selecting the falling prediction model most similar to the falling data values, and providing advance notification of the falling occurrence prediction using the selected prediction result value. Its features include; step.

Here, in particular, the labeling may be performed by selecting and labeling a section when the change of the acceleration value in the feature vector is greater than or equal to a predetermined value in the labeling step.

In particular, the selection of the interval in the feature vector is characterized in that the window block data of the action-specific frame and the time variable of the feature point extraction of the interval are applied.

In particular, the feature vector is characterized in that it uses a Gaussian function based peak detection algorithm.

In particular, the method further includes removing gravity acceleration and noise from the measured user motion acceleration data before calculating the feature vector.

In particular, the plurality of fall prediction models of the machine learning algorithm are characterized in that it includes five machine learning algorithms and one deep learning algorithm.

In addition, the fall prediction system according to the present invention for achieving the above object, a vector calculation unit for extracting a feature point of the acceleration data for the user's motion measured from the inertial measurement means to calculate a feature vector; An interval detector for labeling the calculated feature vector; And apply the labeled feature vectors to a plurality of fall prediction models of a machine learning algorithm, and compare the data values calculated from the plurality of fall prediction models with previously stored fall data values to predict the fall data most similar to the fall data values. After the model is selected, it is characterized in that it comprises a prediction unit for providing advance notification for the fall occurrence prediction with the selected prediction result value.

According to the present invention, it is possible to prevent injuries due to a fall by applying a fall prediction model to various types of falls and injuries to predict the fall, and informing the user of the fall risk before the fall occurs.

1 is a view schematically showing that the acceleration data measured from the inertial measurement means of the user of the present invention is linked to the fall prediction system.
2 is a diagram schematically showing the configuration of a fall prediction system according to an embodiment of the present invention.
FIG. 3 is a flowchart schematically illustrating a fall prediction method according to an exemplary embodiment of the present invention. FIG.
4 is a diagram illustrating an example of extracting feature points and calculating feature vectors using acceleration data of the present invention.
5 shows an example of a process of labeling a feature vector of the present invention.
6 is a diagram illustrating an example of the windowing technique and the prior fall prediction time derivation process of the present invention.
FIG. 7 is a diagram illustrating an example of an experimental process for daily living operation and falls using the machine learning algorithm of the present invention.
8 is a diagram showing an example of preliminary prediction results through a fall prediction model of the present invention.

As the inventive concept allows for various changes and numerous embodiments, particular embodiments will be illustrated in the drawings and described in detail in the written description. However, this is not intended to limit the present invention to specific embodiments, it should be understood to include all changes, equivalents, and substitutes included in the spirit and scope of the present invention.

In describing the present invention, when it is determined that the detailed description of the related well-known technology may unnecessarily obscure the gist of the present invention, the detailed description thereof will be omitted. In addition, numerals (eg, first, second, etc.) used in the description process of the present specification are merely identification symbols for distinguishing one component from another component.

Further, in the present specification, when one component is referred to as "connected" or "connected" with another component, the one component may be directly connected or directly connected to the other component, but in particular It is to be understood that, unless there is an opposite substrate, it may be connected or connected via another component in the middle.

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

1 is a diagram schematically illustrating the interlocking with a fall prediction system by receiving acceleration data measured from an inertial measurement means of a user of the present invention, and FIG. 2 illustrates a configuration of a fall prediction system according to an embodiment of the present invention. It is a schematic drawing.

As shown in Figures 1 and 2, the fall prediction system 200 of the present invention, the inertial measurement means 100, the communication unit 210, the acceleration detector 220, the vector calculator 230, the interval detector ( 240, a fall prediction unit 250, and a database 260.

The inertial measurement means 100 is a device worn by a user, and wears one inertial measurement means 100 on a user's body part, for example, the left pelvis, and then through a wireless communication unit (Bluetooth, etc.) (not shown). Each motion acceleration value that is subdivided into the fall prediction system is transmitted in real time. Here, although the detailed configuration of the inertial measurement means 100 is not shown, it may include an acceleration sensor, a controller, an input / output device, a notification unit, a communication unit, and the like, but is not limited thereto.

The inertial measurement means 100 extracts the acceleration and inclination information of the user detected by the acceleration sensor using a three-axis acceleration sensor or the like. In general, a three-axis acceleration sensor measures a data signal (raw data) that includes both the acceleration component due to acceleration and deceleration and the gravity acceleration component due to slope in the acceleration value.

The communication unit 210 receives each of the granular movement acceleration values of the user transmitted from the communication unit (not shown) of the inertial measurement means 100.

The acceleration detector 220 removes the motion acceleration component, the gravity acceleration component and the noise component from the received acceleration data of the user. More specifically, since the acceleration component is located in the high frequency component in the frequency domain, the gravity acceleration component is located in the low frequency component, the acceleration detection unit using the HPF (High Pass Filter) to the acceleration value (Ax, Ay, Az), LPF Extract the gravitational acceleration values (Tx, Ty, Tz) using (Low Pass Filter). For example, a signal obtained by extracting a gravity acceleration component using a low pass filter (LPF) is shown, and a signal obtained by extracting a motion acceleration component using a high pass filter (HPF). In addition, by using a noise removing filter, for example, a median filter may be used to remove the noise signal.

The vector calculating unit 230 calculates a feature vector by extracting feature points of the acceleration data of the user's motion measured from the inertial measurement means 100.

More specifically, the vector calculator 230 calculates 28 features by processing the acceleration data extracted by the inertial measurement means. For example, after extracting 26 features for each frame using the measured three-axis acceleration values, the maximum, minimum, average, median, change, skewness, and kurtosis of each window are calculated to calculate a total of 182 features. It can be represented as a vector.

The interval detector 240 labels the calculated feature vector. In this case, the feature vector for each window is used as an input value of the machine learning-based prior fall prediction model. In other words, unlike general unsupervised learning, algorithms based on machine learning or deep learning must be labeled according to the frame of each action. It will label the interval from start to end.

On the other hand, the fall of the window section can be divided into four sections (pre-impact, impact (fall), aftermath, and posture), and in the case of pre-fall prediction, it must be determined that the fall before impact. Therefore, input data is processed by extracting only pre-impact and impact data, for example, specifying a window size of 10, and then increasing the lead time by moving the window from the impact point to the starting point of the movement.

The fall prediction unit 250 applies the labeled feature vector to a plurality of fall prediction models of a machine learning algorithm, and compares data values calculated from the plurality of fall prediction models with previously stored fall data values. After selecting the fall prediction model most similar to the data value, the selected prediction result value provides a preliminary notification for the fall occurrence prediction.

More specifically, fall data for general mobile daily activities and various types of falls are compared with the database data. That is, the calculated data is applied to the plurality of fall prediction models of the machine learning algorithm, respectively, to determine the fall data closest to the previously stored fall data as the fall occurrence prediction. The machine learning algorithm selects an algorithm model representing the highest accuracy, sensitivity, and specificity using a total of five machine learning and one deep learning technique.

The database 260 stores the acceleration value transmitted from the inertial measurement means 100 in the database 260. In addition, the database of the fall data for the general mobile daily activities and various types of falls.

3 is a flowchart schematically illustrating a fall prediction method according to an embodiment of the present invention. As shown in FIG. 3, in the fall prediction method according to the present invention, first, gravity acceleration and noise are removed from the measured user motion acceleration data (S310).

More specifically, the motion acceleration component, the gravity acceleration component and the noise component are removed from the measured acceleration data of the user. Here, the motion acceleration component is located in the high frequency component in the frequency domain, the gravity acceleration component is located in the low frequency component, and the motion acceleration values (Ax, Ay, Az) using the HPF (High Pass Filter), LPF (Low Pass Filter) Extract the gravitational acceleration values (Tx, Ty, Tz) using. For example, a signal obtained by extracting a gravity acceleration component using a low pass filter (LPF) is shown, and a signal obtained by extracting a motion acceleration component using a high pass filter (HPF). In addition, by using a noise removing filter, for example, a median filter may be used to remove the noise signal.

4 is a diagram illustrating an example of extracting feature points and calculating feature vectors using acceleration data of the present invention.

Next, as illustrated in FIG. 4, a feature vector is calculated by extracting feature points from the acceleration data from which the acceleration, gravity acceleration, and noise are removed (S320).

More specifically, the acceleration data measured by the inertial measurement means is processed to calculate the 28 characteristics. For example, 28 features are calculated using the measured 3-axis acceleration values, which are extracted as 26 features for each frame. A total of seven maximum, minimum, average, median, change, skewness, and kurtosis for each window can be calculated to represent a total of 182 features as one vector. That is, in FIG. 4, the three-axis acceleration values are used to extract 26 frames using 28 features, wherein 1 to 3 are reference acceleration features of the X, Y, and Z axes, and 4 to 6 are the first. In the first frame, the acceleration feature is extracted for each difference of the X, Y, and Z axes, and the remaining 26 times are extracted. Therefore, the total 26 features mentioned above are obtained by calculating the maximum, minimum, average, median, change, skewness, and kurtosis of a total of seven areas for each frame, and a total of 182 features are obtained. That is, a total of 182 feature vectors are extracted to 30 for each action according to each action including both daily life and fall actions, and the final feature vectors are selected again.

FIG. 5 is a diagram illustrating an example of a process of labeling a feature vector of the present invention, and FIG. 6 is a diagram illustrating an example of a window technique and a prior fall prediction time derivation process of the present invention. As shown in FIG. 5, the step of labeling the calculated feature vector is performed (S330).

More specifically, the calculated feature vector is selected for a section when the change in the acceleration value is more than a predetermined value to label the section. Here, the selection of the interval in the feature vector is applied to the window block data of the action-specific frame and the time variable of the feature point extraction of the interval, and for this segmentation operation, for example, using a peak detection algorithm based on a Gaussian function. Preferred but not limited to this. In this case, the feature vector for each window is used as an input value of a pre-fall prediction model based on machine learning. In other words, unlike general unsupervised learning, algorithms based on machine learning or deep learning must be labeled according to the frame of each action. It will label the interval from start to end. Such a Gaussian function-based algorithm is optimized for finding peaks that occur repeatedly.

On the other hand, as shown in Figure 6, the fall of the window section is pre-impact, impact (fall), aftermath (postmath) and posture of the four sections In the case of a pre-fall prediction, it is necessary to determine that the fall is before impact. Thus, only pre-impact and impact data are extracted. For example, after setting the window size 10, the input value is processed by increasing the lead time by moving the window from the impact point to the start point of the movement.

Next, applying the labeled feature vector to each of the plurality of falling prediction models of the machine learning algorithm is performed (S340). Here, the plurality of fall prediction models of the machine learning algorithm may include five machine learning algorithms and one deep learning algorithm.

More specifically, a machine learning based algorithm applies a learning style. In this case, as a machine learning algorithm applied to the plurality of fall prediction models, SVM (Support Vector Machine), K-NN (Nearest Neighboring), DT (Decision Tree), QDA (Quadratic discriminant analysis), NB (Naive Bayes) And ANN (Artificial Neural Network) may be applied, but is not limited thereto.

Here, the support vector machine (SVM) completes the fall prediction model based on the acceleration data collected by the classification algorithm based on the algorithm, and after the accuracy test, sets the machine learning algorithm or completes the programming according to the accuracy. The fall prediction model algorithm of K-NN (Nearest Neighboring) is an algorithm that finds who is the k nearest neighbor data around me and decides my class as the majority of the class of neighbors. In other words, when there is train data that is already labeled and belongs to supervised learning for classification, it finds out what labeling is appropriate based on the existing data when new data appears, and belongs to the non-gigido learning. The data is not labeled, and it is knn that belongs to the classification to compare features and tie similar things together, that is, to learn to do the labeling yourself. The DA (Decision Tree Algorithms) is a decision tree algorithm, also called a "decision tree," to generate a decision model based on the values of attributes in real data. It is mainly used to solve classification and regression problems. It is generally used in machine learning because of its high speed and high accuracy. The QDA (Quadratic discriminant analysis) is a stochastic algorithm that can estimate the parameters of the distribution of random variables according to each class. The NB (Naive Bayes) is a simple technique for creating a classifier, and is trained using various algorithms based on general principles, not training through a single algorithm. The ANN (Artificial Neural Network) is an artificial neural network algorithm, which is a model that mimics the structure and function of a neuronal network of biology. It is a form of pattern matching and is mainly used to solve classification and regression problems, but there are numerous subfields of hundreds of algorithms, and variants of algorithms to solve all types of problems.

Then, by comparing the data values calculated from the plurality of falling prediction models with previously stored falling data values, selecting the falling prediction model most similar to the falling data values (S350), and then notifying the prediction of the falling occurrence with the selected prediction result value. To provide a step (S360).

More specifically, the optimal fall prediction model most similar to the fall data value is selected by comparing the data values calculated in the plurality of fall prediction models with previously stored fall data values, and then the fall occurrence prediction is performed using the selected prediction result value. Providing advance notification. That is, the calculated data is applied to the plurality of falling prediction models of the machine learning algorithm, respectively, to determine the result value of the falling data having the closest approximation with the previously stored falling data as the fall occurrence prediction. Here, the pre-stored data is a database of falling data for various general mobile daily activities and various types of falls.

The machine learning algorithm uses a total of five machine learning and one deep learning technique to select an algorithmic fall prediction model representing the highest accuracy, sensitivity, and specificity.

On the other hand, Figure 7 is a view showing an example of the experiment process for everyday life behavior and falls using the machine learning algorithm of the present invention, Figure 8 shows an example of the pre-prediction results through the fall prediction model of the present invention Drawing.

As shown in FIG. 7 and FIG. 8, experimental data of a total of 26 different motions are acquired to distinguish the everyday life motion from the fall, and the result of the previous fall prediction model is shown. In Figure 8, the pre-fall and daily motion prediction accuracy when reduced from the impact point (100%) to 50%, the six machine learning accuracy of about 99% or more at 100% starting point from the impact point, Although reduced to 50%, the ANN shows a slight change in accuracy. In addition, prior fall and daily motion prediction sensitivity and specificity can be seen when the window size is reduced by 50% from the impact point (100%).

Therefore, the field of application of the present invention can be applied to a machine learning-based classification modeling method capable of predicting various types of falls and a wearable device capable of measuring falling data, and it is possible to inform a wearer of a fall before the danger. The wearable device as a user inertial measurement device can be combined with an appropriate control device of a fall-related protection mechanism such as an airbag or a protector to minimize or prevent injury of the elderly due to the fall.

The scope of the present invention is not limited to the above-described embodiments but may be implemented in various forms of embodiments within the appended claims. Without departing from the gist of the invention claimed in the claims, it is intended that any person skilled in the art to which the present invention pertains falls within the scope of the claims described in the present invention.

100 --- Inertial measurement means 200 --- Fall prediction system
210 --- Communication unit 220 --- Acceleration detector
230 --- vector calculator 240 --- interval detector
250 --- Fall prediction 260 --- Database

Claims (7)

Calculating a feature vector by extracting feature points from user acceleration data measured from a user's measuring device;
Labeling the calculated feature vector;
Applying each of the labeled feature vectors to a plurality of fall prediction models of a machine learning algorithm; And
Comparing the data values calculated by the plurality of falling prediction models with previously stored falling data values to select a falling prediction model most similar to the falling data values, and then providing a preliminary notification of the falling occurrence prediction using the selected prediction result value; Fall prediction method including;
The method of claim 1,
And in the labeling step, selecting and labeling a section when the change in the acceleration value in the feature vector is greater than or equal to a predetermined value.
The method of claim 2,
The selection of the interval in the feature vector is a fall prediction method characterized in that to apply the window block data of the action-specific frame and the time variable of the feature point extraction of the interval.
The method of claim 3,
And the feature vector uses a Gaussian function based peak detection algorithm.
The method of claim 1,
And removing gravity acceleration and noise from the measured user motion acceleration data prior to calculating the feature vector.
The method of claim 1,
The plurality of fall prediction models of the machine learning algorithm include five machine learning algorithms and one deep learning algorithm.
A vector calculator configured to extract a feature point of acceleration data of a user's motion measured from inertial measurement means and calculate a feature vector;
An interval detector for labeling the calculated feature vector; And
Applying the labeled feature vector to a plurality of falling prediction models of a machine learning algorithm, and comparing the data values calculated from the plurality of falling prediction models with previously stored falling data values, the falling prediction model most similar to the falling data values. After selecting the fall prediction system comprising a fall prediction unit for providing a pre-notification for the fall occurrence prediction with the selected prediction result value.

Images